CN104424105B - The read-write processing method and device of a kind of internal storage data - Google Patents

Abstract

The invention discloses a method and device for reading and writing memory data, relates to the technical field of computer systems, and solves the problem of wasting a lot of overhead on data request packets in the prior art when accessing memory data with a small amount of data , leading to the problem of low transmission efficiency. The method includes: the source node acquires each request packet to be sent; parsing the destination node identifiers of each request packet to be sent, and judging whether the destination node identifiers of each request packet to be sent are the same; At least two request packets to be sent are respectively used as sub-request packets and packaged into a merge request packet; the merge request packet is sent to the destination node corresponding to the destination node identifier, and the merge request packet is used to indicate the destination node Perform read and/or write operations on memory data of the destination node. The embodiment of the present invention is applied in the process of reading and writing memory data.

Description

Method and device for reading and writing processing of memory data

technical field

The invention relates to the technical field of computer systems, in particular to a method and device for reading and writing data in memory.

Background technique

In a multi-node computer system, the efficiency of data interaction between local and remote nodes directly affects the performance of the entire system. Remote direct memory access (Remote Direct Memory Access, RDMA) is to quickly move memory data from a system to remote system memory without any impact on the operating system, for example, without CPU intervention, memory data of a node It is directly transmitted to the physical memory of another node, thereby reducing the impact of data transmission on processing components and improving system performance.

In RDMA read and write operations, the granularity of data exchange between memory and Cache (cache memory) is a Cache Line, with a size of 64Bytes (bytes) or 128Bytes. If the data of each Cache Line is packaged into a request data There is a certain degree of link bandwidth waste in this small-scale data transmission. For example, the data packet header of PCIE (Peripheral Component Interface Express, fast peripheral interconnection standard) may have 16Bytes, and if the Payload (payload) length is 16Bytes, the utilization rate of the link bandwidth is only 50%, so multiple remote data access operations with small data volumes will waste more overhead on data request packets, resulting in low transmission efficiency.

Contents of the invention

Embodiments of the present invention provide a method and device for reading and writing memory data, which improves the utilization efficiency of links.

In order to achieve the above object, embodiments of the present invention adopt the following technical solutions:

In a first aspect, the present invention provides a method for reading and writing memory data, including:

Obtaining each request packet to be sent, the request packet to be sent includes: a read request packet, and/or a write request packet;

Analyzing the destination node identifiers of the respective request packets to be sent, and judging whether the destination node identifiers of the respective request packets to be sent are the same;

The at least two request packets to be sent with the same destination node identifier are respectively used as sub-request packets, and packaged into a merge request packet, and the merge request packet includes a public header and at least two sub-request packets, and the public header includes Contains a merged package identifier, a sub-request package number identifier and a source node identifier, the merged package identifier is used to indicate that the request package is a combined package of the at least two sub-request packages, and the sub-request package number identifier is used In order to indicate the number of sub-request packets carried by the merge request packet, the sub-request packet includes a sub-request packet header and sub-request information;

Sending the merge request packet to the destination node corresponding to the destination node identifier, where the merge request packet is used to instruct the destination node to perform read and/or write operations on the memory data of the destination node.

In a first possible implementation manner of the first aspect, the packaging into one merge request package includes:

Setting the combined package identifier to a predetermined value, the predetermined value indicating that the request package is a combined package of the at least two sub-request packages;

Obtain the number of sub-request packets, and write the number into the number identifier of the sub-request packets.

In combination with the first aspect or the first possible implementation manner of the first aspect, in the second possible implementation manner, the parsing of the destination node identifier of each request packet to be sent, and judging that each request packet to be sent Whether the destination node identifiers of the packets are the same includes:

Obtain the destination node identifier carried in each request packet to be sent;

Judging whether the destination node identifiers of the respective request packets to be sent are the same;

The packages packaged as a merge request include:

Extracting the target node identifier in the sub-request package, and placing the target node identifier in the common header of the merge request package.

In combination with the first aspect or the first possible implementation of the first aspect, in a third possible implementation, the parsing of the destination node identifier of each request packet to be sent, and judging the Whether the destination node identifiers of the packets are the same includes:

Obtain the memory address of the destination node carried in each request packet to be sent;

According to the memory address of the destination node, look up the destination node identification to which the memory address of the destination node belongs in the address attribution table, where the address attribution table is a correspondence table between the memory address of the destination node and the destination node identification;

It is judged whether the identifiers of the destination nodes of the request packets to be sent are the same.

With reference to the second possible implementation manner or the third possible implementation manner of the first aspect, in a fourth possible implementation manner, the sub-request package includes a write request sub-package and/or a read request sub-packet, and the write The header of the request sub-packet includes a write request identifier identifying the write request sub-packet, and the write request information in the write request sub-packet includes the data to be written to the destination node and the memory address of the destination node to be written The header of the read request sub-packet includes a read request identifier identifying the read request sub-packet, the read request information in the read request sub-packet includes the memory address of the destination node where the data to be read is located, and the method further includes include:

Receive a write request response message sent by the destination node, where the write request response message includes a write request identifier corresponding to the write request subpacket whose write request is successful; and/or,

Receive the read request response message sent by the destination node, the read request response message includes the read request identifier corresponding to the read request subpacket with a successful read request, and the read payload, the payload is the Read the data corresponding to the remote memory address in the request subpacket.

In combination with the first aspect or the first possible implementation of the first aspect or the second possible implementation or the third possible implementation or the fourth possible implementation, in the fifth possible implementation , the parsing of the destination node identifiers of the respective request packets to be sent, and judging whether the destination node identifiers of the respective request packets to be sent are identical include:

According to the destination node identifier corresponding to each request packet to be sent, the routing port corresponding to the destination node identifier is searched in the routing table, and the routing table includes the routing port corresponding to the destination node identifier, and the routing port is the routing port corresponding to the destination node identifier. Describe the port on the path where the request packet to be sent reaches the destination node;

It is judged whether the routing ports of the request packets to be sent are the same.

In a second aspect, the present invention provides a method for reading and writing memory data, including:

Receive request packet;

Judging whether the received request packet carries a combined package identifier, the combined package identifier is used to indicate that the request package is a combined request package composed of at least two sub-request packages, the sub-request package includes: a read request sub-packet, and/ Or write request subpackage;

When the request package carries a merge package identifier, it is determined that the request package is a merge request package, and the merge request package includes a public header and a sub-request package, and the public header includes a merge package identifier and the number of sub-request packages An identifier and a source node identifier, the identifier of the number of sub-request packets is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packets include sub-request packet headers and sub-request information;

According to the number of sub-request packets carried in the merge request packet and the header of each sub-request packet, extracting the number of sub-request packets corresponding to the number of sub-request packets in the merge request packet;

According to the header of each sub-request packet and the sub-request information, the corresponding sub-request packet is processed.

In the first possible implementation manner of the second aspect, the processing the corresponding sub-request packet according to the header and sub-request information of each sub-request packet includes:

Write the data to be written into the destination according to the write request identifier in the header of the write request subpacket, the data to be written in the write request information of the write request subpacket, and the memory address of the destination node to be written. node memory address; and/or,

According to the read request identifier in the header of the read request sub-packet and the memory address of the destination node where the data to be read in the read request information of the read request sub-packet is located, read the data to be read in the memory address of the destination node read data.

With reference to the first possible implementation of the second aspect, in the second possible implementation, after extracting the number of sub-request packages corresponding to the number of sub-request packages in the merge request package ,Also includes:

The source node identifier is written into the sub-request packet header of each sub-request packet, so that the destination node sends a response message to the source node.

With reference to the second possible implementation manner of the second aspect, in a third possible implementation manner, the destination node sending the response message to the source node includes:

Judging whether the source node identifiers corresponding to the response messages to be sent are the same;

Packing the to-be-sent response messages with the same source node identifier as sub-response messages into a combined response message package, the combined response message package includes a write request sub-response message and/or a read request sub-response message;

Send the combined response message packet to the source node corresponding to the source node identifier.

In a fourth possible implementation of the second aspect, when the destination node is a routing port, in the extracting the sub-request packets corresponding to the number of sub-request packets in the merge request packet After that, also include:

Extract the source node identifier in the public header of the merge request package, and write it into the header of each sub-request package.

In a third aspect, the present invention provides a device for reading and writing memory data, including:

An acquisition unit, configured to acquire each request packet to be sent, wherein the request packet to be sent includes: a read request packet, and/or a write request packet;

A judging unit, configured to parse the destination node identifiers of the respective request packets to be sent, and judge whether the destination node identifiers of the respective request packets to be sent are the same;

A processing unit, configured to pack the at least two request packets to be sent with the same destination node identifier as sub-request packets into a combined request package, wherein the combined request package includes a common header and at least two sub-request packages, The public packet header includes a merged package identifier, a sub-request packet number identifier, and a source node identifier, and the merged package identifier is used to indicate that the request package is a combined package of the at least two sub-request packages, and the sub-request The packet number identifier is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packets include sub-request packet headers and sub-request information;

A sending unit, configured to send the merge request packet to a destination node corresponding to the destination node identifier, where the merge request packet is used to instruct the destination node to perform read and/or write operations on the memory data of the destination node .

In a first possible implementation manner of the third aspect, the processing unit includes:

A first identification module, configured to set the combined package identifier to a predetermined value, the predetermined value indicating that the request package is a combined package of the at least two sub-request packages;

The second identification module is configured to obtain the number of sub-request packets, and write the number into the identifier of the number of sub-request packets.

With reference to the third aspect or the first possible implementation manner of the third aspect, in a second possible implementation manner, the judging unit includes:

The first obtaining module is used to obtain the destination node identification carried in each request packet to be sent;

The first judging module is used to judge whether the destination node identifiers of the request packets to be sent are the same;

The processing unit also includes:

The third identification module is configured to extract the destination node identifier in the sub-request package, and place the destination node identifier in the common header of the merge request package.

With reference to the third aspect or the first possible implementation manner of the third aspect, in a third possible implementation manner, the judging unit includes:

The second obtaining module is used to obtain the memory address of the destination node carried in each request packet to be sent;

The first search module is used to search the destination node identifier to which the destination node memory address belongs in the address attribution table according to the destination node memory address, and the address attribution table is the corresponding relationship between the destination node memory address and the destination node identifier surface;

The second judging module is configured to judge whether the destination node identifiers of the request packets to be sent are the same.

With reference to the second possible implementation manner or the third possible implementation manner of the third aspect, in a fourth possible implementation manner, the sub-request package included in the merge request package obtained by the processing unit includes writing request subpackage and/or read request subpackage, the header of the write request subpackage includes a write request identifier identifying the write request subpackage, and the write request information in the write request subpackage includes the information to be written to the destination node data and the memory address of the destination node to be written, the header of the read request sub-packet includes a read request identifier identifying the read request sub-packet, and the read request information in the read request sub-packet includes the location of the data to be read memory address of the destination node, the device also includes:

A receiving unit, configured to receive a write request response message sent by the destination node, where the write request response message includes a write request identifier corresponding to the write request subpacket whose write request is successful; and/or, used to receive the The read request response message sent by the destination node, the read request response message includes the read request identifier corresponding to the read request subpacket for which the read request is successful, and the read payload, the payload is the read request subpacket The data corresponding to the remote memory address in the packet.

In combination with the third aspect or the first possible implementation manner or the second possible implementation manner or the third possible implementation manner or the fourth possible implementation manner of the third aspect, in the fifth possible implementation manner , the judging unit also includes:

The second search module is configured to search the routing table for the routing port corresponding to the destination node ID according to the destination node ID corresponding to each request packet to be sent, and the routing table includes the routing port corresponding to the destination node ID , the routing port is a port on the path where the request packet to be sent reaches the destination node;

The third judging module is used for judging whether the routing ports of the request packets to be sent are the same.

In a fourth aspect, the present invention provides a device for reading and writing memory data, including:

a receiving unit, configured to receive the request packet;

A judging unit, configured to judge whether the received request packet carries a combined package identifier, the combined package identifier used to indicate that the request package is a combined request package composed of at least two sub-request packages, and the sub-request package includes: a read request subpackage, and/or write request subpackage;

A determining unit, configured to determine that the request package is a merge request package when the request package carries a merge package identifier, the merge request package includes a public header and a sub-request package, and the common header includes a merge package identifier, A sub-request packet number identifier and a source node identifier, the sub-request packet number identifier is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packet includes a sub-request packet header and sub-request information;

An extracting unit, configured to extract the number of sub-request packets corresponding to the number of sub-request packets in the merge request packet according to the number of sub-request packets carried in the merge request packet and the header of each sub-request packet ;

The processing unit is configured to process the corresponding sub-request packet according to the header of each sub-request packet and the sub-request information.

In a first possible implementation manner of the fourth aspect, the processing unit includes:

The first processing module is configured to write the data to be written according to the write request identifier in the header of the write request subpacket, the data to be written in the write request information of the write request subpacket, and the memory address of the destination node to be written. Incoming data is written to the memory address of the destination node;

The second processing module is configured to, according to the read request identifier in the header of the read request sub-packet and the memory address of the destination node where the data to be read in the read request information of the read request sub-packet is located, in the memory of the destination node Read the data to be read from the address.

With reference to the first possible implementation manner of the fourth aspect, in a second possible implementation manner, the device further includes:

The identifier writing unit is configured to write the source node identifier into the sub-request packet header of each sub-request packet, so that the destination node sends a response message to the source node.

With reference to the second possible implementation of the fourth aspect, in a third possible implementation, the judging unit is further configured to judge whether the source node identifiers corresponding to the response messages to be sent are the same;

The processing unit is further configured to pack the to-be-sent response messages with the same source node identifier as sub-response messages into a combined response message package, the combined response message package including the write request sub-response message and/or or read request sub-response message;

The device also includes:

A sending unit, configured to send the combined response message packet to the source node corresponding to the source node identifier.

In a fourth possible implementation of the fourth aspect, when the destination node is a routing port, the processing unit is further configured to extract the source node identifier in the common header of the merge request packet, and write In the header of a sub-request packet.

In the method and device for reading and writing memory data provided by the embodiments of the present invention, the source node obtains each request packet to be sent, and the request packet to be sent includes: a read request packet and/or a write request packet, and then parses the The destination node identifier of each request packet to be sent, and judge whether the destination node identifier of each request packet to be sent is the same, and at least two request packets to be sent with the same destination node identifier are respectively used as sub-request packets, and are packaged as A merge request packet, the merge request packet is sent to the destination node corresponding to the destination node identifier, and the merge request packet is used to instruct the destination node to perform read and/or write operations on the memory data of the destination node . The embodiment of the present invention solves the problem in the prior art that a plurality of memory data read and write operations with a small amount of data waste more overhead on data request packets, resulting in low transmission efficiency, and improves link utilization efficiency.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required in the embodiments of the present invention. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For Those of ordinary skill in the art can also obtain other drawings based on these drawings without making creative efforts.

FIG. 1 is a flow chart of a method for reading and writing memory data provided by an embodiment of the present invention;

Fig. 2 is a schematic diagram of a read request packet format;

Fig. 3 is a schematic diagram of a write request packet format;

FIG. 4 is a schematic diagram of another read request packet format;

FIG. 5 is a schematic diagram of another write request packet format;

FIG. 6 is a schematic diagram of a combined write request packet format provided by an embodiment of the present invention;

FIG. 7 is a schematic diagram of a merged read request packet format provided by an embodiment of the present invention;

FIG. 8 is a schematic diagram of a combined read-write request packet format provided by an embodiment of the present invention;

FIG. 9 is a schematic diagram of another combined write request packet format provided by an embodiment of the present invention;

FIG. 10 is a schematic diagram of another merged read request packet format provided by an embodiment of the present invention;

FIG. 11 is a schematic diagram of another combined read-write request packet format provided by an embodiment of the present invention;

12 is a flow chart of another method for reading and writing memory data provided by an embodiment of the present invention;

13 is a flow chart of another method for reading and writing memory data provided by an embodiment of the present invention;

FIG. 14 is a structural diagram of a device for reading and writing memory data provided by an embodiment of the present invention;

15 is a structural diagram of a processing unit in a device for reading and writing memory data provided by an embodiment of the present invention;

Fig. 16 is a structural diagram of another device for reading and writing memory data provided by an embodiment of the present invention;

17 is a structural diagram of a judging unit in a device for reading and writing memory data provided by an embodiment of the present invention;

FIG. 18 is a structural diagram of another device for reading and writing memory data provided by an embodiment of the present invention;

19 is a structural diagram of a judging unit in a device for reading and writing memory data provided by an embodiment of the present invention;

FIG. 20 is a structural diagram of yet another device for reading and writing memory data provided by an embodiment of the present invention;

FIG. 21 is a structural diagram of a processing unit in yet another device for reading and writing memory data provided by an embodiment of the present invention;

FIG. 22 is a structural diagram of yet another device for reading and writing memory data provided by an embodiment of the present invention;

Fig. 23 is a structural diagram of yet another device for reading and writing memory data provided by an embodiment of the present invention;

FIG. 24 is a hardware structural diagram of a device for reading and writing memory data provided by an embodiment of the present invention;

FIG. 25 is a hardware structural diagram of another device for reading and writing memory data provided by an embodiment of the present invention.

detailed description

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

When the local memory interacts with the remote memory, more overhead will be wasted on data request packets for data read and write operations on memory with a small amount of data, resulting in low transmission efficiency. The embodiment of the present invention provides a memory data read Write a processing method to combine requests with the same destination node address to improve the utilization rate of the link. The execution subject source node of the method in the embodiment of the present invention can be a computer node (also called a computing node), as shown in Figure 1 As shown, the method includes the following steps:

101. Acquire each request packet to be sent, where the request packet to be sent includes: a read request packet, and/or a write request packet.

Each to-be-sent request packet acquired by the source node includes a read request packet and/or a write request packet.

102. Analyze the destination node identifiers of the request packets to be sent, and determine whether the destination node identifiers of the request packets to be sent are the same.

Wherein, the above step 102 can be realized in two ways.

Way 1: first obtain the destination node identifiers carried in each request packet to be sent, and then determine whether the destination node identifiers of the request packets to be sent are the same.

In the first way, before the source node reads and writes memory data to the destination node, referring to the prior art, the source node sends a memory borrowing request to the destination node, and the destination node decides according to the state of its own memory Accept or reject, when the destination node accepts the request of the source node, the destination node sends the allocated memory address and address length to the source node, and the source node is the memory of the destination node The address allocates a local memory address, and establishes an address mapping table, where the address mapping table is a mapping list of the local memory address, the destination node identifier, and the memory address of the destination node.

When the source node sends a read or write request packet to the destination node, the request packet includes the source node identifier, the destination node identifier and the memory address of the destination node, wherein the destination node identifier and the memory address of the destination node is obtained by the source node after obtaining the request carrying the local memory address, using the address mapping table to search through the local memory address, as shown in Figure 2 The read request packet to be sent is shown in Figure 3 as the write request packet to be sent. By acquiring the destination node identifiers carried in the request packets to be sent shown in FIG. 2 and/or FIG. 3 , and judging whether the destination node identifiers of the request packets to be sent are the same.

Method 2: Obtain the memory address of the destination node carried in each request packet to be sent; according to the memory address of the destination node, look up the destination node identifier of the memory address of the destination node in the address attribution table, and the address attribution table is the destination node A table of correspondence between node memory addresses and destination node identifiers; judging whether the destination node identifiers of the request packets to be sent are the same.

In mode 2, using PCIE as an example, the read request packet to be sent is shown in Figure 4, and the write request packet to be sent is shown in Figure 5, wherein the read request packet to be sent and the pending The address carried in the sending write request packet is the memory address of the destination node, and the destination node identification to which the memory address of the destination node belongs is searched in the address attribution table, and the address attribution table is the correspondence between the memory address of the destination node and the destination node identification relational table, for example, in the address attribution table, when the memory address of the destination node is 0H1000-0H1fff, the corresponding destination node ID is 1; when the memory address of the destination node is 0H2000-0H2fff, the corresponding destination node The identifier is 2, and when the memory address of the destination node is 0H3000-0H2fff, the corresponding destination node identifier is 3. After the memory address of the destination node carried in the request packet to be sent is obtained, the destination node identifier corresponding to the memory address of the destination node is obtained according to the address attribution table.

For example, when the memory address of the destination node carried in the read request packet to be sent is 0H1001, and the memory address of the destination node carried in the write request packet to be sent is 0H1ff0, the read request packet to be sent and The write request packets to be sent all exist within the memory address range of the destination node corresponding to the destination node ID 1, so the read request packets to be sent and the write request packets to be sent have the same destination node ID.

103. Pack the at least two to-be-sent request packets with the same destination node identifier as sub-request packets, and package them into a combined request package, where the combined request package includes a public header and at least two sub-request packages, and the common The packet header includes a combined package identifier, a number of sub-request packages and a source node identifier, the combined package identifier is used to indicate that the request package is a combined package of the at least two sub-request packages, and the number of sub-request packages is The identifier is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packets include sub-request packet headers and sub-request information.

When the sub-request package is packaged into a merge request package, the merge request package includes a public package header and a sub-request package, and the public package header contains a merge package identifier (Combine, C), which is used to indicate that the request package is all A combined package of the at least two sub-request packages, and set the combined package identifier to a predetermined value, which indicates that the request package is a combined package of the at least two sub-request packages, for example, C="1 " indicates that the request package is a combined request package; it includes the sub-request package number identifier (Combined Requests Number, CR NUM), which is used to indicate the number of sub-request packages carried by the combined request package, when packaged into a combined request package , to obtain the number of sub-request packets, and write the number into the identifier of the number of sub-request packets; it also includes a source node identifier (Requester ID), which is used to indicate the source node from which the merge request packet is sent, and also It is the identifier of the local source node. In addition, the public header of the merge request package also includes the public information in the headers of each sub-request package. The sub-request packet in the merge request packet includes a sub-request packet header and sub-request information, and the sub-request packet header includes the type (Type) and length of the sub-request packet, wherein the length is specifically the transaction layer packet length (Transaction Lay Packet, TLP length). Wherein, the type is used to distinguish whether the sub-request packet is a write request sub-packet or a read request sub-packet.

Among them, the write request subpackages with the same destination node ID can be packaged into a combined write request package, or the read request subpackets with the same destination node ID can be packaged into a combined read request package, or in order to further improve link utilization , pack the write request sub-package and the read request sub-packet with the same destination node ID into a merged read-write request package, which is not limited here.

From the merge request package obtained by packaging the first method in step 102, it is necessary to extract the destination node identifier in the sub-request packet, and place the destination node identifier in the public header of the merge request packet. As shown in FIG. 6 , it is the obtained merged write request packet, as shown in FIG. 7 , it is the obtained merged read request packet, and as shown in FIG. 8 , it is the obtained merged read and write request packet.

In the combined request package obtained by packaging the second method in step 102 above, the combined write request package obtained is shown in Figure 9, the combined read request package obtained is shown in Figure 10, and the obtained combined read request package is shown in Figure 11. Merge read and write request packets. In addition, the header of each sub-request package in the combined request package obtained in the second method also includes a request code, which is the code of the source node for the sub-request package and is the unique identifier of each sub-request package.

104. Send the merge request packet to the destination node corresponding to the destination node identifier, where the merge request packet is used to instruct the destination node to perform read and/or write operations on memory data of the destination node.

In the method for reading and writing memory data provided by the embodiment of the present invention, the source node obtains each request packet to be sent, and analyzes the destination node identifier of each request packet to be sent, and determines the destination node of each request packet to be sent identify whether they are the same, then use the at least two request packets to be sent with the same destination node identifier as sub-request packets, package them into a merge request packet, and finally send the merge request packet to the corresponding destination node identifier For a destination node, the merge request packet is used to instruct the destination node to perform read and/or write operations on memory data of the destination node. The embodiment of the present invention solves the problem in the prior art that a plurality of memory data access operations with a small amount of data waste more overhead on data request packets, resulting in low transmission efficiency, and improves link utilization efficiency.

The embodiment of the present invention also provides a method for reading and writing memory data. The target node of the execution subject of the method of the embodiment of the present invention may be a remote computer node, as shown in FIG. 12 , the method includes the following steps:

1201. Receive a request packet;

1202. Determine whether the received request packet carries a combined package identifier, where the combined package identifier is used to indicate that the request package is a combined request package composed of at least two sub-request packages, and the sub-request package includes: a read request sub-packet, and/or write requests subpackage.

When the destination node receives the request packet, it first judges whether the request packet contains the merged packet identifier C, for example, when it is judged that C=“1”, it means that the request packet is a merged request packet.

1203. When the request package carries a merge package identifier, determine that the request package is a merge request package, the merge request package includes a public header and a sub-request package, and the public header includes a merge package identifier, a sub-request package A number identifier and a source node identifier, the sub-request packet number identifier is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packet includes a sub-request packet header and sub-request information;

1204. According to the identifier of the number of sub-request packets carried in the merge request packet and the header of each sub-request packet, extract the number of sub-request packets corresponding to the identifier of the number of sub-request packets in the merge request packet.

When it is determined that the received request package is a merge request package, in addition to the merge package identifier, the common header of the merge request package also includes a sub-request package number identifier and a source node identifier, wherein the sub-request package number identifier is used for Indicates the number of sub-request packets carried by the merge request packet, and the source node identifier is used to indicate the source node from which the merge request packet is sent. Include sub-request packet header and sub-request information in each sub-request packet.

Through the number of sub-request packets carried in the merge request packet and the header of each sub-request packet, extract the number of sub-request packets corresponding to the number of sub-request packets in the merge request packet, wherein, Sub-request packet number identifier and the type and length in each sub-request packet header, or use the sub-request packet number identifier and the format, type and length in each sub-request packet header to extract the merge request packet and the The number of sub-request packets identifies the corresponding number of sub-request packets.

1205. Process the corresponding sub-request packet according to the header of each sub-request packet and the sub-request information.

Wherein, since the merge request package may contain a write request subpackage and/or a read request subpackage, the write request subpackage and the read request subpackage can be distinguished according to the format and/or type in the header of the subrequest packet, so that The write request sub-package and the read request sub-packet of different requests are processed differently.

With reference to the prior art, for a sub-write request, write the payload into the memory address of the destination node according to the memory address and payload of the destination node carried in the sub-packet of the write request; for the sub-read request , according to the memory address of the destination node carried in the read request subpacket, read the content on the memory address of the destination node, and return it to the source node.

In the method for reading and writing memory data provided by an embodiment of the present invention, the destination node judges whether the received request packet carries a merged package identifier, and when the request packet carries the merged packet identifier, determines that the request packet is a merged request package, and then according to the number of sub-request packets carried in the merge request packet and the header of each sub-request packet, extract the sub-request packets corresponding to the number of sub-request packets in the merge request packet, and finally According to the header of each sub-request packet and the sub-request information, the corresponding sub-request packet is processed. The embodiment of the present invention solves the problem in the prior art that a plurality of memory data access operations with a small amount of data waste more overhead on data request packets, resulting in low transmission efficiency, and improves link utilization efficiency.

The embodiment of the present invention can be applied to RDMA, pack and combine the request packets to be sent with the same destination node in the memory data of a computing node, and send them together to the remote computing node corresponding to the destination node, specifically, the local Cache will request Send it to the local memory controller for packaging and merging processing, then send it to the remote memory controller for unpacking processing, and send the obtained request to the remote Cache. The present embodiment takes PCIE as an example to illustrate, and Fig. 4 is the format of the read request packet in the standard PCIE, and Fig. 5 is the format of the write request packet, wherein the format and type are used to distinguish whether the request packet is a read request packet or a write request packet, and things The layer packet length is used to indicate the length of the request packet, the source node identifier is used to indicate the source node sending the request packet, the address in the read request packet represents the address of the memory data to be read by the read request packet, and the The address in the write request packet indicates the payload to be carried in the write request packet, the memory address to be written, and the meaning of other fields can refer to the packet format of PCIE in the prior art. The following uses the above application scenario as an example to introduce a method for reading and writing memory data in detail, as shown in Figure 13. The method includes the following steps:

1301. The source node acquires each to-be-sent request packet, where the to-be-sent request packet includes: a read request packet, and/or a write request packet;

1302. The source node parses the destination node identifiers of the request packets to be sent, and judges whether the destination node identifiers of the request packets to be sent are the same.

Obtain the memory address of the destination node carried in each request packet to be sent; according to the memory address of the destination node, look up the destination node identification of the memory address of the destination node in the address attribution table, and the address attribution table is the memory address of the destination node A correspondence table with the destination node ID; judging whether the destination node IDs of the request packets to be sent are the same.

The read request packet to be sent is shown in Figure 4, and the write request packet to be sent is shown in Figure 5, wherein the address carried in the read request packet to be sent and the write request packet to be sent is the For the memory address of the destination node, look up the destination node identifier of the memory address of the destination node in the address attribution table, the address attribution table is a correspondence table between the destination node memory address and the destination node identifier, for example, in the address attribution table Among them, when the memory address of the destination node is 0H1000-0H1fff, the corresponding destination node ID is 1, and when the memory address of the destination node is 0H2000-0H2fff, the corresponding destination node ID is 2, and the destination node memory address is When 0H3000-0H2fff, the corresponding destination node ID is 3. After the memory address of the destination node carried in the request packet to be sent is obtained, the destination node identifier corresponding to the memory address of the destination node is obtained according to the address attribution table.

For example, when the memory address of the destination node carried in the read request packet to be sent is 0H1001, and the memory address of the destination node carried in the write request packet to be sent is 0H1ff0, the read request packet to be sent and The write request packets to be sent all exist within the memory address range of the destination node corresponding to the destination node ID 1, so the read request packets to be sent and the write request packets to be sent have the same destination node ID.

1303. The source node takes the at least two request packets to be sent with the same destination node ID as sub-request packets, and packs them into a merge request packet, and the merge request packet includes a common header and at least two sub-request packets , the common packet header includes a combined package identifier, a number of sub-request packages and a source node identifier, the combined package identifier is used to indicate that the request package is a combined package of the at least two sub-request packages, and the sub-request package The identifier of the number of request packets is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packets include sub-request packet headers and sub-request information.

Among them, the write request subpackages with the same destination node ID can be packaged into a combined write request package, or the read request subpackets with the same destination node ID can be packaged into a combined read request package, or in order to further improve link utilization , packing the write request sub-package and the read request sub-packet with the same destination node identifier into a combined read-write request package, so the combined request package includes the write request sub-packet and/or the read request sub-packet.

When the sub-request package is packaged into a merge request package, the merge request package includes a public header and a sub-request package, and the public header contains a merge package identifier (Combine, C), which is used to indicate that the request package is multiple A combined request package composed of sub-request packages, for example, C=“1” indicates that the request package is a combined request package; it includes the sub-request package number identifier (Combined Requests Number, CR NUM), which is used to indicate the combined requests carried by the combined request package The number of sub-request packages also includes the source node identifier (Requester ID), which is used to indicate the source node from which the merge request package is sent, that is, the identifier of the local source node. In addition, the public header of the merge request package also includes Common information in the header of each sub-request packet. The sub-request package in the merge request package contains the sub-request package header and sub-request information, and the sub-request package header contains the format (Fmt), type (Type) and length of the sub-request package, where the length is specifically the transaction layer Packet length (Transaction Lay Packet, TLP length). Wherein, the format and type are used to distinguish whether the sub-request packet is a write request sub-packet or a read request sub-packet.

In addition, the header of each sub-request packet also includes a request code, which is the code of the source node for the sub-request packet, and is the unique identifier of each sub-request packet, so that the source node can receive the response returned by the destination node message, the completion status of the sub-request is determined according to the request code carried in the response message.

Specifically, as shown in Figure 10, two read request subpackages are combined to obtain a combined read request package. The combined package identifier and subrequest package number identifier in the figure are newly added additional information, wherein the subrequest package number identifies the length is 4 bits, and may indicate that the number of read request sub-packets included in the merged read request packet is 16. PCIE originally contains a request encoding (that is, a tag) field with a length of 8 bits. In the combined read request packet, the upper 4 bits of this field are shared, and the 0-3 bits are used to distinguish different read requests. The sub-packets can distinguish 16 read request sub-packets, which match the number indicated by the sub-request package number identification field. Among them, communication classification (TC), pollution reserved bit (EP), attribute (Attr) and source node identification (Requester ID) belong to the public header in the merged read request packet; format (Fmt), type (Type), whether ECRC (TD), transaction layer packet length (TLP length), request coding (Request Coding), last byte alignment (LastDW BE) and first byte alignment (1st DW BE) belong to the header of the read request subpacket, and the above fields For the meaning, please refer to the format of the read request packet in PCIE in the prior art.

As shown in Figure 9, two write request subpackages are combined to obtain a combined write request package, the meaning of the fields contained in the combined write request package is similar to that of the fields contained in the combined read request package shown in Figure 10, except that each A write request subpacket also contains the payload to be written to the destination node.

In order to further improve link utilization, write request sub-packets and read request sub-packets with the same destination node identifier can be packaged into a combined request package, as shown in FIG. 11 .

For the sub-requests in the above merge request package, the sub-requests shall be packaged and merged according to the order in which the sub-requests are sent.

1304. The source node sends the merge request packet to the destination node identified by the destination node, where the merge request packet is used to instruct the destination node to read and/or write the memory data of the destination node operate.

1305. The destination node receives the request packet;

1306. The destination node judges whether the received request packet carries a merged packet identifier, and the merged packet identifier is used to indicate that the request packet is a merged request packet composed of at least two sub-request packets, and the sub-request packet includes: read request subpackage, and/or write request subpackage.

When the destination node receives the request packet, it first judges whether the request packet contains the merged packet identifier C, for example, when it is judged that C=“1”, it means that the request packet is a merged request packet.

1307. When the request package carries a merge package identifier, determine that the request package is a merge request package, the merge request package includes a public header and a sub-request package, and the public header includes a merge package identifier, a sub-request package A number identifier and a source node identifier, the sub-request packet number identifier is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packet includes a sub-request packet header and sub-request information;

1308. According to the identifier of the number of sub-request packets carried in the merge request packet and the header of each sub-request packet, extract the number of sub-request packets corresponding to the identifier of the number of sub-request packets in the merge request packet.

When it is judged that the request packet is a combined request packet, each sub-request packet is obtained according to the identifier of the number of sub-request packets, and the format, type and length of each sub-request packet. In addition, since the merge request package may contain a write request subpacket and/or a read request subpacket, the write request subpackage and the read request subpacket can be distinguished according to the format and type in the sub-request packet header, so that different The write request sub-packet and the read request sub-packet of the request are processed differently.

1309. Process the corresponding sub-request packet according to the header of each sub-request packet and the sub-request information.

Write the data to be written into the destination according to the write request identifier in the header of the write request subpacket, the data to be written in the write request information of the write request subpacket, and the memory address of the destination node to be written. node memory address; and/or, according to the read request identifier in the header of the read request subpacket, and the destination node memory address where the data to be read in the read request information of the read request subpacket is located, at the destination node The data to be read is read from the memory address.

Further, after the destination node processes the sub-request packet, it is convenient to send a response message to the source node, and the source node identifier is written into the sub-request packet header of each sub-request packet, and the Write other information of the public header in the merge request packet into the sub-request packet header of each sub-request packet, so that each sub-request packet is converted into the format before merging.

In addition, after processing the sub-requests, the destination node may return a response message to the source node to inform the source node of the completion of each sub-request. Feedback can be given in two ways:

method one:

The destination node sends out the response messages of the processed sub-request packets respectively:

Obtain the source node identifier in the write request subpackage;

Sending a write request response message to the source node corresponding to the source node identifier, where the write request response message includes the write request identifier corresponding to the write request sub-packet whose write request is successful.

Wherein, when the write request sub-packet is a non-push write request, the destination node needs to send feedback to the source node, or the write request sub-packet required to complete the write request sends a response to the source node information.

The source node receives the write request response message sent by the destination node, and the write request response message includes a write request identifier corresponding to the write request sub-packet whose write request is successful.

Referring to the prior art, after the source node receives the write request response message, it compares the carried write request identifier with the write request identifier in the sent write request subpacket to obtain which write request subpackets are The destination node writes successfully, and which write request subpackages are not written, and the request packets that do not have a successful write request can be sent again.

and / or,

The destination node obtains the source node identifier in the read request subpacket;

Sending a read request response message to the source node corresponding to the source node identifier, the read request response message includes the request code corresponding to the read request subpackage for which the read request is successful, and the obtained payload, and the payload is The data obtained by the sub-read request.

The source node receives the read request response message sent by the destination node, and the read request response message includes the read request identifier corresponding to the read request subpacket whose read request is successful, and the obtained payload, and the payload The data fetched for the subread request.

The payload in the read request sub-packet is the data obtained by the destination node according to the remote memory address in the read request sub-packet, and the destination node will successfully obtain the read request corresponding to the sub-read request of the payload The response message is sent to the source node, and after the source node receives the payload, according to which addresses of the local end need to read data recorded when sending the read request subpacket, according to the read request identifier, the The corresponding payload can be read back to the corresponding address.

Method 2:

Judging whether the source node identifiers corresponding to the response messages to be sent are the same;

Packing the to-be-sent response messages with the same source node identifier as sub-response messages into a combined response message package, the combined response message package includes a write request sub-response message and/or a read request sub-response message;

Send the combined response message packet to the source node corresponding to the source node identifier.

After the source node obtains the combined response message packet, it obtains the corresponding response message according to the write request identifier and/or read request identifier in the packet, and the processing method after the acquisition is the same as that described in method 1, here No longer.

The method in the embodiment of the present invention can also be applied to data transmission between multiple ports in the system. According to the destination node identifiers corresponding to the respective request packets to be sent, search the routing table for the destination node identifiers corresponding to the destination node identifiers. Routing port, the routing table includes the routing port corresponding to the destination node identifier, the routing port is the port on the path where the request packet to be sent reaches the destination node, and the routing of each request packet to be sent is determined Whether the ports are the same, combine the request packets to be sent with the same routing port to obtain a combined request package, and send the combined request package to the routing port. After receiving the merge request packet, the routing port performs an unpacking operation, extracts the source node identifier in the common header of the merge request packet, writes it into the header of each sub-request packet, and then continues the following forwarding operation.

An embodiment of the present invention provides a memory data read-write processing device 14, the memory data read-write processing device 14 may exist in a computing node, or may be a computing node, as shown in Figure 14, the device includes An acquiring unit 141 , a judging unit 142 , a processing unit 143 and a sending unit 144 .

Wherein, the acquiring unit 141 is configured to acquire each request packet to be sent, and the request packet to be sent includes: a read request packet, and/or a write request packet;

Judging unit 142, configured to analyze the destination node identifiers of the respective request packets to be sent, and judge whether the destination node identifiers of the respective request packets to be sent are the same;

The processing unit 143 is configured to pack the at least two to-be-sent request packets with the same destination node identifier as sub-request packets, and package them into a combined request package, wherein the combined request package includes a public header and at least two sub-request packages , the common packet header includes a combined package identifier, a number of sub-request packages and a source node identifier, the combined package identifier is used to indicate that the request package is a combined package of the at least two sub-request packages, and the sub-request package The number of request packets is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packets include sub-request packet headers and sub-request information;

The sending unit 144 is configured to send the merge request packet to the destination node corresponding to the destination node identifier, the merge request packet is used to instruct the destination node to read and/or write the memory data of the destination node operate.

A read-write processing device for memory data provided by an embodiment of the present invention obtains each request packet to be sent, and analyzes the destination node identifier of each request packet to be sent, and judges whether the destination node identifier of each request packet to be sent is The same, then the at least two request packets to be sent with the same destination node identifier are respectively used as sub-request packets, packaged into a merge request packet, and finally the merge request packet is sent to the destination node corresponding to the destination node identifier , the merge request packet is used to instruct the destination node to perform read and/or write operations on memory data of the destination node. The embodiment of the present invention solves the problem in the prior art that multiple remote data access operations with small data volumes waste more overhead on data request packets, resulting in low transmission efficiency, and improves link utilization efficiency.

Optionally, as shown in Figure 15, the processing unit includes:

The first identification module 151 is configured to set the combined package identifier to a predetermined value, the predetermined value indicating that the request package is a combined package of the at least two sub-request packages;

The second identification module 152 is configured to obtain the number of sub-request packets, and write the number into the identifier of the number of sub-request packets.

Optionally, as shown in Figure 16, the judging unit includes:

The first obtaining module 161 is used to obtain the destination node identification carried in each request packet to be sent;

The first judging module 162 is used to judge whether the destination node identifiers of the request packets to be sent are the same;

The processing unit also includes:

The third identification module 163 is configured to extract the destination node identifier in the sub-request packet, and place the destination node identifier in the common header of the merge request packet.

Optionally, as shown in Figure 17, the judging unit includes:

The second obtaining module 171 is used to obtain the memory address of the destination node carried in each request packet to be sent;

The first search module 172 is configured to search the address attribution table for the destination node identifier to which the destination node memory address belongs according to the destination node memory address, and the address attribution table is the correspondence between the destination node memory address and the destination node identifier Relational tables;

The second judging module 173 is configured to judge whether the destination node identifiers of the request packets to be sent are the same.

Optionally, the sub-request package included in the merge request package obtained by the processing unit includes a write request sub-packet and/or a read request sub-packet, and the write request sub-packet header includes a write request identifying the write request sub-packet. Request identification, the write request information in the write request subpackage includes the data to be written to the destination node and the memory address of the destination node to be written, and the header of the read request subpackage includes a The read request identifier, the read request information in the read request subpackage includes the memory address of the destination node where the data to be read is located, as shown in Figure 18, the device also includes:

The receiving unit 181 is configured to receive a write request response message sent by the destination node, where the write request response message includes a write request identifier corresponding to the write request subpacket whose write request is successful; and/or, used to receive the Describe the read request response message sent by the destination node, the read request response message includes the read request identifier corresponding to the read request subpacket for which the read request is successful, and the read payload, the payload is the read request The data corresponding to the remote memory address in the subpackage.

Optionally, as shown in Figure 19, the judging unit further includes:

The second search module 191 is configured to search the routing table for the routing port corresponding to the destination node ID according to the destination node ID corresponding to each request packet to be sent, and the routing table includes the route corresponding to the destination node ID port, the routing port is a port on the path of the request packet to be sent to the destination node;

The third judging module 192 is configured to judge whether the routing ports of the request packets to be sent are the same.

Correspondingly, the embodiment of the present invention also provides a memory data read-write processing device 20, and the memory data read-write processing device 20 may exist in a computing node, or may be a computing node, as shown in FIG. 20 , The device includes a receiving unit 201 , a judging unit 202 , a determining unit 203 , an extracting unit 204 and a processing unit 205 .

A receiving unit 201, configured to receive a request packet;

The judging unit 202 is configured to judge whether the received request packet carries a merged packet identifier, the merged packet identifier is used to indicate that the request packet is a merged request packet composed of at least two sub-request packets, and the sub-request packet includes: read request subpackage, and/or write request subpackage;

A determining unit 203, configured to determine that the request package is a merge request package when the request package carries a merge package identifier, the merge request package includes a public header and a sub-request package, and the common header includes a merge package identifier , a sub-request packet number identifier and a source node identifier, the sub-request packet number identifier is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packet includes a sub-request packet header and sub-request information ;

Extraction unit 204, configured to extract the number of sub-requests corresponding to the number of sub-request packets in the merge request packet according to the number of sub-request packets carried in the merge request packet and the header of each sub-request packet Bag;

The processing unit 205 is configured to process the corresponding sub-request packet according to the header of each sub-request packet and the sub-request information.

An embodiment of the present invention provides a device for reading and writing memory data, which determines whether the received request packet carries a merged package identifier, and when the request packet carries the merged packet identifier, determines that the request packet is a merged request packet, Then according to the number of sub-request packets carried in the merge request packet and the header of each sub-request packet, extract the sub-request packets corresponding to the number of sub-request packets in the merge request packet, and finally according to the Describe the header and sub-request information of each sub-request packet, and process the corresponding sub-request packet. The embodiment of the present invention solves the problem in the prior art that multiple remote data access operations with small data volumes waste more overhead on data request packets, resulting in low transmission efficiency, and improves link utilization efficiency.

Optionally, as shown in Figure 21, the processing unit includes:

The first processing module 211 is configured to, according to the write request identifier in the header of the write request sub-packet, the data to be written in the write request information of the write request sub-packet, and the memory address of the destination node to be written, write the data to be written Writing data into the memory address of the destination node;

The second processing module 212 is configured to, according to the read request identifier in the header of the read request sub-packet and the memory address of the destination node where the data to be read in the read request information of the read request sub-packet is located, on the destination node The data to be read is read from the memory address.

Optionally, as shown in Figure 22, the device further includes:

The identifier writing unit 221 is configured to write the source node identifier into the sub-request packet header of each sub-request packet, so that the destination node sends a response message to the source node.

Optionally, the judging unit is further configured to judge whether the source node identifiers corresponding to the response messages to be sent are the same;

The processing unit is further configured to pack the to-be-sent response messages with the same source node identifier as sub-response messages into a combined response message package, the combined response message package including the write request sub-response message and/or or read request sub-response message;

As shown in Figure 23, the device also includes:

The sending unit 231 is configured to send the combined response message packet to the source node corresponding to the source node identifier.

Optionally, when the destination node is a routing port, the processing unit is further configured to extract the source node identifier in the common header of the merge request packet, and write it into the header of each sub-request packet.

For the operation process of the above-mentioned device 14 and device 20, refer to the processing process of the above-mentioned method for reading and writing memory data.

The above-mentioned reading and writing processing device for memory data shown in FIGS. 14 to 19 can be realized based on the hardware structure of a computer or other network equipment. As shown in FIG. 24 , the hardware structure of the reading and writing processing device 24 for memory data includes a memory 241, Transceiver 242 , processor 243 and bus 244 .

Wherein, the processor 243 , the memory 241 and the transceiver 242 are connected through a bus 244 .

The memory 241 may be a read only memory (Read Only Memory, ROM), a static storage device, a dynamic storage device or a random access memory (Random Access Memory, RAM). The memory 241 may store instructions for an operating system and other application programs as well as application data. The instructions stored in the memory 241 are executed by the processor 243 .

In the embodiment of the present invention, the memory 241 may be used to store instructions for implementing each functional unit in the embodiments shown in FIG. 14 to FIG. 19 .

The transceiver 242 is used to realize the communication between the memory data read-write processing device 24 and other devices.

In the embodiment of the present invention, the transceiver 242 is used to send a merge request packet and receive a response message.

The processor 243 may adopt a general-purpose central processing unit (Central Processing Unit, CPU), a microprocessor, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits for executing related programs.

In the embodiment of the present invention, the processor 243 is used to execute the instructions in the memory 241, specifically:

Obtaining each request packet to be sent, the request packet to be sent includes: a read request packet, and/or a write request packet;

Analyzing the destination node identifiers of the respective request packets to be sent, and judging whether the destination node identifiers of the respective request packets to be sent are the same;

The at least two request packets to be sent with the same destination node identifier are respectively used as sub-request packets, and packaged into a merge request packet, and the merge request packet includes a public header and at least two sub-request packets, and the public header includes Contains a combined package identifier, a sub-request package number identifier and a source node identifier, the combined package identifier is used to indicate that the request package is a combined package of the at least two sub-request packages, and the sub-request package number identifier is used In order to indicate the number of sub-request packets carried by the merge request packet, the sub-request packet includes a sub-request packet header and sub-request information;

The merge request packet is sent to the destination node corresponding to the destination node identifier through the transceiver 242, and the merge request packet is used to instruct the destination node to perform read and/or write operations on memory data of the destination node.

A read-write processing device for memory data provided by an embodiment of the present invention obtains each request packet to be sent, and analyzes the destination node identifier of each request packet to be sent, and judges whether the destination node identifier of each request packet to be sent is The same, then the at least two request packets to be sent with the same destination node identifier are respectively used as sub-request packets, packaged into a merge request packet, and finally the merge request packet is sent to the destination node corresponding to the destination node identifier , the merge request packet is used to instruct the destination node to perform read and/or write operations on memory data of the destination node. The embodiment of the present invention solves the problem in the prior art that multiple remote data access operations with small data volumes waste more overhead on data request packets, resulting in low transmission efficiency, and improves link utilization efficiency.

Wherein, for the specific processing process in the processor 243, reference may be made to the method embodiment part and the device embodiment part shown in FIG. 14 to FIG. 19 , which will not be repeated here.

The bus 244 may include a path for transferring information between various components of the processing device (eg, processor 243 , memory 241 , and transceiver 242 ) for reading and writing data in memory.

The read-write processing device of memory data shown in above-mentioned Fig. 20 to Fig. 23 all can realize based on the hardware structure of computer, as shown in Figure 25, the hardware structure of the read-write processing device 25 of memory data comprises memory 251, transceiver 252 , processor 253 and bus 254.

Wherein, the processor 253 , the memory 251 and the transceiver 252 are communicatively connected through the bus 254 .

The memory 251 may be a read only memory (Read Only Memory, ROM), a static storage device, a dynamic storage device or a random access memory (Random Access Memory, RAM). The memory 251 may store instructions for an operating system and other application programs as well as application data. The instructions stored in the memory 251 are executed by the processor 253 .

In the embodiment of the present invention, the memory 251 may be used to store instructions for implementing each functional unit in the embodiments shown in FIG. 20 to FIG. 23 .

The transceiver 252 is used to realize the communication between the memory data reading and writing processing device 25 and other devices.

In the embodiment of the present invention, the transceiver 252 is used to receive a merge request packet and send a response message.

The processor 253 may adopt a general-purpose central processing unit (Central Processing Unit, CPU), a microprocessor, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits for executing related programs.

In the embodiment of the present invention, the processor 253 is used to execute the instructions in the memory 251, specifically:

Receive the request packet through the transceiver 252;

Judging whether the received request packet carries a combined package identifier, the combined package identifier is used to indicate that the request package is a combined request package composed of at least two sub-request packages, the sub-request package includes: a read request sub-packet, and/ Or write request subpackage;

When the request package carries a merge package identifier, it is determined that the request package is a merge request package, and the merge request package includes a public header and a sub-request package, and the public header includes a merge package identifier and the number of sub-request packages An identifier and a source node identifier, the identifier of the number of sub-request packets is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packets include sub-request packet headers and sub-request information;

According to the number of sub-request packets carried in the merge request packet and the header of each sub-request packet, extracting the number of sub-request packets corresponding to the number of sub-request packets in the merge request packet;

According to the header of each sub-request packet and the sub-request information, the corresponding sub-request packet is processed.

An embodiment of the present invention provides a device for reading and writing memory data, which determines whether the received request packet carries a merged package identifier, and when the request packet carries the merged packet identifier, determines that the request packet is a merged request packet, Then according to the number of sub-request packets carried in the merge request packet and the header of each sub-request packet, extract the sub-request packets corresponding to the number of sub-request packets in the merge request packet, and finally according to the Describe the header and sub-request information of each sub-request packet, and process the corresponding sub-request packet. The embodiment of the present invention solves the problem in the prior art that multiple remote data access operations with small data volumes waste more overhead on data request packets, resulting in low transmission efficiency, and improves link utilization efficiency.

Wherein, for the specific processing process in the processor 253, reference may be made to the method embodiment part and the device embodiment part shown in FIG. 20 to FIG. 23 , which will not be repeated here.

The bus 254 may include a path for transferring information between various components of the processing device (eg, processor 253 , memory 251 , and transceiver 252 ) for reading and writing data in memory.

Those skilled in the art can clearly understand that for the convenience and brevity of description, only the division of the above-mentioned functional modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional modules according to needs. The internal structure of the device is divided into different functional modules to complete all or part of the functions described above. For the specific working process of the above-described system, device, and unit, reference may be made to the corresponding process in the foregoing method embodiments, and details are not repeated here.

In the several embodiments provided in this application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be Incorporation may either be integrated into another system, or some features may be omitted, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor (processor) execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and other media that can store program codes. .

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Anyone skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present invention. Should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims (22)

1. A method for reading and writing processing of memory data, characterized in that, comprising: Obtaining each request packet to be sent, the request packet to be sent includes: a read request packet, and/or a write request packet; Analyzing the destination node identifiers of the respective request packets to be sent, and judging whether the destination node identifiers of the respective request packets to be sent are the same; The at least two request packets to be sent with the same destination node identifier are respectively used as sub-request packets, and packaged into a merge request packet, and the merge request packet includes a public header and at least two sub-request packets, and the public header includes Contains a merged package identifier, a sub-request package number identifier and a source node identifier, the merged package identifier is used to indicate that the request package is a combined package of the at least two sub-request packages, and the sub-request package number identifier is used In order to indicate the number of sub-request packets carried by the merge request packet, the sub-request packet includes a sub-request packet header and sub-request information; Sending the merge request packet to the destination node corresponding to the destination node identifier, where the merge request packet is used to instruct the destination node to perform read and/or write operations on the memory data of the destination node. 2. The method for reading and writing processing of memory data according to claim 1, wherein said packaging as a merge request package comprises: Setting the combined package identifier to a predetermined value, the predetermined value indicating that the request package is a combined package of the at least two sub-request packages; Obtain the number of sub-request packets, and write the number into the number identifier of the sub-request packets. 3. The method for reading and writing processing of memory data according to claim 1 or 2, characterized in that, the parsing of the destination node identification of each request packet to be sent, and judging the destination node of each request packet to be sent Whether the identification is the same includes: Obtain the destination node identifier carried in each request packet to be sent; Judging whether the destination node identifiers of the respective request packets to be sent are the same; The packages packaged as a merge request include: Extracting the target node identifier in the sub-request package, and placing the target node identifier in the common header of the merge request package. 4. The method for reading and writing processing of memory data according to claim 1 or 2, characterized in that, the parsing of the destination node identification of each request packet to be sent, and judging the destination node of each request packet to be sent Whether the identification is the same includes: Obtain the memory address of the destination node carried in each request packet to be sent; According to the memory address of the destination node, look up the destination node identification to which the memory address of the destination node belongs in the address attribution table, where the address attribution table is a correspondence table between the memory address of the destination node and the destination node identification; It is judged whether the identifiers of the destination nodes of the request packets to be sent are the same. 5. The method for reading and writing processing of memory data according to claim 3, wherein the sub-request package includes a write request sub-packet and/or a read request sub-packet, and the write request sub-packet header includes a The write request identifier of the request subpackage, the write request information in the write request subpackage includes the data to be written to the destination node and the memory address of the destination node to be written, and the header of the read request subpackage includes the identification Describe the read request identifier of the read request subpackage, the read request information in the read request subpackage includes the memory address of the destination node where the data to be read is located, and the method further includes: Receive a write request response message sent by the destination node, where the write request response message includes a write request identifier corresponding to the write request subpacket whose write request is successful; and/or, Receive the read request response message sent by the destination node, the read request response message includes the read request identifier corresponding to the read request subpacket with a successful read request, and the read payload, the payload is the Read the data corresponding to the remote memory address in the request subpacket. 6. The method for reading and writing processing of memory data according to any one of claims 1, 2, and 5, characterized in that, the parsing of the destination node identifiers of each of the request packets to be sent, and judging that each of the request packets to be sent Whether the destination node IDs of the request packets are the same includes: According to the destination node identifier corresponding to each request packet to be sent, the routing port corresponding to the destination node identifier is searched in the routing table, and the routing table includes the routing port corresponding to the destination node identifier, and the routing port is the routing port corresponding to the destination node identifier. Describe the port on the path where the request packet to be sent reaches the destination node; It is judged whether the routing ports of the request packets to be sent are the same. 7. A method for reading and writing data in memory, comprising: Receive request packet; Judging whether the received request packet carries a combined package identifier, the combined package identifier is used to indicate that the request package is a combined request package composed of at least two sub-request packages, the sub-request package includes: a read request sub-packet, and/ Or write request subpackage; When the request package carries a merge package identifier, it is determined that the request package is a merge request package, and the merge request package includes a public header and a sub-request package, and the public header includes a merge package identifier and the number of sub-request packages An identifier and a source node identifier, the identifier of the number of sub-request packets is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packets include sub-request packet headers and sub-request information; According to the number of sub-request packets carried in the merge request packet and the header of each sub-request packet, extracting the number of sub-request packets corresponding to the number of sub-request packets in the merge request packet; According to the header of each sub-request packet and the sub-request information, the corresponding sub-request packet is processed. 8. The method for reading and writing processing of memory data according to claim 7, wherein, according to said each sub-request packet header and sub-request information, processing a corresponding sub-request packet includes: Write the data to be written into the destination according to the write request identifier in the header of the write request subpacket, the data to be written in the write request information of the write request subpacket, and the memory address of the destination node to be written. node memory address; and/or, According to the read request identifier in the header of the read request sub-packet and the memory address of the destination node where the data to be read in the read request information of the read request sub-packet is located, read the data to be read in the memory address of the destination node read data. 9. The method for reading and writing processing of memory data according to claim 8, characterized in that, after the sub-request packets corresponding to the number of sub-request packets identified in the merge request packet are extracted, further include: The source node identifier is written into the sub-request packet header of each sub-request packet, so that the destination node sends a response message to the source node. 10. The method for reading and writing memory data according to claim 9, wherein the destination node sending a response message to the source node comprises: Judging whether the source node identifiers corresponding to the response messages to be sent are the same; Packing the to-be-sent response messages with the same source node identifier as sub-response messages into a combined response message package, the combined response message package includes a write request sub-response message and/or a read request sub-response message; Send the combined response message packet to the source node corresponding to the source node identifier. 11. The method for reading and writing processing of memory data according to claim 7, characterized in that, when the destination node is a routing port, in the extraction of the combined request package, the number of sub-request packages corresponding to the sub-request package After the number of subrequest packets, it also includes: Extract the source node identifier in the public header of the merge request package, and write it into the header of each sub-request package. 12. A device for reading and writing data in memory, comprising: An acquisition unit, configured to acquire each request packet to be sent, wherein the request packet to be sent includes: a read request packet, and/or a write request packet; A judging unit, configured to parse the destination node identifiers of the respective request packets to be sent, and judge whether the destination node identifiers of the respective request packets to be sent are the same; A processing unit, configured to pack the at least two request packets to be sent with the same destination node identifier as sub-request packets into a combined request package, wherein the combined request package includes a common header and at least two sub-request packages, The public packet header includes a merged package identifier, a sub-request packet number identifier, and a source node identifier, and the merged package identifier is used to indicate that the request package is a combined package of the at least two sub-request packages, and the sub-request The packet number identifier is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packets include sub-request packet headers and sub-request information; A sending unit, configured to send the merge request packet to a destination node corresponding to the destination node identifier, where the merge request packet is used to instruct the destination node to perform read and/or write operations on the memory data of the destination node . 13. The device for reading and writing processing of memory data according to claim 12, wherein the processing unit comprises: A first identification module, configured to set the combined package identifier to a predetermined value, the predetermined value indicating that the request package is a combined package of the at least two sub-request packages; The second identification module is configured to obtain the number of sub-request packets, and write the number into the identifier of the number of sub-request packets. 14. The device for reading and writing memory data according to claim 12 or 13, characterized in that, The judging unit includes: The first obtaining module is used to obtain the destination node identification carried in each request packet to be sent; The first judging module is used to judge whether the destination node identifiers of the request packets to be sent are the same; The processing unit also includes: The third identification module is configured to extract the destination node identifier in the sub-request package, and place the destination node identifier in the common header of the merge request package. 15. The device for reading and writing data in memory according to claim 12 or 13, wherein the judging unit comprises: The second obtaining module is used to obtain the memory address of the destination node carried in each request packet to be sent; The first search module is used to search the destination node identifier to which the destination node memory address belongs in the address attribution table according to the destination node memory address, and the address attribution table is the corresponding relationship between the destination node memory address and the destination node identifier surface; The second judging module is configured to judge whether the destination node identifiers of the request packets to be sent are the same. 16. The device for reading and writing memory data according to claim 14, wherein the sub-request packets included in the merge request packet obtained by the processing unit include a write request sub-packet and/or a read request sub-packet , the header of the write request sub-packet includes a write request identifier identifying the write request sub-packet, and the write request information in the write request sub-packet includes the data to be written to the destination node and the destination node to be written memory address, the header of the read request sub-packet includes a read request identifier identifying the read request sub-packet, the read request information in the read request sub-packet includes the memory address of the destination node where the data to be read is located, the The device also includes: A receiving unit, configured to receive a write request response message sent by the destination node, where the write request response message includes a write request identifier corresponding to the write request subpacket whose write request is successful; and/or, used to receive the The read request response message sent by the destination node, the read request response message includes the read request identifier corresponding to the read request subpacket for which the read request is successful, and the read payload, the payload is the read request subpacket The data corresponding to the remote memory address in the packet. 17. The device for reading and writing memory data according to any one of claims 12, 13, and 16, wherein the judging unit further comprises: The second search module is configured to search the routing table for the routing port corresponding to the destination node ID according to the destination node ID corresponding to each request packet to be sent, and the routing table includes the routing port corresponding to the destination node ID , the routing port is a port on the path where the request packet to be sent reaches the destination node; The third judging module is used for judging whether the routing ports of the request packets to be sent are the same. 18. A device for reading and writing data in memory, comprising: a receiving unit, configured to receive the request packet; A judging unit, configured to judge whether the received request packet carries a combined package identifier, the combined package identifier used to indicate that the request package is a combined request package composed of at least two sub-request packages, and the sub-request package includes: a read request subpackage, and/or write request subpackage; A determining unit, configured to determine that the request package is a merge request package when the request package carries a merge package identifier, the merge request package includes a public header and a sub-request package, and the common header includes a merge package identifier, A sub-request packet number identifier and a source node identifier, the sub-request packet number identifier is used to indicate the number of sub-request packets carried by the merge request packet, and the sub-request packet includes a sub-request packet header and sub-request information; An extracting unit, configured to extract the number of sub-request packets corresponding to the number of sub-request packets in the merge request packet according to the number of sub-request packets carried in the merge request packet and the header of each sub-request packet ; The processing unit is configured to process the corresponding sub-request packet according to the header of each sub-request packet and the sub-request information. 19. The device for reading and writing processing of memory data according to claim 18, wherein the processing unit comprises: The first processing module is configured to write the data to be written according to the write request identifier in the header of the write request subpacket, the data to be written in the write request information of the write request subpacket, and the memory address of the destination node to be written. Incoming data is written to the memory address of the destination node; The second processing module is configured to, according to the read request identifier in the header of the read request sub-packet and the memory address of the destination node where the data to be read in the read request information of the read request sub-packet is located, in the memory of the destination node Read the data to be read from the address. 20. The device for reading and writing data in memory according to claim 19, wherein the device further comprises: The identifier writing unit is configured to write the source node identifier into the sub-request packet header of each sub-request packet, so that the destination node sends a response message to the source node. 21. The device for reading and writing data in memory according to claim 20, characterized in that: The judging unit is also used to judge whether the source node identifiers corresponding to the response messages to be sent are the same; The processing unit is further configured to pack the to-be-sent response messages with the same source node identifier as sub-response messages into a combined response message package, the combined response message package including the write request sub-response message and/or or read request sub-response message; The device also includes: A sending unit, configured to send the combined response message packet to the source node corresponding to the source node identifier. 22. The device for reading and writing processing of memory data according to claim 18, wherein when the destination node is a routing port, the processing unit is also used to extract the source node in the public header of the merge request packet Identifier, written in the header of each subrequest packet.