JP2008028767A - Network card and information processing apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique for efficiently transmitting streaming data.
A network card including a host connector and a network connector, wherein data to be transmitted via the network connector is a unit of block data having a second size larger than the first size via the host connector. Receiving means, a buffer memory for temporarily storing the received block data, and generating a data frame of the first size or less, and transmitting the data frame on a network connected to the network connector Transmitting means.
[Selection] Figure 4

Description

  The present invention relates to a data transmission technique, and more particularly to a technique for efficiently transmitting broadband streaming data.

  In recent years, streaming services such as video on demand (VOD) have been realized. In addition, the bandwidth per video streaming data transmitted from the distribution server is increasing due to the recent widening of the access network. Furthermore, further broadbandization is expected in the future through the distribution of so-called HD (high definition) video. As the data transmission rate from the distribution server increases, the utilization factor of the internal bus (for example, PCI) of the distribution server tends to increase.

  In general, the maximum data length (MTU) that can be transferred is defined for each network interface according to each standard. For example, in Ethernet (registered trademark), which is also known as the IEEE 802.3 series standard, the maximum is about 1.5 kbytes. Therefore, the device driver generally divides the data to be transmitted on the network into data that does not exceed the above MTU even if the data length that can be transmitted by the internal bus is sufficiently large. Transfer to the interface board via the internal bus. This division is called fragment processing. Thereafter, the processing is performed by the MAC / PHY on the board and sent out on the network.

On the other hand, in a streaming service, a large amount of data can cause packet discard on the network. In this case, forward error correction (FEC) encoding is used in the distribution server so that it is not necessary to issue a retransmission request from the receiving side. For example, encoding as disclosed in Patent Document 1 is used.
International Publication No. WO2005 / 112250

  In general, the internal bus can transfer data in a sufficiently large data unit (data length) in accordance with the transmission unit of streaming data. However, as described above, the data length flowing on the internal bus becomes small as a result of the limitation by the MTU. Therefore, in transmitting streaming data, there is a problem that bus congestion tends to occur, and CPU power by fragment processing is also required.

  Further, in order to execute an encoding process as disclosed in Patent Document 1 for a large amount of data, further CPU power is required.

  The present invention has been made in view of the above-described problems, and an object thereof is to solve at least one of these problems.

  In order to solve at least one of the problems described above, the network card of the present invention has the following configuration.

  That is, a network card having a host connector for connecting to a bus connector provided in the host device and a network connector for connecting to a network, the maximum size of a data frame that can be transmitted through the network connector Receiving means for receiving data to be transmitted via the network connector as a unit of block data of a second size larger than the first size via the host connector, A buffer memory for temporarily storing the block data received by the receiving means, and reading data to be included in the data frame to be transmitted from the buffer memory, and generating a data frame having the first size or less Connected to the network connector It was on the network and transmitting means for transmitting the data frame.

  In order to solve at least one of the problems described above, an information processing apparatus of the present invention has the following configuration.

  That is, an information processing apparatus that includes a host processing unit and a network processing unit connected by a bus and sends streaming data to a network, the host processing unit including data input means for inputting stream data, and at least the stream When the maximum size of a data frame that can be transferred in the network is the first size, the data is transferred to the network processing unit via the bus in units of block data having a second size larger than the first size. The network processing unit for receiving the block data transmitted from the bus transfer unit via the bus, and for temporarily storing the block data received by the receiving unit. Storage means and a data flow to be transmitted from the storage means. Reading data for inclusion in over beam, and generates a data frame to be less than the first size, and transmitting means for transmitting the data frame over a network connected to the network connector.

  According to the present invention, a technique for efficiently sending streaming data can be provided.

  Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. Note that these embodiments are merely examples, and are not intended to limit the scope of the present invention.

(First embodiment)
As a first embodiment of the data transmission apparatus according to the present invention, a streaming distribution apparatus constituted by a general-purpose PC and a network board will be described below as an example.

<Overview>
In the streaming distribution apparatus according to the first embodiment, the fragment processing that has been performed by the CPU of the PC main body by executing the device driver program is processed by hardware on the network board. As a result, the load on the CPU of the PC main body can be reduced, and the data transfer to the network board via the bus can be performed with a larger data length, thereby improving the bus use efficiency.

<System configuration and device configuration>
FIG. 1 is a conceptual diagram of the overall configuration of a streaming distribution system.

  Reference numeral 100 denotes a streaming distribution server, and 110a and 110b denote streaming receiving apparatuses. Reference numerals 101, 111a, and 111b denote network segments to which the distribution server 100 and the receiving apparatuses 110a and 110b belong, respectively. Each network segment 101, 111 a, 111 b is connected via routers 102, 112 a, 112 b and core network 120. In the following description, it is assumed that the Internet protocol (IP) is used for data transfer between the network segments.

  The distribution server 100 packetizes and transmits the stream data in the RTP / UDP / IP format to the receiving terminals 110a and 110b. Here, RTP means real-time transport protocol and UDP means user datagram protocol. The distribution server 100 may perform streaming distribution to each receiving terminal by unicast or distribution by multicast. Further, distribution may be started based on a distribution request from each receiving terminal as in a so-called video on demand (VOD) service.

  FIG. 2 is a diagram illustrating an internal configuration of the distribution server according to the first embodiment. As shown in the figure, the distribution server 100 includes a CPU 201, a RAM 202, a ROM 203, an HDD 204, a user I / F 205, and a network (NW) board 200, and each unit is mutually connected by an internal system bus 210. .

  The CPU 201 executes various programs stored in the ROM 203 and the HDD 204 to control each unit or implement each functional unit described later with reference to FIG. The ROM 203 stores a program executed when the distribution server 100 is activated. The RAM 202 temporarily stores various programs executed by the CPU 201 and various data. The HDD 204 is a large-capacity storage device and stores various programs and various data files. The program includes an operating system (OS) program and a streaming distribution program. A user I / F 205 is a user input device such as a keyboard and a mouse (not shown) and a display output device such as a display (not shown).

  The internal system bus 210 is assumed to be a general-purpose bus including a general PCI bus, but may be a dedicated bus as a matter of course. However, the transfer speed on the bus 210 is higher than the transfer speed on the network 101, and the transferable data length is longer than the data length on the network 101.

  In the following, in order to simplify the description, in the distribution server 100, the network (NW) board 200 may be referred to as “NW board side” and the other parts may be referred to as “server body side”.

  FIG. 3 is a diagram illustrating an internal configuration of the network board in the distribution server according to the first embodiment. As shown in the figure, the network board 200 includes a packet handler 301, a memory 302, a memory controller 303, and a bus I / F 310.

  The memory 302 is a part for temporarily storing data received from the server main body via the bus 210 and the bus I / F 310, and includes a packet buffer 302a therein. Although details will be described later, an area of the packet buffer 302a is secured for each stream.

  The packet handler 301 is a circuit unit that transmits data temporarily stored in the memory 302 in a data format suitable for the network 110. Specifically, the data temporarily stored in the memory 302 is subjected to fragment processing and smoothing processing described later, and then output to the network 110.

<Functional configuration and operation>
FIG. 4 is a diagram illustrating a functional configuration of the distribution server according to the first embodiment.

  The distribution server 100 includes an input unit 401, a bus transfer unit 402, a fragment processing unit 403, and a smoothing processing unit 404 as functional units related to data transmission. The function units of the input unit 401 and the bus transfer unit 402 are realized by the CPU 201 on the server body side executing various programs. On the other hand, the functional units of the fragment processing unit 403 and the smoothing processing unit 404 are realized by hardware on the NW board side. Hereinafter, each functional unit will be described.

  In the following, for the sake of simplicity, only the processing for the streaming data in the RTP / UDP / IP format of each processing unit will be described. Other data is processed as usual. Packets may be distinguished based on the port number described in the IP header, or simply based on the packet data length.

  The input unit 401 is a functional unit that inputs a streaming file to be transmitted via the network board 200. Specifically, it is realized by the CPU 201 executing the streaming distribution software to read the streaming data stored in the HDD 204 or the like into the RAM 202. The input unit 401 corresponds to input means in the claims.

  The bus transfer unit 402 divides the streaming data input to the RAM 202 by the input unit 401 into fixed-length data designated in advance, and stores the data in the RTP / UDP / IP format. It is a functional part to transfer to. Specifically, it is realized by the CPU 201 executing the IP stack program and the device driver program of the NW board 200. The bus transfer unit 402 corresponds to the bus transfer means in the claims.

  However, unlike the case described in the background art, the streaming data packet to be transferred is larger than the data length that can be transmitted to the network 101. For example, when the network 101 is Ethernet (registered trademark), that is, even if the maximum data length (MTU) (first size in the claims) is about 1.5 kbytes, a large data block such as 32 kbytes is used. (Second size in the claims).

  In general, specifications are determined for the data format between the application and the IP stack program and between the IP stack program and the device driver, so that a change requires a large design change. However, it should be noted that the data format between the device driver and the hardware as described above can be designed relatively freely.

  Although details will be described later, in the data (data block), the payload part excluding the headers of IP, UDP, and RTP is represented by a data length corresponding to an integral multiple (or a power of 2) of the minimum processing unit of stream data. It is preferable to do.

  The fragment processing unit 403 is a functional unit that divides data (data blocks) transferred from the bus transfer unit 402 via the bus 210 (bus connector) into data lengths that can be sent to the network 101. Specifically, a data block stored in the memory 302 via a host connector (not shown) is divided into data lengths equal to or less than the MTU of the network 101, and each of IP, UDP, RTP corresponding to the divided data is divided. Regenerate the header. Then, an IP packet having a data length that can be directly transmitted to the network 101 is stored in the packet buffer 302a. The bus I / F 310 corresponds to the receiving unit in the claims, and the memory 302 corresponds to the buffer memory or the storage unit in the claims. The fragment processing unit 402 constitutes part of the transmission means in the claims, and the packet handler 301 corresponds to the transmission means.

  Note that the regeneration of IP, UDP, and RTP headers specifically refers to the following processing. The IP header describes the data length (payload length) of the data included in the IP packet. The UDP header describes the data length of the data included in the UDP packet and the checksum of the data. Further, the RTP header describes the sequence number and time stamp of the data included in the RTP packet. Therefore, since the RTP packet is divided as a result of fragment processing by the fragment processing unit 403, the information described in each header is calculated according to the divided packet and the header information is updated. . In the data (data block) transferred from the bus transfer unit 402, header information is easily calculated by equally dividing the payload portion excluding the IP, UDP, and RTP headers. Therefore, as described above, it is desirable that the payload portion of the data block has a data length corresponding to an integral multiple (or a power of 2) of the minimum processing unit of the stream data.

  The smoothing processing unit 404 is a functional unit that sends out fixed-length IP packets stored in the packet buffer 302 a by the fragment processing unit 403 to the network 101 at equal intervals. Specifically, a transmission interval is calculated based on header information in a fixed-length IP packet stored in the packet buffer 302a, and the IP packets are transmitted in the stored order. The transmission interval can be calculated, for example, from the data length information of the IP header or UDP header and the time stamp information of the RTP header. Note that the IP data may be sent in order so as to have a preset sending interval that is assumed not to have bursty traffic characteristics. The smoothing processing unit 404 corresponds to transmission interval control means in the claims.

  In the above description, transmission control is performed so that IP (RTP) packets after fragmentation are equally spaced in the time direction. However, in general, not only RTP packets but also RTCP packets used for controlling the RTP packet stream or other packets can be transmitted from the NW board 200. For this reason, it is also preferable to secure a usable time slot in addition to the RTP packet in advance, and perform transmission control so that the intervals are equal in a period excluding the time slot.

<Operation flow>
FIG. 5 is a data processing flowchart in the distribution server according to the first embodiment. The following steps are started by receiving a streaming data transmission request from the receiving device 110a (or 110b), for example. Here, it is assumed that the minimum processing unit of stream data is 64 bytes.

  In step S501, the input unit 401 reads the streaming data requested from the receiving device 110a from the HDD 204 or the like and stores it in the RAM 202.

In step S502, the bus transfer unit 402 divides the data stored in the RAM 202 in step S501 into data blocks having a data length of 16 kbytes (= 64 bytes × 2 ⁸ ), for example. Then, IP, UDP, and RTP headers for the data block are generated, stored in the RTP / UDP / IP format, and transferred to the NW board side via the bus 210.

In step S503, the fragment processing unit 403 converts the payload data in the data block transferred from the bus transfer unit 402 via the bus 210 in step S502 to a data length of, for example, 512 bytes (= 64 bytes × 2 ³ ). Divide into data. That is, the data block is divided so that the data length is equal to or less than the MTU in the network 101. Then, the IP, UDP, and RTP headers are regenerated for each 512-byte data generated by division and stored in the RTP / UDP / IP format. Then, the regenerated IP packet is stored in the packet buffer 302a.

  In step S504, the smoothing processing unit 404 sends the IP packets stored in the packet buffer 302a in step S503 to the network 101 at regular intervals.

  In the above flowchart, for the sake of simplicity, the description has been made so that single streaming data is transmitted from the distribution server 100. However, of course, it is possible to execute the above processing for each streaming data. In particular, the streaming data reaching the network segments 111a and 111b to which the receiving devices 110a and 110b belong by performing the above-described processing of the smoothing processing unit 404 for each streaming data, Therefore, there is an advantage that data loss hardly occurs.

  As described above, according to the distribution server of the first embodiment, the load on the bus 210 due to fragment processing (congestion) and the load on the CPU 201 due to execution of fragment processing can be greatly reduced. Therefore, the bottleneck caused by the transfer capability of the bus 210 or the processing capability of the CPU 201 can be greatly relieved. As a result, streaming data can be transmitted more efficiently.

(Second Embodiment)
<Overview>
In the second embodiment, in addition to the configuration of the first embodiment, a forward error correction code (FEC) encoder is arranged on the NW board. The forward error correction code referred to here includes a loss compensation code. With such a configuration, the CPU power consumed for the FEC encoding process can be greatly reduced. In addition, the bus usage rate (traffic) can be reduced.

  Note that the overall configuration of the streaming distribution system (FIG. 1) and the internal configuration of the distribution server (FIG. 2) are the same as those in the first embodiment, and a description thereof will be omitted.

<FEC code>
In the present invention, it is particularly effective to use a loss compensation code as the FEC code. Therefore, in the second embodiment, it is assumed that a Raptor code, which is an FEC code developed by the US Digital Fountain Company, is used as the FEC code. Of course, it is also possible to use a general Reed-Solomon (RS) -based code. Hereinafter, the Raptor code will be briefly described. For details, refer to Patent Document 1 described in the background art.

  In the Raptor code, a stream file is divided into sections having a specific data length (s × k bytes), and data in each section is divided into k pieces of data having the same data length (s bytes) called “input symbols”. Then, based on an index value called a key, one or more input symbols are selected from the divided k input symbols, the selected input symbols are subjected to XOR operation for each bit, and s called “output symbols” Generate data with a data length of bytes. Such output symbols are generated sequentially for different keys.

  On the other hand, on the receiving side, k + α (α is smaller than k) output symbols are received stochastically, and the input symbols are restored by XORing the output symbols. At this time, since k + α output symbols can be arbitrarily selected, it has an excellent characteristic that it can be restored regardless of which packet is lost during transfer.

<Device configuration>
FIG. 6 is a diagram illustrating an internal configuration of the network board in the distribution server according to the second embodiment. As shown in the figure, the network board 600 includes an encoding engine 604 and an encoding control unit 605 in addition to a packet handler 601, a memory 602, a memory controller 603, and a bus I / F 610. Below, the FEC encoding engine 604 and the encoding control part 605 which are different parts from 1st Embodiment are demonstrated.

  The FEC encoding engine 604 is a circuit that executes an XOR operation by hardware. It is well known to those skilled in the art that logical operations including XOR operations can be easily configured by hardware.

  The encoding control unit 605 is a functional unit that realizes the above-described encoding operation of the Raptor code by controlling the FEC encoding engine 604. It is preferable that the encoding control unit 605 is configured as a flash memory that stores a CPU and a control program (not shown), so that it can be easily changed to another FEC encoding algorithm. Note that the encoding control unit 605 and the FEC encoding engine 604 correspond to encoding means in the claims.

  Specifically, the encoding control unit 605 sequentially generates output symbols by selecting one or more input symbols from data (input symbols) temporarily stored in the memory 602 and inputting them to the FEC encoding engine 604. The generated output symbol is temporarily stored in the memory 602.

  However, as described in the description of the above-described Raptor code, it should be noted that although the output symbol and the input symbol have the same data length, the number increases by at least α.

  The packet handler 601 is a circuit unit that transmits data composed of output symbols temporarily stored in the memory 602 in a data format suitable for the network 110. Specifically, the data temporarily stored in the memory 602 is subjected to fragment processing and smoothing processing, and then output to the network 110.

<Functional configuration and operation>
FIG. 7 is a diagram illustrating a functional configuration of the distribution server according to the second embodiment.

  The distribution server 100 includes an encoding processing unit 705 in addition to an input unit 701, a bus transfer unit 702, a fragment processing unit 703, and a smoothing processing unit 704 as functional units related to data transmission. Below, the part relevant to the encoding process part 705 which is a different part from 1st Embodiment is demonstrated.

  The encoding processing unit 705 is a functional unit that performs FEC encoding processing on data (data block) transferred from the bus transfer unit 702 via the bus 210. Specifically, an output symbol is generated by regarding the data block realized by the encoding engine 604 and the encoding control unit 605 and stored in the memory 302 as the above-described input symbol.

  The fragment processing unit 703 is a functional unit that divides the output symbols (data blocks) encoded by the encoding processing unit 705 into data lengths that can be transmitted to the network 101. Specifically, the data block stored in the memory 602 is divided into data lengths equal to or less than the MTU of the network 101, and the IP, UDP, and RTP headers corresponding to the divided data are regenerated. Then, an IP packet having a data length that can be directly transmitted to the network 101 is stored in the packet buffer 602a.

  However, as described above, redundant data is added by the encoding processing by the encoding processing unit 705, and as a result, the output rate is higher than the input rate of the encoding processing unit 705. Specifically, when k + α output symbols are generated from k input symbols, since the data length is the same, the output rate is (k + α) / k times the input rate. At this time, the coding rate is expressed as k / (k + α). For example, the time stamp of the RTP header is reset based on the coding rate. That is, the time interval is set to be shorter by about k / (k + α) times compared with the case of non-coding. Here, the data amount of one input symbol and one output symbol corresponds to the first data amount in the claims. A data amount corresponding to k input symbols corresponds to a second data amount in the claims.

  The smoothing processing unit 704 is a functional unit that sends out fixed-length IP packets stored in the packet buffer 602 a by the fragment processing unit 703 to the network 101 at equal intervals. Specifically, the transmission interval is calculated based on the header information in the fixed-length IP packet stored in the packet buffer 602a, and the IP packets are transmitted in the stored order. As described above, since the time stamp of the RTP header is set short, as a result, the transmission interval is also set short by about k / (k + α) times.

  As described above, according to the distribution server of the second embodiment, the load on the CPU 201 is greatly reduced by executing the FEC encoding processing on the NW board 600 in addition to the fragment processing described in the first embodiment. I can do it. Further, since redundant data by FEC encoding does not flow on the bus 210, it is possible to reduce the bus usage rate (traffic). As a result, streaming data can be transmitted more efficiently.

(Modification)
In the above description, a fixed-length packet (512 bytes) smaller than the maximum transfer size (about 1.5 kbytes) of the network directly connected to the network board (here, Ethernet (registered trademark)) is set as the transfer size. . However, of course, it may be set almost equal to the MTU. In general, a plurality of various networks can coexist from the distribution server to the receiving terminal, and there is a possibility that each MTU is different. Further, since the route is different for each receiving terminal, there is a possibility that the “path MTU” is different for each receiving terminal. For this reason, path MTU discovery may be used in advance to detect a path MTU, and the packet size may be dynamically set at the start of stream distribution so as to be equal to or less than the path MTU corresponding to each terminal.

It is a figure which shows the whole structure of a streaming delivery system exemplarily. It is a figure which shows the internal structure of the delivery server which concerns on 1st Embodiment. It is a figure which shows the internal structure of the network board in the delivery server which concerns on 1st Embodiment. It is a figure which shows the function structure relevant to the data transmission of the delivery server which concerns on 1st Embodiment. It is a data processing flowchart in the delivery server which concerns on 1st Embodiment. It is a figure which shows the internal structure of the network board in the delivery server which concerns on 2nd Embodiment. It is a figure which shows the function structure relevant to the data transmission of the delivery server which concerns on 2nd Embodiment.

Claims (8)

A network card comprising a host connector for connecting to a bus connector provided in a host device, and a network connector for connecting to a network,
When the maximum size of the data frame that can be transmitted via the network connector is the first size,
Receiving means for receiving data to be transmitted via the network connector in units of block data having a second size larger than the first size via the host connector;
A buffer memory for temporarily storing block data received by the receiving means;
Transmitting means for reading data to be included in a data frame to be transmitted from the buffer memory, generating a data frame having the first size or less, and transmitting the data frame on a network connected to the network connector; ,
A network card comprising:   The network card according to claim 1, wherein the transmission unit further includes a transmission interval control unit that transmits the one or more data frames to the network at substantially equal intervals in a time axis direction.   The network card according to claim 2, further comprising an encoding unit configured to perform a forward error correction code (FEC) encoding process on data included in the block data. The transmitting unit generates a data frame having a size equal to or smaller than the second size based on the data encoded by the encoding unit;
4. The network card according to claim 3, wherein the transmission interval control means determines the interval for sending the data frame based on a coding rate used by the coding means. The encoding means performs the encoding process by repeatedly executing a logical operation in units of a predetermined first data amount,
5. The network card according to claim 4, wherein the transmission unit includes data having a size that is an integral multiple of the first data amount in the data frame. The encoding means performs the encoding process in units of a predetermined second data amount,
6. The network card according to claim 4, wherein the receiving unit receives the block data set to a size that is an integral multiple of the second data amount. An information processing apparatus including a host processing unit and a network processing unit connected by a bus, and sending streaming data to a network,
The host processing unit
Data input means for inputting stream data;
When at least the stream data has a maximum size of a data frame that can be transferred in the network as the first size, the network processing is performed via the bus in units of block data having a second size larger than the first size. Bus transfer means for transferring to the
With
The network processing unit
Receiving means for receiving block data transmitted via the bus from the bus transfer means;
Storage means for temporarily storing block data received by the receiving means;
Transmitting means for reading data to be included in a data frame to be transmitted from the storage means, generating a data frame having the first size or less, and transmitting the data frame on a network connected to the network connector; ,
An information processing apparatus comprising: An information processing apparatus including a host processing unit and a network processing unit connected by a bus, and sending streaming data to a network,
The host processing unit
A data input unit for inputting stream data;
When at least the stream data has a maximum size of a data frame that can be transferred in the network as the first size, the network processing is performed via the bus in units of block data having a second size larger than the first size. A bus transfer unit for transferring to the unit,
With
The network processing unit
A receiving unit for receiving block data transmitted from the bus transfer unit;
A buffer memory for temporarily storing block data received by the receiving unit;
A transmission unit that reads data to be included in a data frame to be transmitted from the buffer memory, generates a data frame that is equal to or smaller than the first size, and transmits the data frame on a network connected to the network connector; ,
An information processing apparatus comprising:

Images