JP2010226275A - Communication equipment and communication method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To effectively utilize resources of a multicore processor when processing a packet by using the multicore processor. <P>SOLUTION: A transmission control part 112 includes a plurality of output buffers in which processing-completed data packets are stored. The output buffer is provided, for instance, in each output port. A work control part 120 divides processing for a received data packet into a plurality of subworks and allocates each of the subworks to any of a plurality of threads 131 executed on a plurality of cores 130. In the allocation, the allocation of a subwork related to an output queue the use rate of which exceeds a threshold to the thread 131, is restricted. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a packet transfer processing technique.

  Conventionally, there are various communication devices for performing packet transfer processing, which have a plurality of input ports and a plurality of output ports, and output packets input from the input ports from the output ports according to their destinations and QoS classes. Is known (see, for example, Patent Document 1).

  The communication device described in Patent Document 1 is configured by using a multi-core processor, and processes for packets belonging to different IP flows (defragment processing, IPSec decoding processing, network address conversion processing, etc.) are different processor cores. The throughput is improved by processing in parallel. Here, the multi-core processor is a microprocessor in which two or more processor cores are integrated in one package. Further, packets belonging to different IP flows are packets having different 5-tuple information about the IP header and the TCP header. The 5-tuple information refers to five pieces of information: “destination IP address”, “source IP address”, “protocol number” of the IP packet, “destination port number” of the TCP / UDP header, and “source port number”.

Special table 2008-512950 gazette

  By the way, a packet that has been processed by the processor is sent to the outside via an output queue corresponding to the output destination or QoS class. However, when the corresponding output queue is congested and a buffer overflow occurs, the processed packet is processed. Will be discarded in the output queue. In this case, since resources such as processor processing capacity and memory are used for the finally discarded packet, the resources are not effectively used. However, Patent Document 1 does not consider this at all. Such a problem is not unique to a communication device using a multi-core processor, but also occurs in a communication device using a single-core processor.

[Object of invention]
Therefore, an object of the present invention is to efficiently use resources and to efficiently process a packet.

The first communication device according to the present invention is:
Processor cores,
Multiple output queues where processed packets are stored;
A work control unit that divides the processing for the received packet into a plurality of sub-work, and assigns each sub-work to a thread executed on the processor core;
When there is a congested output queue among the plurality of output queues, the work control unit assigns a thread to a subwork related to a packet having the congested output queue as a store destination. Limit.

A first communication method according to the present invention includes:
A first step in which the work control unit divides the processing for the received packet into a plurality of sub-work;
When the work control unit assigns a sub work to any one of a plurality of threads executed on the processor core, an output queue that is congested in a plurality of output queues in which processed packets are stored The second step of restricting thread allocation for subwork related to a packet having the congested output queue as a store destination.

  According to the present invention, it is possible to direct allocation of resources such as processor processing capacity and memory, which has been aimed at processing a packet having a high possibility of being discarded, to processing of other packets having a low possibility of being discarded. Thus, it is possible to obtain an effect that the resources provided in the communication device can be effectively used.

FIG. 1 is a block diagram showing a configuration example of a first embodiment of a communication apparatus according to the present invention. FIG. 2 is a diagram for mainly explaining the processing of the reception control unit 111. FIG. 3 is a diagram showing the relationship among the work table, the packet element table, and the action table. FIG. 4 is a block diagram illustrating a configuration example of the transmission control unit 112. FIG. 5 is a block diagram illustrating a configuration example of the work control unit 120. FIG. 6 is a diagram illustrating a configuration example of the packet element table 210. FIG. 7 is a diagram illustrating a configuration example of the work table 211. FIG. 8 is a flowchart showing a processing example of the thread 131. FIG. 9 is a block diagram showing a configuration example of the work control unit 120 used in the second embodiment of the communication apparatus according to the present invention. FIG. 10 is a diagram illustrating a configuration example of the work table 211 used in the second embodiment.

  Next, a first embodiment of the present invention will be described in detail with reference to the drawings.

  Referring to FIG. 1, the communication apparatus according to the first embodiment of the present invention performs a data packet transfer process, and includes a reception interface unit 101 that receives a data packet (sometimes simply referred to as a packet). And a transmission interface unit 102 for transmitting the processed data packet. Although not shown, the reception interface unit 101 and the transmission interface unit 102 include a plurality of reception ports and transmission ports, respectively.

  The reception interface unit 101 and the transmission interface unit 102 are connected to the reception control unit 111 and the transmission control unit 112 in the multi-core processor unit 105 via switch units 103a and 103b, respectively. Here, the switch unit 103a has a function of determining which multi-core processor unit among the plurality of multi-core processor units 150 is to output the data packet. In this case, the output destination multi-core processor unit is determined based on the value calculated using the hash function from information such as the “destination IP address”, “source IP address”, “protocol number”, etc. of the IP data packet. It is possible to take the following method. That is, it is possible to determine the output multicore processor unit according to the IP flow of the data packet. The switch unit 103b has a function of outputting the data packet from each multi-core processor unit 150 to the transmission interface unit 102.

  The reception control unit 111 and the transmission control unit 112 are connected to the internal bus interface 160 of the multi-core processor unit 150, and each has a function of outputting a data packet passed from the reception interface unit 101 to the bus interface 160, and a bus. It has a function of passing a data packet sent to the outside of the multi-core processor unit 150 transmitted via the interface 160 to the transmission interface unit 102 via the switch 103b.

  A plurality of cores 130 are arranged inside the multi-core processor unit 150, and each core 130 is connected to the work control unit 120, the main memory 140, the reception control unit 111, and the transmission control unit 112 via the bus interface 160, respectively. It is connected. Each core 130 is configured to be able to execute a plurality of threads 131 simultaneously. The reception control unit 111 stores a newly received data packet in the main memory 140, and passes the data packet processing to the work control unit 120 in the form of a work, and requests assignment of the thread 131 to the work. Have

  The work control unit 120 has a function of dividing a work into a plurality of subwork and assigning a thread 131 to be executed on the core 130 to each subwork. Here, as the sub work, for example, a process for determining which output queue of the plurality of output queues included in the transmission control unit 112 stores the data packet, a defragment process, an IPSec decoding process, an IPSec There are encryption processing and network address change processing. In addition, when there is a congested output queue among the plurality of output queues provided in the transmission control unit 112, the work control unit 120 performs sub work related to the packet that stores the output queue. Has a function of restricting the assignment of threads 131. This function can prevent resources from being consumed by packets that are likely to be discarded in the future.

  When a sub work is assigned by the work control unit 120, the thread 131 executed on each core 130 executes a processing instruction corresponding to the assigned sub work. Input information for work processing and an instruction (action) to be performed are obtained from the main memory 140, and output information as a processing result is written in the main memory 140. Further, it is possible to issue a further subdivided subwork obtained in the course of processing to the work control unit 120 and request assignment of another thread 131 to the subwork.

  The transmission control unit 112 is provided with an output queue for each transmission destination (for each output port) of each data packet or for each QoS class, and after scheduling the output timing, the transmission control unit 112 is connected to the transmission interface unit 102 via the switch unit 103b. Send. Further, the transmission control unit 112 has a function of passing information indicating the output queue in which congestion occurs to the work control unit 120 when congestion occurs in any of the plurality of output queues.

  The data packet received by the reception interface unit 101 is transferred to the reception control unit 111 of the multi-core processor unit 150 via the switch unit 103a. Referring to FIG. 2, the reception control unit 111 assigns unique ID information called an element ID and unique ID information called a work ID to a newly received data packet 201. The numbered element ID is used to identify the data packet 201, and the work ID is the highest level sub-work when the processing (work) for the data packet 201 is hierarchically divided into a plurality of sub-work. Used to identify.

  Further, the reception control unit 111 stores the received data packet 201 and information including the numbered element ID and work ID in the main memory 140 in a form called a packet element table 210, and the numbered work ID and Information including the element ID and the processing content of the work is stored in the main memory 140 in a form called a work table 211. Furthermore, the reception control unit 111 transmits the numbered element ID and work ID to the transmission control unit 112 and the work control unit 120, respectively. The packet element table 210 and the work table 211 will be described later in detail with reference to FIGS.

  The transmission control unit 112 stores the element ID passed from the reception control unit 111 in the packet queue group 220 and performs scheduling using the packet transmission scheduler 230. When the data packet to be transmitted is determined, the data packet is read from the main memory 140 and sent to the transmission interface unit 102. The detailed configuration and operation of the transmission control unit 112 will be described later with reference to FIG.

  Further, the work control unit 120 stores the work ID passed from the reception control unit 111 in the work queue group 221, and then determines which sub-work is assigned to which thread 131 using the work scheduler 231. Then, the work ID of the sub work determined as the assignment destination is passed to the assignment destination thread 131. Also, the work control unit 120 performs the same processing when a sub work ID is passed from the thread 131 operating on the core 130. The detailed configuration and operation of the work control unit 120 will be described later in detail with reference to FIG.

  Here, with reference to FIG. 3, the relationship between the packet element table 210 that manages the data packet 201 and the work management table 211 that manages each sub-work for the data packet 201 will be described.

  When receiving the data packet 201, the reception control unit 111 receives the packet element table 210 for managing the data packet 201, and the sub-work of the highest layer (first layer) among the sub work for the data packet 201. A work table 211 for managing the work is generated. In these two tables 210 and 211, the same element ID and work ID 301 are set, and both tables 210 and 211 are associated with each other by the element ID. In the work table 211 of each sub work related to the data packet 201, a work ID unique to the work table is set, and a work ID indicating a parent and a child of the own work table is set. The hierarchical relationship can be recognized.

  Specific processing contents performed in each subwork are called actions, and processing instructions are described in an action table 310 stored in the main memory 140. In the action table 310, processing instructions related to each sub work are registered in association with different action IDs 302, and in the work table 211 of each sub work, an action ID indicating a processing instruction to be executed in that sub work is stored. Is registered. Therefore, it is possible to associate a subwork with a processing instruction executed in the subwork by the action ID.

  Referring to FIG. 4, the transmission control unit 112 includes a packet queue group 220 and a packet transmission scheduler 230.

  The packet queue group 220 includes an output destination distribution packet queue 402, a process waiting packet output queue group 401, and an output queue group 403.

  The waiting packet output queue group 401 includes a waiting packet output queue for each data packet output interface (output port) or QoS class, and the output queue group 403 also has an output queue for each output interface or QoS class. I have. In FIG. 4, the process waiting packet output queue group 401 and the output queue group 403 are each composed of N process waiting packet output queues and output queues. An identifier called the output queue ID of #N is assigned. The element IDs output from the process waiting packet output queues whose output queue IDs are # 1 to #N are input to output queues having the same output queue ID. Also, as shown in FIG. 4, an output destination distribution packet queue 402 is arranged in the preceding stage of the processing-waiting packet output queue group 401, and the element ID is stored for a data packet for which the output queue ID has not yet been determined. To do.

  The element ID of the data packet 201 sent from the reception control unit 111 to the transmission control unit 112 is first stored in the output destination distribution packet queue 402. The queue 402 has a function of distributing element IDs to any queue in the processing-waiting packet output queue group 401 in accordance with FIFO (First In First Out) principles. When the output destination distribution packet queue 402 outputs an element ID from the queue, a packet element table corresponding to the output element ID (the same element ID as the above element ID is set as information for identifying the table). Packet element table (see FIG. 6). If the output queue ID has already been recorded in this packet element table (if the output queue ID valid bit is “1”), the corresponding element ID is distributed to the corresponding queue in the processing-waiting packet output queue group 401. . For example, if # 3 is recorded as the output queue ID in the packet element table, out of the N queues in the process waiting packet output queue group 401, the element is output to the process waiting packet output queue whose output queue ID is # 3. Assign IDs. If the output queue ID is not recorded in the packet element table, it continues to wait until it is recorded. Note that the recording of the output queue ID in the packet element table 210 is performed by the thread 131 executing in the core 130. The operation of the thread 131 will be described in detail later with reference to FIG.

  The element ID stored in each queue in the process waiting packet output queue group 401 is transmitted to the output queue in the output queue group 403 having the same output queue ID according to the FIFO principle. The process waiting packet output queue group 401 checks the corresponding packet element table when outputting the element ID from the queue. When all sub-work related to the corresponding data packet is completed and the work completion bit is “1”, the element ID is transferred to the corresponding output queue in the output queue group 403. When the work completion bit of the corresponding packet element table is “0”, the process continues to wait until the work completion bit becomes “1”. The work completion bit is set by the thread 131 running on the core 130.

  The packet transmission scheduler 230 sequentially reads out the element ID stored at the head of each output queue constituting the output queue group 403, reads out the data packet corresponding to the read element ID from the main memory 140, and Transmit to the corresponding output port (the output port corresponding to the output queue from which the element ID was read). Further, the packet transmission scheduler 230 monitors the usage rate of the output queue in order to detect a congested output queue in the output queue 403 group. When an output queue whose usage rate exceeds a predetermined threshold occurs, it is determined that congestion has occurred in the output queue, and a congestion notification including the output queue ID of the output queue is sent to the work control 120. Send to. When an output queue whose usage rate has returned to a threshold value or less occurs, a congestion resolution notification including the output queue ID of the output queue is transmitted to the work control unit 120. Note that the threshold can be set to about 80%, for example.

  Referring to FIG. 5, the work control unit 120 includes a work output queue group 501 and an output destination distribution work queue 502 that form the work queue group 221 illustrated in FIG. 2, a work scheduler 231, and round robin function units 511 and 512. And an output queue ID checkpoint 521. The work output queue group 501 is composed of N work output queues similarly to the process waiting packet queue group 401 and the output queue group 403, and can be identified by the output queue IDs # 1 to #N.

  When a work ID is input from the thread 131 running on the reception control unit 111 or the core 130, the work control unit 120 generates a packet element table to which the work ID belongs (generates a subwork indicated by the work ID). In accordance with the state of the original packet element table), a queue to be a distribution destination (store destination) of this work ID is determined. Specifically, if the output queue ID (see FIG. 6) has been recorded in the packet element table 210, the queue specified by the output queue ID among the N queues constituting the work output queue group 501. Distribute the work ID. On the other hand, if the output queue ID is not recorded in the packet element table, the work ID is distributed to the output destination assignment work queue 502.

  The N work output queues that make up the work output queue group 501 and the work IDs stored in the output destination distribution work queue 502 pass through the round robin function unit 511 and the round robin function unit 512, respectively, according to the FIFO principle. Then, it is delivered to the work scheduler 231. In other words, the work scheduler 231 checks the processing status of the threads 131 belonging to each core 130, and if it finds a thread 131 that is not processing a work, it passes through the round robin function unit 511 and the round robin function unit 512. The work ID is taken from the N work output queues constituting the work output queue group 501 or the output destination assignment work queue 502, and the work ID is assigned to the found thread 131. Here, the output queue ID check point 521 checks again the state of the packet element table of the data packet to which the corresponding work ID belongs when the work ID information is drawn from the output destination work queue 502, and the output queue ID Is recorded in the packet element table, the work ID is redistributed to the work output queue 501. If the output queue ID is not already recorded in the packet element table, the work scheduler 231 accepts the work ID as it is.

  The work control unit 120 performs the above-described process when the usage rates of the N output queues constituting the output queue group 403 in the transmission control unit 112 are all equal to or less than the threshold value. On the other hand, when an output queue whose usage rate exceeds the threshold value is generated, the output of the work ID from the work output queue corresponding to the output queue is limited. The reason for this is that if the output queue of data packets is in a congested state for some reason, the corresponding output queue may overflow. When the corresponding output queue overflows and the corresponding data packet is discarded, the processing capacity of the multi-core processor spent on the work corresponding to the corresponding data packet is wasted. Therefore, in this embodiment, when an output queue that exceeds a certain threshold with a usage rate among N output queues constituting the output queue group 403 of data packets occurs, the corresponding output queue 403 is in a congested state. By assigning the thread 131 to the processing (sub work) for a data packet that is likely to be discarded in the future by limiting the output of the work ID from the work output queue corresponding to the corresponding output queue (Assignment) can be postponed, and processing (subwork) for other packets that are less likely to be discarded can be assigned to the thread 131.

  Here, an operation when an output queue having a usage rate exceeding a threshold value occurs will be described with a specific example. Now, for example, it is assumed that the usage rate of the output queue with the output queue ID # 1 out of N output queues constituting the output queue group 403 exceeds the threshold. In the following description, an output queue whose output queue ID is #j, a work output queue, or the like may be referred to as an output queue #j or a work output queue #j. When the packet transmission scheduler 230 detects that the usage rate of the output queue # 1 exceeds the threshold value, the packet transmission scheduler 230 transmits a congestion notification including the output queue ID “# 1” to the work control unit 120.

  When the work scheduler 231 in the work control unit 120 receives a congestion notification including the output queue ID “# 1” from the transmission control unit 112, the N work output queues # constituting the work output queue group 501 are sent. Limit the output of work ID from work output queue # 1 among 1 to #N. Specifically, for example, the round robin function unit 512 is instructed to prohibit reading of the work ID from the work output queue # 1. As a result, the work ID is not output from the work output queue # 1. Alternatively, for example, the round robin function unit 512 is instructed to read the work ID stored only once per M scans (M is an integer of 2 or more) from the work output queue # 1. As a result, the number of work IDs output from the work output queue # 1 is 1 / M of the number of work IDs output from the other work output queues # 2 to #N.

  Further, the packet transmission scheduler 230 in the transmission control unit 112 transmits a congestion resolution notification including the output queue ID “# 1” to the work control unit 120 when the usage rate of the output queue # 1 returns below the threshold value. As a result, the work scheduler 231 in the work control unit 120 instructs the round robin function unit 512 to release the restriction made when the congestion notification is received. As a result, the work ID is output from the packet output queue # 1 as well as the other packet output queues # 2 to #N.

Next, the packet element table 210 and the work table 211 will be described with reference to FIGS.

  First, each component of the packet element table 210 will be described with reference to FIG.

Element ID: The element ID assigned by the reception control unit 111 when the data packet 201 is received is set.
Element ID valid bit: Displays whether the element ID is valid, that is, whether the corresponding packet element table 210 is valid.

Work ID: The work ID assigned by the reception control unit 111 when the data packet 201 is received is set.
• Work ID valid bit: Displays whether the work ID is valid.

-Work processing result: The work processing result is set.
Work completion bit: Displays whether or not the work related to the data packet 201 has been completed.

Output queue ID: For the packet queue 220 of the transmission control unit 112, describe ID information indicating from which output queue 403 the corresponding packet is output [output queue ID to be stored in the data packet 201 (output queue ID) is set].
• Output queue ID valid bit: Displays whether the output queue ID is valid.

Packet queue type display field: The element ID of the corresponding packet element table 210 is assigned to any of the output destination distribution packet queue 402, the processing waiting packet output queue group 401 or the output queue group 403 provided in the transmission control unit 112. Displays whether it is contained.

Packet queue preceding element ID: In the packet queue group 220, the element ID accommodated prior to the preceding stage of the element ID of the data packet that is the generation source of the packet element table 210 is displayed.
Packet queue preceding element ID valid bit: In the packet queue group 220, it is determined whether or not there is an element ID accommodated before the element ID of the data packet that is the generation source of the packet element table 210. indicate.

Packet Queue Subsequent Element ID: In the packet queue group 220, the element ID accommodated after the element ID of the data packet that is the generation source of the packet element table 210 is displayed.
Packet Queue Subsequent Element ID Valid Bit: Indicates whether or not there is an element ID subsequently stored in the packet queue group 220 following the element ID of the data packet that is the generation source of this packet element table 210 To do.

Data packet: The contents of the data packet are stored.

  Next, the components of the work table 211 will be described with reference to FIG.

Work ID: ID information for specifying a sub work (corresponding sub work) managed by the work table 211.
Work ID valid bit: Indicates whether the work ID is valid, that is, whether the information in the corresponding work table 211 is valid.

Element ID: An element ID indicating a data packet that is a generation source of the corresponding subwork is set. In other words, the element ID of the data packet to be processed by the corresponding subwork is set.
• Element ID valid bit: Displays whether the element ID is valid.

Action ID: Indicates ID information of the action table 310 (see FIG. 3) indicating the contents of the processing (action) actually performed by the subwork.
• Action ID valid bit: Displays whether the action ID is valid.

Assigned thread ID: The ID of the thread 131 to which the subwork is assigned is set.
Assigned thread completion bit: Displays whether or not the assigned thread ID for the corresponding sub-work has been completed and the assigned thread ID has become valid.

-Parent work ID: If there is a parent subwork from which the corresponding subwork is generated, the ID information of the parent subwork is displayed.
-Parent work ID valid bit: Displays whether parent sub work ID information is valid. Set to “0” if there is no parent subwork that has generated the subwork.

Work queue ID: Displays in which queue in the work control unit 120 the work ID of this work table 121 is stored.
Work queue assignment completion bit: Displays whether or not the work ID information of the corresponding work table 121 has been assigned to the work output queue 501.

Work queue preceding work ID: In the work queue group 221, the work ID accommodated prior to the work ID of the work table 121 is displayed.
Work queue preceding work ID valid bit: Displays whether or not there is a work ID stored in the work queue group 221 prior to the work ID of the work table 121.

Work queue subsequent work ID: In the work queue group 221, the work ID stored subsequent to the work ID of the work table 121 is displayed.
Work queue subsequent work ID valid bit: Displays whether or not there is a work ID stored subsequent to the work ID of the work table 121 in the work queue group 221.

Process progress information: The result of executing the process of the corresponding subwork is updated and recorded in this field as needed.

-Subwork ID: If a subwork exists under the subwork, add the subwork ID to the table.
・ Subwork ID issuance completion bit: Displays whether or not the subwork has been issued.

Subwork action ID: Records the ID information of the action executed by the subwork.
-Subwork input data: Enter any value you want to use as input data when issuing the subwork.

Subwork ID result information: Records the execution result information obtained when the corresponding subwork is completed.
Subwork ID result valid bit: Displays whether or not the subwork ID result information is valid after execution of the subwork is completed.

  In the work table at the lowest level, the items "subwork ID issue completion bit", "subwork ID", "subwork action ID", "subwork input data", "subwork ID result valid bit", "subwork ID result" There is no “information”.

  Next, packet element table generation processing and work table generation processing performed when reception control 111 receives data packet 201 will be described with reference to FIGS. 6 and 7. FIG.

  First, generation processing of the packet element table 210 will be described with reference to FIG. In the generation process of the packet element table 210, the following settings are made for each item.

Element ID: Sets the element ID assigned when receiving the data packet 201.
• Element ID valid bit: Set “1” to indicate that the element ID is valid.

Work ID: Set the work ID assigned when receiving the data packet 201.
• Work ID valid bit: Set “1” to indicate that the work ID is valid.

-Work processing result: Set that it is unprocessed.
• Work completion bit: Set “0” to indicate that the work is not completed.

-Output queue ID: Set nothing. The output queue ID is set by the thread to which the sub work is assigned.
Output queue ID valid bit: Set “0” to indicate that the output queue ID is invalid.

• Packet queue type display field: Nothing is set. This field is set by the transmission control unit 112.

-Packet queue preceding element ID: Nothing is set. This item is set by the transmission control unit 112.
Packet queue preceding element ID valid bit: “0” indicating invalid is set.

-Packet queue subsequent element ID: Nothing is set. This item is set by the transmission control unit 112.
• Packet queue subsequent element ID valid bit: “0” indicating invalidity is set.

Data packet: The received data packet 201 is set.

  Next, processing for generating the work table 210 will be described with reference to FIG. In the process of generating the work table 210, the following settings are made for each item.

Work ID: Set the work ID assigned when receiving the data packet 201.
• Work ID valid bit: Set “1” to indicate that the work ID is valid.

Element ID: Sets the element ID assigned when receiving the data packet 201.
• Element ID valid bit: Set “1” to indicate that the element ID is valid.

-Action ID: Set the action ID that defines the action at the highest sub-work.
• Action ID valid bit: Set “1” to indicate that the action ID is valid.

-Assigned thread ID: Nothing is set. This item is set by the work control unit 120.
Assigned thread completion bit: “0” indicating that assignment to a thread has not been completed is set.

-Parent work ID: Since it is the sub-work of the highest hierarchy and there is no parent work, nothing is set.
• Parent work ID valid bit: “0” indicating that the parent work ID is invalid is set.

Work queue ID: Since this item is set by the work control unit 120, nothing is set.
Work queue assignment completion bit: “0” indicating that assignment to the work queue group is not completed is set.

Work queue preceding work ID: Since this item is set by the work control unit 120, nothing is set.
Work queue preceding work ID valid bit: “0” indicating that the work queue preceding work ID is invalid is set.

Work queue subsequent work ID: Since this item is set by the work control unit 120, nothing is set.
Work queue subsequent work ID valid bit: “0” indicating that the work queue subsequent work ID is invalid is set.

・ Processing progress information: Arbitrary (It is also possible to set specific fields in the data packet 201)

Subwork ID: Number and set a subwork ID for each of the subwork predetermined for the subwork.
• Subwork ID issuance completion bit: Set "0" to indicate that no work has been issued.

-Subwork input data: When issuing the subwork, record any values you want to use as input data.
-Subwork action ID: Set the action ID of the action to be executed in the subwork.

Subwork ID result information: This item is set by the thread, so nothing is set.
Subwork ID result valid bit: “0” indicating that the subwork ID result information is invalid is set.

  When the reception control unit 111 generates the packet element table 210 and the work table 211 on the main memory 140, the reception control unit 111 sends the work ID of the sub-work of the highest hierarchy to the work control unit 120 and assigns the thread to the corresponding sub-work. And the element ID of the data packet 201 is sent to the transmission control unit 112. In the following, the sub-work at the highest level is referred to as sub-work W1. Further, it is assumed that the work ID of the sub work W1 is ID “IDW1” and the element ID of the data packet 201 is “ID201”.

  The work control unit 120 stores the work ID “IDW1” of the top-level sub-work W1 sent from the reception control unit 111 in the output destination work queue 502, and the transmission control unit 112 receives from the reception control unit 111. The transmitted element ID “ID201” is stored in the output destination distribution packet queue 402.

  Thereafter, the work scheduler 231 in the work control unit 120 assigns the thread 131 to the sub-work W1 in the highest hierarchy. When the work scheduler 231 assigns the thread 131 to the uppermost sub work W1, the work scheduler 231 sets the ID of the thread as the assigned thread ID in the corresponding work table 211 and sets the assigned thread completion bit to “1”.

  When the work ID “IDW1” of the sub-work W1 in the highest hierarchy is assigned, the thread 131 starts the process illustrated in the flowchart of FIG.

  First, the thread 131 performs exclusive control on the work table 211 of the corresponding work ID “IDW1” (step 1). Here, the procedure for setting and releasing exclusive control is defined below.

[Exclusive control setting procedure]
-Check whether other threads are performing exclusive control on the work table 211 with the corresponding work ID "IDW1".
-If other threads are performing exclusive control, continue to wait until exclusive control is released. When other threads are performing exclusive control, the following operations are performed.
-> The corresponding thread cannot execute the sub-work of the corresponding work ID-> The corresponding thread cannot access the work table 211 of the corresponding work ID-If the exclusive control is not performed by another thread, the work table of the corresponding work ID "IDW1" Exclusive control is set for 211 so that other threads cannot execute the work with the corresponding work ID and cannot access the work table 211 with the corresponding work ID “IDW1”.

[Procedure for releasing exclusive control]
・ Release exclusive control for the work table 211 with the corresponding work ID “IDW1” so that other threads can execute the work with the corresponding work ID “IDW1” and access the work table 211 with the corresponding work ID “IDW1”. It can be so.

  Thereafter, the thread 131 refers to the action ID set in the corresponding work table 211 and executes an instruction (instruction sequence) described in the corresponding action table (step 2). The input data delivered when executing the instruction and the output data expected after execution are as follows. After executing the instruction, the action ID valid bit is set to “0”. It should be noted that the instruction sequence can have different processing contents when the sub-work is issued for the first time and when it is re-issued.

[Input data] Work ID, action ID, process progress information, result information of each sub work ID [Output data] Process progress information (update information), sub work ID, sub work action ID, sub work input data, output Queue ID

  Now, for example, assuming that the output queue for storing the data packet 201 is determined in step 2, the thread 131 uses the output queue ID (for example, # 1) of the output queue as a packet element table for the data packet 201. In addition, the output queue ID valid bit is set to “1”. As a result, the transmission control unit 112 sets the element ID for specifying the data packet 201 stored in the output destination distribution packet queue 402 to the output queue ID # of the queues constituting the processing-waiting packet output queue group 401. Store in queue 1 When updating the packet element table 210, such as setting an output queue ID in the packet element table 210, the exclusive control is performed on the packet element table 210, and then the exclusive control is released. To do.

  Thereafter, the thread 131 determines whether there is an unissued sub-work based on the sub-work ID issuance completion bit of the work table 211 for the sub-work W1 (step 3). If all are “1”, it is determined that there is no unissued subwork, and the process proceeds to step 4; otherwise, it is determined that there is an unissued subwork, and the process proceeds to step 13. When the item “subwork ID issuance completion bit” does not exist in the work table, it is determined that there is no unissued subwork. The sub-work W1 to be processed now is the top-level sub-work and has not yet performed the sub-work issue process, so the thread 131 proceeds to step 13.

  In step 13, the thread 131 performs issuing processing for unissued subwork (second-level subwork). Specifically, referring to the work table 211 of the highest sub work W1, a new work table is generated on the main memory 140 for each sub work ID for which the sub work ID issuance completion bit is “0”. To do. Thereafter, the corresponding sub-work ID (sub-work ID of the second-level sub-work) is sent to the work control unit 120. The method for generating the work table is the same as the method for generating the work table in the highest hierarchy described above except for the following points. In this example, it is assumed that m subworkes have been issued, and each subwork is denoted as subwork W21 to W2m. In addition, if the data packet to which the subwork belongs is discarded, the subwork is not issued (in this case, it can be identified by the fact that the element ID of the packet element table to which the work belongs belongs to “0”). is there).

Action ID: Sets the action ID in the work table 211 of the highest sub work W1.
Parent work ID: Set the work ID of the sub work W1 that has issued the corresponding sub work W2i (1 ≦ i ≦ m).
• Parent work ID valid bit: Set “1”.
-Process progress information: Sub-work ID input data is used.

  Thereafter, the thread 131 releases the exclusive control of the sub work W1 with respect to the work table 211, and ends the process.

  On the other hand, the work control unit 120 to which the sub-work ID (IDW2i) of the sub-work W2i in the second hierarchy is passed is the output queue ID of the packet element table 210 of the data packet 201 that is the generation source of this sub-work W2i (# 1 is set), the sub work ID “IDW2i” is stored in the queue whose output queue ID is # 1 among the work output queues constituting the work output queue group 501. Thereafter, the work scheduler 231 in the work control unit 120 may be the same thread as the thread 131 (the thread assigned the top-level sub-work ID to the second-level sub-work ID “IDW2i” or different. May be a thread).

  The thread 131 to which the sub-work ID “IDW2i” of the second hierarchy is assigned determines whether or not there is an unissued sub-work after performing the processing of Steps 1 and 2 described above (Step 3). If the second layer is the lowest layer, the item “subwork ID issuance completion bit” does not exist. Therefore, the thread 131 determines that there is no unissued subwork and proceeds to step 4.

  In step 4, it is determined whether or not there is an incomplete subwork. Whether or not there is an incomplete subwork is determined based on the subwork ID result valid bit set in the work table of the subwork W2i currently being processed. That is, if all the subwork ID result valid bits are “1”, it is determined that there is no incomplete subwork, and if not, it is determined that there is an incomplete subwork. Also, if there is no item “subwork ID result valid bit” in the work table, it is determined that there is no incomplete subwork. The sub-work W2i that is currently being processed is the sub-work of the lowest hierarchy, and since the item “sub-work ID result valid bit” does not exist in the work table, the thread 131 has no incomplete sub-work. Judge and go to step 5.

  In step 5, it is determined whether or not there is a parent sub-work that has issued the sub-work W2i currently being processed. Specifically, if the parent work ID valid bit registered in the work table of the sub work W2i is “1”, it is determined that there is a parent sub work, and if it is “0”, it becomes the parent. Judge that there is no subwork. If the parent sub-work exists, the process proceeds to step 6; otherwise, the process proceeds to step 14. In this example, the sub-work W2i to be processed is the second-level sub-work, and the parent sub-work W1 exists, so the process proceeds to step 6.

  In step 6, exclusive control is performed on the work table of the sub work W1 which is the parent of the sub work W2i. In the next step 7, the sub work ID result information set in the sub work W1 is updated with the work ID “IDW2i” that is currently being executed, and the sub work ID result valid bit is set to “1”. Change to "".

  In the next step 8, in subwork W1 that is the parent of the subwork W2i to be processed, whether or not all subwork W21 to W2m are completed, subwork ID result valid bit of subwork W1 Is determined based on whether or not all are “1”. If all of them are “1”, it is determined that all the sub-work W21 to W2m of the sub-work W1 are completed, and the process proceeds to step 9; otherwise, the process proceeds to step 10.

  In step 9, the work ID “IDW1” of the sub work W1 that is the parent of the sub work W2i is transmitted to the work control unit 120, and the assignment of the thread 131 to the sub work W1 is requested. In the next step 10, exclusive control for the work table of the sub work W1 is released. In step 11, it is determined that the processing of the sub work W2i and lower sub work is completed, and the work ID valid bit of the work table of the sub work W2i is set to “0”. Finally, the exclusive control for the work table of the sub work W2i that is currently being processed is released (step 12).

  The work ID “IDW1” of the sub-work W1 of the highest hierarchy transmitted to the work control unit 120 in step 9 is stored in the queue having the output queue ID “# 1” in the work output queue group 501, and the work scheduler 231 The thread 131 is assigned by.

  The thread 131 to which the sub-work W1 of the highest hierarchy is assigned performs the processing of Steps 1 and 2 described above, and then checks whether there is an unissued sub-work (Step 3). In this example, since the sub work W1 has already issued all the sub work W21 to W2m and the sub work ID issuance completion bit of the corresponding work table is “1”, the determination result of step 3 is N (No), and the process of step 4 is performed.

  In step 4, it is checked whether or not there is an incomplete subwork in subwork W1. In this example, since there is no incomplete subwork, the determination result in step 4 is N, and the process in step 5 is performed.

  In step 5, it is checked whether or not a sub work as a parent exists in the sub work W1. In this example, since the sub work W1 is the sub work of the highest hierarchy, the determination result is N, and the process of step 14 is performed.

  In step 14, the processing progress information of the work table of the sub work W1 and the sub work ID result information of the work tables of the sub work W21 to W2m under the sub work W1 are used to determine the data packet 201 that is the work generation source. Update the packet element table 210 (for example, rewrite the address of the data packet). In addition, when updating the packet element table 210, after performing exclusive control, an update process is performed.

  According to the present embodiment, the allocation of resources such as processor processing capacity and memory, which has been directed to processing data packets that are likely to be discarded, can be used for processing other data packets that are unlikely to be discarded. Therefore, it is possible to obtain an effect that it is possible to effectively use resources included in the communication device. The reason is that when there is a congested output queue among the queues constituting the output queue group 403, the thread of the subwork related to the packet storing the congested output queue is stored in the thread. This is because the allocation is limited.

  In this embodiment, since data packets belonging to the same flow are stored in the same output queue and all sub-work related to the data packet is completed, it is sent out from the output queue. A plurality of sub-work can be executed in parallel by an arbitrary plurality of threads 131, and the processing efficiency is improved.

  Furthermore, in this embodiment, since sub-work can be assigned to any thread 131, even if the number of cores or the number of threads increases due to device upgrade, the processing algorithm can be easily improved without correcting the processing algorithm. Can be achieved.

[Second embodiment of the present invention]
Next, a second embodiment of the communication apparatus according to the present invention will be described. The present embodiment is characterized in that a priority is assigned to a sub work and a thread is preferentially assigned to a sub work having a high priority. Hereinafter, differences from the first embodiment will be described.

  In the present embodiment, the work table shown in FIG. 10 is used in which the item “work priority” is added to the work table used in the first embodiment. In this embodiment, it is assumed that the work priorities are k levels from the first level to the k-th level. Assume that the first level has the highest priority.

  In the present embodiment, the work control unit 120 having the configuration shown in FIG. 9 is used. The work control unit 120 according to the present embodiment includes a work output queue group 501a instead of the work output queue group 501, and a round robin function unit 512a instead of the round robin function unit 512. However, it is different from the work output queue group 501 of the first embodiment shown in FIG.

  The work output queue group 501 of the first embodiment is composed of N queues (the same number as the number of queues constituting the output queue group 403), whereas the work output queue group 501a of the present embodiment. Is composed of N × k queues. That is, the work output queue group 501a of the present embodiment is composed of queues corresponding to the number of stages of work priority (k pieces) for one output queue. In other words, k queues for work priorities “first level to k-th level” are provided in association with one output queue ID. When the work ID is passed from the core 130, and the output queue ID of the queue to be stored is already determined, the work control unit 120 includes the k queues identified by the output queue ID. , Store in a queue according to the work priority of the work ID. In FIG. 9, the Y-th level queue corresponding to the output queue ID “X” is denoted by the symbol LX-Y.

  When the work queue 231 specifies the output queue ID of the queue that stops or suppresses the output of the work ID (decreases the selected ratio from other queues), the round robin function unit 512a uses the output queue ID described above. Stops or suppresses output of work IDs from the corresponding k queues. Further, the round robin function unit 512a preferentially outputs work IDs stored in a queue having a high work priority. For example, the work ID is output in all k scans from the first level queue with the highest work priority, and (k-1) of the k scans are output from the second level queue. The work ID is output, and the work ID is output in one of the k scans from the k-th level queue having the lowest work priority.

  The setting of the work priority in the work table is performed in the work control unit 120 and in step 13 of the flowchart shown in FIG.

  The work control unit 120 sets a predetermined work priority when generating a work table corresponding to the sub-work of the highest hierarchy.

  In step 13, for example, the thread 131 gives a high work priority to the following sub work.

-Subwork for calculating the output queue ID of the data packet: By preferentially processing the operation for obtaining the output queue ID of the data packet, the resources of the multi-core processor unit 150 can be effectively distributed to the subwork. It is considered to be. For this reason, the work priority of the sub work for calculating the output queue ID of the data packet is increased.
• Subwork of data packets that are congested in the queued packet output queue (queues that constitute the queued packet output queue group 401). Since the processing is considered to be delayed, raise the priority of the corresponding subwork. Note that whether or not the processing-waiting packet output queue is congested is determined based on, for example, whether or not the queue usage rate exceeds a threshold value.
・ Sub work corresponding to data packet with high QoS class ・ Other work with large processing unit, work with large number of sub-work under it, work with large amount of memory, etc.

  According to the present embodiment, in addition to the effect obtained in the first embodiment, it is possible to obtain an effect that the processing can be performed more efficiently. The reason is that threads are preferentially assigned to work having a high work priority.

  The present invention can be used in a communication apparatus that performs packet transfer processing.

101 ... Receiving interface section
102: Transmission interface
103a, 103b ... Switch part
111 ... Reception control section
112 ... Transmission control section
120… Work control section
130 ... Core
131 ... Thread
140 ... Main memory
150 ... multi-core processor
160 ... Bus interface
201 ... Data packet
210 ... Packet element table
211… Work table
220 ... Packet queue group
221 ... Work queue group
230 ... Packet transmission scheduler
231 ... Work scheduler
401: Processing packet output queue group
402: Output destination packet queue
403 ... Output queue group
501,501a ... Work output queue group
502: Output destination work queue
511,512,512a… Round robin function

Claims (13)

Processor cores,
Multiple output queues where processed packets are stored;
A work control unit that divides the processing for the received packet into a plurality of sub-work, and assigns each sub-work to a thread executed on the processor core;
When there is a congested output queue among the plurality of output queues, the work control unit assigns a thread to a subwork related to a packet having the congested output queue as a store destination. A communication device characterized by restricting. The communication device according to claim 1,
The communication device, wherein the congested output queue is an output queue whose usage rate exceeds a first threshold value. The communication device according to claim 1 or 2,
The communication apparatus according to claim 1, wherein the limitation on the assignment of threads to the subwork is to prohibit the assignment of threads to the subwork. The communication device according to claim 1 or 2,
The communication device is characterized in that the restriction on the assignment of threads to the subwork is to reduce the frequency of assignment of threads to the subwork compared to other threads. The communication apparatus according to any one of claims 1 to 4,
The communication apparatus is characterized in that the packet stored in the output queue is sent out after all sub-work related to the packet is completed. The communication apparatus according to any one of claims 1 to 5,
The work control unit, when assigning the plurality of sub-work to a thread executed on the processor core, preferentially assigns a thread from a sub-work having a high priority. The communication device according to claim 6.
A communication apparatus characterized in that the priority of a subwork for determining an output queue to be a packet storage destination is higher than that of other subwork. The communication device according to claim 6.
A process waiting packet output queue for each output queue, which stores a process unfinished packet and outputs a process completed packet to an output queue corresponding to its own process waiting packet output queue. Prepared,
A communication apparatus characterized in that a priority of a subwork related to a processing-waiting packet output queue whose usage rate is equal to or higher than a second threshold is higher than that of other subwork. The communication device according to any one of claims 1 to 8,
2. A communication apparatus comprising a plurality of the processor cores. A first step in which the work control unit divides the processing for the received packet into a plurality of sub-work;
When the work control unit assigns a sub work to any one of a plurality of threads executed on the processor core, an output queue that is congested in a plurality of output queues in which processed packets are stored A sub-work related to a packet that stores the congested output queue as a store destination, the second step of restricting thread allocation. The communication method according to claim 10,
The communication method, wherein the congested output queue is an output queue whose usage rate exceeds a first threshold value. The communication method according to claim 10 or 11,
The communication method according to claim 1, wherein the limitation on the assignment of threads to the subwork is to prohibit the assignment of threads to the subwork. The communication method according to claim 10 or 11,
The communication method is characterized in that the limitation of the thread allocation to the subwork is to reduce the frequency of thread allocation to the subwork as compared to other threads.

Images