JP2008219252A - Network communication system, and application notifying method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a network communication system capable of reducing a load of a CPU because an application side is not required to retrieve a command boundary. <P>SOLUTION: An application processing server 1 has the CPU 3 and a network interface card 2 for receiving a packet transmitted from an opposite host, reconstructing a message stream on the basis of the received packet and writing the reconstructed message stream in a main storing part 5. An application operating on the CPU 3 interprets the message stream written in the main storing part 5 as a command and executes communication processing with the opposite host. When there is a pattern adaptable to a pattern for detection which is preregistered in the reconstructed message stream, the network interface card 2 determines the adaptable pattern as a break of the command and writes the reconstructed message stream in the main storing part 5 in each command. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention is an upper layer of a protocol corresponding to packet retransmission and message reassembly (reconstruction) processing, represented by TCP (Transmission Control Protocol), and includes SOAP (Simple Object Access Protocol) and SMTP (Simple Mail Transfer). The present invention relates to a network communication system that performs communication using a message-based application protocol such as SIP (Session Initiation Protocol). More specifically, the present invention relates to a method of receiving a packet from an opposite host by a network interface card on a network communication system and notifying an application on the network communication system of a message stream reconstructed from the received packet. The present invention relates to an access method for writing and copying a reconstructed message stream to a main memory.

  Conventionally, based on a protocol such as TCP that resends packets that are discarded until they reach the opposite host, and a protocol that reconstructs the message stream based on the data sequence number information of the received packet. Communication systems using application protocols such as SOAP, SIP, and SMTP are provided. Most of the communication systems are server devices including a network interface card, a CPU (Central Processing Unit), and a main memory.

  Each time a network interface card receives a packet, the network interface card writes the received packet into the main memory and notifies an OS (Operating System) operating on the CPU of information related to the written packet. The OS reconstructs the message stream from the received packet sequence number information and stores it in the socket buffer.

  By issuing a read () system call, the application reads data from the socket buffer to the user area of the application and performs protocol processing. According to the POSIX UNIX system call convention, the read () system call consists of the following arguments:

(Return value: Actual read length) = read (file descriptor, pointer to the start address of the acquired data, requested data length)
By issuing a read () system call after establishing a TCP session, the OS stores data up to the length of the data requested by the read () system call in the socket buffer in response to the read () system call. Is read and written to the address indicated by the pointer in the user area of the application, and the length read from the socket buffer is passed to the application as a return value.

  In addition, many network interface cards perform writing to the main memory at the timing of receiving a packet, and protocol processing such as TCP message stream reconstruction processing and TCP response processing operates on the CPU. Leave it to the OS. In this case, since the CPU needs to perform both TCP processing and upper layer processing, sufficient CPU resources for performing upper layer application processing may not be allocated.

  Further, when the network interface card writes the packet to the main memory at the timing of receiving the packet, the memory access delay in the main memory is large, so that it is required to reduce the number of memory accesses.

  In order to solve the problems of CPU load and memory access, in the TOE (TCP Offload Engine) architecture such as TCP Chimney, the received TCP packet is reconstructed in the form of a message stream on the network interface card, and the reconstructed message is reconstructed. The format that writes the stream to the socket buffer is adopted. As a result, the CPU load for TCP processing and the number of accesses to the main memory are reduced to improve performance.

  In TCP Chimney, if any of the following three conditions is satisfied, a message stream is passed to the upper layer and a completion notification is sent (see Non-Patent Document 1).

(1) Receive packet with PSH flag (2) Receive buffer is full (3) Timeout (200msec) occurs However, these transfer conditions do not take into account the command delimiter required by the upper application layer . Therefore, the application needs to search the command delimiter of the application after reading the message stream from the socket buffer by the read () system call. Since character string search has a heavy load in application processing, reduction of the character string search load is required.
"Scalable Networking: Network Protocol Offload-Introducing TCP Chimney", Microsoft Corporation, WinHEC 2004 Version-April 9, 2004 (see page 19, "Receive Completions")

  In the conventional method in which the network interface card reconstructs the message stream from the received packet and notifies the application side, the command boundary required by the application is not considered in the notification unit of the message stream notified to the application In addition, it is necessary to perform a character string search process for searching the command boundary on the application side, which increases the load on the CPU.

  An object of the present invention is to provide a network communication system that solves the above-described problems and does not require an application side to search for command boundaries and can reduce the load on the CPU.

In order to achieve the above object, the network communication system of the present invention provides:
A main memory unit;
A network interface unit that receives a packet sent from the opposite host, reconstructs a message stream based on the received packet, and writes the reconstructed message stream in the main storage unit;
An application execution unit that interprets the message stream written in the main storage unit as a command and executes communication processing with the opposite host;
When there is a pattern that matches a detection pattern registered in advance in the reconstructed message stream, the network interface unit determines that the conformance pattern is a delimiter of the command, and determines the reconstructed message stream as Each command is written to the main storage unit.

  According to the above configuration, the network interface unit writes the reconstructed message stream to the main storage unit for each command required by the application execution unit, and the application execution unit generates a message stream for each command from the main storage unit. read out. As described above, the notification unit of the reconstructed message stream to the application is performed in units of commands required by the application. Therefore, it is not necessary to perform a character string search process for searching for command boundaries on the application side.

  According to the present invention, it is not necessary to search for an application protocol command boundary on the application side, and accordingly, the load on the CPU can be reduced, and protocol processing can be performed at high speed.

  Next, embodiments of the present invention will be described with reference to the drawings.

(First embodiment)
FIG. 1 is a block diagram showing a configuration of an application processing server according to the first embodiment of the present invention. As a communication protocol, TCP (Transmission Control Protocol) is assumed in the transport layer, and in a higher layer, SOAP (Simple Object Access Protocol), SMTP (Simple Mail Transfer Protocol), SIP (Session Initiation Protocol), etc. Assume a text-based application protocol.

  Referring to FIG. 1, an application processing server 1 constitutes a network communication system or a monitoring system, and includes a network interface card 2 that performs packet reception processing and protocol processing from the opposite host through the opposite network 10, and an application. A CPU (central processing unit) 3 on which an operating system (OS) and an OS (Operating System) operate, an access arbitration unit 4 that arbitrates read / write processing with respect to the CPU 3 and the main storage unit 5 of the network interface card 2, and a packet A main storage unit 5 in which data and application data (including programs) are stored.

  The network interface card 2 is a network interface unit, and performs protocol processing as a TOE (TCP Offload Engine) such as session resolution of received TCP data, sequence number consistency check, and response ACK packet transmission (in the case of a communication system). Or reconstructing the message stream from the received TCP packet and storing the data in the main storage unit 5.

  The network interface card 2 stores a packet reception unit 21 that performs normality check of received packets and protocol processing, a message reconstruction unit 22 that performs message stream reconstruction processing from received TCP packets, and a reconstructed message stream The received message holding unit 23 and the pattern detection unit 24 that detects a pattern required by the application from the message stream. The detection pattern is registered in advance, and the registered detection pattern is notified to the pattern detection unit 24 during communication with the opposite host by the application program. The pattern detection unit 24 stores the notified detection pattern. The detection pattern includes a start pattern and an end pattern.

  The main memory unit 5 is a main memory represented by a DDR (Double Data Rate) memory, and more specifically is configured by a DDR DIMM (Dual In-line Memory Module) or the like. The main storage unit 5 includes a network interface command receiving unit 51, a protocol processing data storage unit 52, and an application user data storage unit 53.

  The network interface command receiving unit 51 manages message stream information received from the network interface card 2. The network interface card 2 writes the message stream to the protocol processing data storage unit 51 and then writes information about the message stream that has been transmitted to the network interface command reception unit 51.

  The protocol processing data storage unit 52 corresponds to a socket buffer in terms of OS. The protocol processing data storage unit 52 cannot be directly referenced from the application, and copies data to the application user data storage unit 53 in response to a read () system call issued by the application.

  Next, the operation of the application processing server 1 will be described.

  2 and 3 are flowcharts for explaining the operation of the network interface card 2. For convenience, the flow of the operation is divided into two in FIGS. 2 and 3, but these operations are a series of operations. The operation of the network interface card 2 will be described below with reference to FIGS.

  When a TCP packet is received from the opposite network 10 (step A1 in FIG. 2), the packet receiving unit 21 checks the protocol consistency of the TCP packet (TCP field consistency) (step A2 in FIG. 2), and TCP TCP flow information is identified from the packet information (step A3 in FIG. 2). Further, the packet receiving unit 21 refers to the sequence number and determines whether the arrival order of the TCP packets is In-Order (received in the order transmitted from the opposite host) or Out-of-Order (transmitted from the opposite host). (Step A4 in FIG. 2). When the application processing server 1 is a communication device that terminates TCP data, the network interface card 2 performs an ACK packet transmission process or the like.

  After the protocol processing of the TCP packet, the message reconstruction unit 22 reconstructs a message stream from the TCP packet based on the received sequence information of the TCP packet (step A5 in FIG. 2). Then, the message reconstructing unit 22 writes the reconstructed message stream in the received message holding unit 23 (step A6 in FIG. 2).

  After the message stream is written in the received message holding unit 23, the network interface card 2 checks whether a detection pattern is registered in the pattern detection unit 24 from the application (step A7 in FIG. 3). There are two types of detection patterns: a data start pattern and an end pattern required by the application. The start pattern and end pattern are registered using a socket parameter registration function such as a setsockopt () system call for registering socket options when a TCP socket is established. These patterns can be registered as regular expressions that represent several character strings in one predetermined format.

  The pattern detection unit 24 includes regular expression pattern search logic, and executes a search process for determining whether or not a registered pattern exists for the input message stream. If it is determined that the message stream being searched is compatible with the start pattern, the pattern matching flag which is an internal parameter is turned ON. The initial value of the pattern matching flag is OFF.

  If it is determined in step A7 that the detection pattern is registered, the pattern detection unit 24 checks whether the pattern matching flag is ON (step A8 in FIG. 3). If it is determined in step A7 that the detection pattern has not yet been registered, the process proceeds to step A16 in FIG.

  If it is determined in step A8 that the pattern matching flag is OFF, the pattern detection unit 24 first searches the message stream for the registered start pattern (A9 in FIG. 3). When the start pattern is found in the message stream, the pattern detection unit 24 turns on the pattern matching flag (A11 in FIG. 3). Thereafter, the process proceeds to step A12 in FIG. If the start pattern is not found in the message stream, the process proceeds to step A15 in FIG.

  If it is determined in step A8 that the pattern matching flag is ON, it indicates that a start pattern has been found in the message stream registered in the received message holding unit 23 or in the past message stream.

  If it is determined in step A8 that the pattern matching flag is ON, or if it is determined in step A11 that the pattern matching flag is ON, the pattern detection unit 24 registers the end pattern registered. The message stream is searched for (step A12 in FIG. 3). When the end pattern is found, the pattern detection unit 24 transfers the message stream from the start pattern to the end pattern as application data from the received message holding unit 23 to the protocol processing data storage unit 52, and at the same time the pattern is being matched. The flag is turned OFF (step A14 in FIG. 3).

  Here, it should be noted that when both the start pattern and the end pattern are present in the message stream stored in the received message holding unit 23 and the message stream is transferred as an application message, the order of the streams As a result, the start pattern must be positioned before the end pattern. If the pattern matching flag is ON and only the end pattern exists in the message stream, the area from the beginning of the received message stream to the end pattern stored in the received message holding unit 23 is used for protocol processing. The data is transferred to the data storage unit 52.

  After transferring the message stream, the pattern detection unit 24 writes information (command) about the transferred message stream to the network interface command reception unit 51 on the main storage unit 5 (step A15 in FIG. 3). Thereafter, the process returns to the data reception process.

  When the determination in any of steps A7, A10, and A13 is “No”, the network interface card 2 determines whether or not any of the following three conditions C1 to C3 is satisfied (FIG. 3 A16).

C1: PSH flag of received packet is ON
C2: Receive buffer is in full state C3: Time-out (200 msec) occurs When any of the conditions C1 to C3 is satisfied, the message stream existing in the receive buffer (received message holding unit 23) is stored in the protocol processing data storage unit 52. The information (command) related to the transferred message stream is written in the network interface command receiving unit 51 (step A18 in FIG. 3). Thereafter, the process returns to the data reception process.

  If none of the conditions C1 to C3 is satisfied, the process returns to the data reception process to perform the next packet process.

  Next, the operation of the CPU 3 will be described. FIG. 4 shows an operation flow of the CPU 3. The operation will be described below with reference to FIGS.

  The OS and application operate on the CPU 3 and use the main storage unit 5 as an execution area and data area for the OS and application. The OS performs protocol processing. At this time, the protocol processing data storage unit 52 is used as a socket buffer. The application uses the application user data storage unit 53 as an application user area. Due to the write protection by the OS, the application cannot refer to the protocol processing data storage unit 52 corresponding to the socket buffer. This is a limitation of general-purpose OSs such as Linux and BSD.

  In FIG. 4, steps B1, B2, and B7 indicated by thick frames are processing on the application side, and steps B3 to B6 indicated by thin frames are processing on the OS side.

  First, the application registers a pattern indicating the start and end of an application command by using the setsockopt () system call or a unique function for a command required by the application for protocol processing (step B1 in FIG. 4). ). The registered pattern is registered in the pattern detection unit 24 of the network interface card 2. Thereafter, the application issues a read () system call in order to acquire protocol data from the OS, and enters a state of waiting for data from the OS (standby state) (step B2 in FIG. 4).

  The OS refers to the network interface command receiving unit 51 on the main storage unit 5 (step B3 in FIG. 4), and information (command) related to the message stream transferred from the network interface card 2 to the protocol processing data storage unit 52 arrives. It is determined whether or not (Step B4 in FIG. 4). With the arrival of this command, it can be determined that the message stream has been transferred to the protocol processing data storage unit 52. The OS waits until a command arrives.

  When the command arrives, the OS copies the message stream area transferred from the network interface card 2 to the protocol processing data storage unit 52 to the application user data storage unit 53 (step B5 in FIG. 4). The application is notified (step B6 in FIG. 4).

  The application that has received the notification ends the read () system call (step B7 in FIG. 4). The application can use the data obtained by the read () system call in response to the notification from the OS. Thereafter, the processing returns to the CPU processing in order to perform the next packet processing.

  Conventionally, when transferring a message stream to an upper layer application, the command delimiter required by the upper layer application layer is not considered, so the application reads the message stream from the socket buffer using the read () system call. Later, it was necessary to search the command delimiter of the application. On the other hand, according to the present embodiment, by registering the pattern of the processing unit of the application in advance, the network interface card notifies the application when the data of the processing unit of the application is received. Will receive data at each command break. Therefore, it becomes unnecessary for the application to search for a command delimiter for each processing unit, the CPU load is reduced correspondingly, and protocol processing can be performed at high speed. In particular, when the server of this embodiment is applied to an SIP or SOAP server that accepts application processing from a plurality of clients, a great effect can be obtained in reducing the load on the CPU and speeding up the protocol processing.

(Second Embodiment)
When performing application protocol processing such as SMTP generally used in electronic mail, a unit of the SMTP command is always delimited by a line feed code (<CR> + <LF>). That is, for SMTP processing, the command end pattern is always a line feed code pattern. Here, an embodiment optimal for a system that executes command processing for each line feed such as SMTP processing will be described.

  Since the basic configuration of the application processing server according to the second embodiment of the present invention is the same as the configuration shown in FIG. 1, detailed description of each part is omitted. However, the detection pattern registered in the application is only a line feed code. When there is only one type of detection pattern, the upper layer is notified whenever a pattern matching the detection pattern is found in the message stream.

  The processing from receiving data from the opposite host to writing the data in the received message holding unit 23 is as shown in the flow of FIG. 2, but the processing after step A7 is the same as that of the first embodiment. Different. FIG. 5 shows an operation flow of the application processing server of this embodiment. The operation will be described below with reference to FIGS.

  In step A7, it is determined whether or not the line feed code (<CR> + <LF>) is registered in the pattern detection unit 24 as a detection pattern in the network interface card 2. When the line feed code is not registered, the process proceeds to step A12-1 in FIG.

  When the line feed code is registered, the pattern detection unit 24 searches the message stream held in the received message holding unit 23 for the line feed code (step A8-1 in FIG. 5), and whether the line feed code is found. It is determined whether or not (step A9-1 in FIG. 5).

  When a line feed code is found, the pattern detection unit 24 transfers the message from the beginning of the message stream held in the received message holding unit 23 to the line feed code to the protocol processing data storage unit 52 (step A10 in FIG. 5). -1). After transferring the message stream, the pattern detection unit 24 writes information (command) about the transferred message stream into the network interface command reception unit 51 on the main storage unit 5 (step A11-1 in FIG. 5). Thereafter, the process returns to the data reception process.

  If “No” is determined in any of the determinations in steps A7 and A9-1, the network interface card 2 determines whether or not any of the above-described conditions C1 to C3 is satisfied (FIG. 5). A12-1).

  When any of the conditions C1 to C3 is satisfied, the message stream existing in the reception buffer (reception message holding unit 23) is transferred to the protocol processing data storage unit 52 (step A13-1 in FIG. 5), and the transferred message is transferred. Information (command) about the stream is written in the network interface command receiving unit 51 (step A14-1 in FIG. 5). Thereafter, the process returns to the data reception process.

  If none of the conditions C1 to C3 is satisfied, the process returns to the data reception process to perform the next packet process.

  According to the present embodiment, for a protocol that performs command processing in linefeed units such as SMTP, an application can receive data for each linefeed from a read () system call by registering a linebreak pattern in advance. Yes, it can reduce the processing load for searching for line breaks on the application side.

(Third embodiment)
In contrast to the regular expression search that expresses several character strings in one format, in the fixed pattern search, it is only necessary to determine whether or not a specific pattern is included in the message stream. Therefore, by applying a fixed pattern search to the pattern search logic, the number of logics can be reduced and high-speed search becomes possible. Here, an embodiment to which fixed pattern search is applied will be described.

  FIG. 6 is a block diagram showing a configuration of an application processing server according to the third embodiment of the present invention. Referring to FIG. 6, the application processing server 100 includes a fixed pattern detection unit 25 having logic specialized for searching for a fixed pattern in the pattern detection unit 24, and this point is the same as that of the first embodiment. Different from the one.

  When the pattern notified from the application is a fixed pattern, the search logic of the pattern search unit 24 is dynamically switched to the search logic of the fixed pattern search unit 25. Then, instead of the regular expression search by the pattern search unit 24, a fixed pattern search by the fixed pattern search unit 25 is performed. Other operations are the same as those in the first embodiment.

  Thus, by dynamically switching from the search logic of the pattern search unit 24 to the search logic of the fixed pattern search unit 25, it becomes possible to search at high speed, and as a result, the time required for pattern detection of the received message stream Shortened.

  Hereinafter, a specific operation example of the application processing server of the present invention will be described.

  Here, an operation example when applied to SOAP will be described.

  SOAP is a protocol for software to exchange messages (objects) (remote procedure call-remote procedure call). Generally, in an application processing server that implements the SOAP protocol, SOAP messages for remote procedures are sent from a plurality of clients. However, since SOAP protocol processing becomes a bottleneck, it is required to reduce the load on the server application. ing.

In the case of a shopping site, the following SOAP message is sent from the client to the application processing server.
[SOAP message]
<SOAP-ENV: Envelope xmlns: SOAP-ENV = "http://schemas.xmlsoap.org/soap/envelope/">
<SOAP-ENV: Body>
<getProductDetails xmlns = "http://warehouse.example.com/ws">
<productId> 827635 </ productId>
</ getProductDetails>
</ SOAP-ENV: Body>
</ SOAP-ENV: Envelope>
As described above, the SOAP message has a pattern surrounded by the character string of <SOAP-ENV: Envelope>... </ SOAP-ENV: Envelope>. This character string is approximately 270 bytes of data. Assume that this message arrives from the opposite network after the session is established. In the conventional network interface, even if the SOAP message arrives at the network interface card,
(1) TCP PSH flag is ON,
(2) Receive buffer is FULL
(3) The network interface card does not transfer the arrived SOAP message to the main storage unit until one of the three conditions such as occurrence of timeout is met. For example, the TCP flag of the received packet is normally only ACK for data and the size of the reception buffer is 64 Kbytes. In this case, the next buffer is received and the reception buffer becomes FULL or a timeout (TCP Chimney Until 200 msec), the received packet remains stored in the network interface card.

  On the other hand, the application processing server of the present invention operates as follows.

  In the application processing server, the start pattern and end pattern are registered with the following regular expressions in advance using the setsockopt () function or the like.

Start pattern: / \ <SOAP-ENV \: Envelope * /
End pattern / \ <\ / SOAP-ENV \: Envelope \> /
In the above case, the setsockopt () function includes the following.
setsockopt () function
setsockopt (socket, SOL_SOCKET, SO_REGX, & rcv_regx, sizeof (struct rcv_regx));
struct rcv_regx [
char start_reg [STR_LEN]; =>"/ \ <SOAP-ENV \: Envelope * /"
char end_reg [STR_LEN]; =>"/ \ <\ / SOAP-ENV \: Envelope \>/"
];
The start pattern matches the first line of the SOAP message and the end pattern matches the last line of the SOAP message. When the network interface card 2 receives a TCP packet having this message stream as a payload from the opposite network 10, first, the packet receiving unit 21 checks the received TCP packet and performs TCP protocol processing. Thereafter, the processing is passed to the message reconstruction unit 22.

  The message reconstruction unit 22 constructs a message stream by referring to the TCP packet received this time and the previously received TCP data. The TCP packet received this time is combined with the packet received earlier than that and stored in the received message holding unit 23 as a message stream.

  In the pattern detection unit 24, the application data start and end patterns registered in advance are registered by the application. When a message stream that is a pattern detection target (hereinafter referred to as a detection target message stream) is stored in the received message holding unit 23, a pattern is not being detected for a message stream stored before the detection target message stream. In this case, the pattern detection flag of the internal parameter is OFF.

  If the pattern detection flag is OFF, the pattern detection unit 24 checks whether there is a registered start pattern for the detection target message stream. In this case, since the start pattern is found for the detection target message stream, the pattern matching flag is turned ON. Thereafter, it is checked whether a registered end pattern exists for the detection target message stream. As for the detection target message stream, since the end pattern is detected at a position after the start pattern, the network interface card 2 determines the message from the start of the detection target message stream held in the received message holding unit 23 to the end pattern. Is transferred to the protocol processing data storage unit 52, and information (command) related to the transferred message is written into the network interface command receiving unit 51.

  On the other hand, since the CPU 3 receives a message sent from the network interface card 2, it issues a read () system call and is in a data waiting state (standby state). In this standby state, the CPU 3 polls the network interface command receiving unit 51 to check whether a command has arrived. Instead of this polling processing, it is also possible to use a format in which the network interface 2 sends an interrupt notification to the CPU 3 after transferring the command. In the case of polling processing, when the message stream is written in the main storage unit 5, the network interface command receiving unit 51 is updated, so that the CPU 3 can recognize packet reception based on the update.

  When the packet reception is recognized by the CPU 3, the OS determines that the message stream transferred from the network interface card 2 to the protocol processing data storage unit 52 conforms to the pattern based on the information of the network interface command reception unit 51. Judge whether there is. If the pattern conforms, the OS copies the message stream stored in the protocol processing data storage unit 52 to the application user data storage unit 53. If the pattern does not conform, the OS refers to the argument of the system call, and the message stream having the length received from the application among the message streams stored in the protocol processing data storage unit 52, Copy to the application user data storage unit 53.

  After copying of the message stream to the application user data storage unit 53 is completed, the OS notifies the application to that effect. The application can refer to the data returned from the read () system call upon receiving a notification from the OS.

  In this way, by performing message transfer in consideration of the protocol command boundary of the application, the processing load (CPU load) of command boundary search on the application side is reduced.

  Here, an operation example when applied to SMTP will be described.

The SMTP protocol performs command transmission / reception processing in line feed units. Here, it is assumed that the application processing server 1 is a mail server that performs SMTP processing, and packets 1-6 are sent from the client side in the following order.
[Packet 1]
HELO [client-side hostname]
[Packet 2]
MAIL FROM: <sender's email address>
[Packet 3]
RCPT TO: <destination email address>
[Packet 4]
DATA
[Packet 5]
Mail content
[Packet 6]
QUIT
In the example of the packet 1-6, all the mail contents other than the mail content of the packet 5 are SMTP commands separated by line feed codes. Normally, SMTP commands are sent with the PSH flag attached, but the PSH flag may disappear if there is a TCP implementation or a proxy and application firewall that terminates TCP in the middle of a passing point. The present embodiment is effective for protocol processing that receives an SMTP command without such a PSH flag and performs command processing in line feed units.

  Specific operations will be described below. Here, it is assumed that a line feed code (<CR> + <LF>) is registered as a detection pattern.

  The network interface card 2 receives the TCP data containing the SMTP command from the opposite network 10, reconstructs the message stream from the received TCP data, and stores it in the received message holding unit 23. This process is basically the same as the process in the first embodiment.

  Subsequently, the pattern detection unit 24 searches the message stream held in the received message holding unit 23 for the line feed code. When a line feed code is found, a message from the beginning of the message stream held in the received message holding unit 23 to the line feed code is transferred from the network interface card 2 to the protocol processing data storage unit 52. After the message stream transfer, the network interface card 2 writes information (command) about the transfer message stream to the network interface command receiving unit 51 on the main storage unit 5.

  The CPU 3 refers to the network interface command receiving unit 51 and recognizes packet reception. When packet reception is recognized, the OS copies the message stream stored in the protocol processing data storage unit 52 to the application user data storage unit 53 and notifies the application to that effect. In this way, the application is notified of receipt with line breaks.

  In application protocols such as SMTP, where it is known that command processing is performed in units of line breaks, it is possible to reduce the search load for line break codes on the application side by registering only line break codes as patterns. .

  Here, an operation example when applied to SIP will be described.

In the SIP protocol, a SIP message including a SIP header field, a blank line, and a SIP body is processed. A SIP message of the following format is sent from the opposite network 10 to the network interface card 2 as a SIP request.
[SIP message]
INVITE sip: bob@biloxi.com SIP / 2.0
Via: SIP / 2.0 / UDP pc33.atlanta.com; branch = z9hG4bK776asdhds
Max-Forwards: 70
To: Bob <sip: bob@biloxi.com>
From: Alice <sip: alice@atlanta.com>; tag = 1928301774
Call-ID: a84b4c76e66710@pc33.atlanta.com
CSeq: 314159 INVITE
Contact: <sip: alice@pc33.atlanta.com>
Content-Type: application / sdp
Content-Length: 142
(Blank line)
(Hereafter, detailed information field of transmitted data)
In the SIP message, the field before the blank line is SIP command information. Detailed fields of data to be transmitted (for example, compression codec of audio file, sampling rate) and the like follow in the field after the blank line.

  After receiving the SIP message from the client, the SIP server performs application processing including processing for determining whether the destination of the SIP message is for itself or another host. Here, the application processing server 1 is a SIP server.

FIG. 7 shows an operation flow of the network interface card 2 mounted on the SIP server. In this example, it is assumed that the following patterns are registered as detection patterns.
-Start pattern (Is the INVITE character string at the beginning of the line?) "/ ^ \ INVITE /"
-End pattern (blank <CR> + <LF> + <CR> + <LF>) "/ \ r \ n \ r \ n /"
The network interface card 2 receives the TCP data containing the SIP message from the opposite network 10, reconstructs the message stream from the received TCP data, and stores it in the received message holding unit 23. This process is basically the same as the process in the first embodiment.

  Subsequently, it is determined whether a start pattern and an end pattern are registered in the pattern detection unit 24 as detection patterns (step A7 in FIG. 7). If the detection pattern is not registered, steps A16 to A18 in FIG. 7 are executed. These steps A16 to A18 are the same steps as the flow shown in FIG.

  If the detection pattern is registered, it is determined whether or not the pattern matching flag is ON. If the pattern matching flag is OFF, the pattern detection unit 24 checks whether there is a character string that matches the start pattern in the message stream held in the received message holding unit 23 (steps A9 and A10 in FIG. 7). ). If the message stream is the above SIP message, the first “INVITE sip: bob@biloxi.com SIP / 2.0” line matches the start pattern.

  After the start pattern match is confirmed, the pattern detection unit 24 turns on the pattern matching flag (step A11 in FIG. 7), and proceeds to step A12-2 in FIG. If the start pattern match is not confirmed, steps A16 to A18 are executed.

  When the determination in step A8 is “Yes” or when step A11 is executed, the pattern detection unit 24 checks whether there is a character string that matches the end pattern in the message stream (FIG. 7). Step 12-2, A13-2). If the message stream is the above SIP message, a blank line matches the end pattern. This pattern matching process is a process for checking whether there is a blank line after the start pattern.

  After the end pattern match is confirmed, the pattern detection unit 24 stores the data area from the head holding position to the blank line of the message stream held in the received message holding unit 23 from the received message holding unit 23 into the main memory. The data is transferred to the protocol processing data storage unit 52 of the unit 5, and the pattern matching flag is turned OFF (step A14-2 in FIG. 7).

  After transferring the message stream, the network interface card 2 writes information (command) about the transferred message stream to the network interface command receiving unit 51 on the main storage unit 5 (step A15 in FIG. 7).

  The CPU 3 refers to the network interface command receiving unit 51 and recognizes packet reception. When packet reception is recognized, the OS copies the message stream stored in the protocol processing data storage unit 52 to the application user data storage unit 53 and notifies the application to that effect. In this way, the application is notified of reception at the SIP command pattern delimiter.

  With the above operation, it is possible to reduce the search load on the application-side SIP command pattern break.

  In the network communication system (application processing server) of the present invention described above, the operation of the application and OS and the operation of the network interface unit can be basically realized by a program. The CPU and the network interface unit execute processing corresponding to each operation according to the program. The program may be provided by a recording medium such as a CD-ROM or a DVD.

  The user can input and register necessary information through an input device (such as a keyboard and a mouse) of a network communication system (application processing server). Specifically, the user can register an arbitrary pattern as the detection pattern. Also, the user can re-register the detection pattern arbitrarily during packet reception.

It is a block diagram which shows the structure of the application processing server which is the 1st Embodiment of this invention. 3 is a flowchart for explaining the operation of the network interface card shown in FIG. 1. 3 is a flowchart for explaining the operation of the network interface card shown in FIG. 1. 3 is a flowchart for explaining the operation of a CPU shown in FIG. 1. It is a flowchart for demonstrating operation | movement of the application processing server which is the 2nd Embodiment of this invention. It is a block diagram which shows the structure of the application processing server which is the 3rd Embodiment of this invention. It is a flowchart for demonstrating operation | movement of a network interface card at the time of applying the network communication system of this invention to a SIP server.

Explanation of symbols

1 Application processing server 2 Network interface card 3 CPU
4 access arbitration unit 5 main storage unit 10 opposite network 21 packet reception unit 22 message reconstruction unit 23 received message holding unit 24 pattern detection unit 51 network interface command reception unit 52 protocol processing data storage unit 53 application user data storage unit

Claims (18)

A main memory unit;
A network interface unit that receives a packet sent from the opposite host, reconstructs a message stream based on the received packet, and writes the reconstructed message stream in the main storage unit;
An application execution unit that interprets the message stream written in the main storage unit as a command and executes communication processing with the opposite host;
When there is a pattern that matches a detection pattern registered in advance in the reconstructed message stream, the network interface unit determines that the conformance pattern is a delimiter of the command, and determines the reconstructed message stream as A network communication system, wherein each command is written to the main storage unit. The network interface unit
A received message holding unit for holding the reconstructed message stream;
A pattern detection unit that checks whether or not there is the matching pattern in the message stream held by the received message holding unit,
The data from the head of the message stream held in the received message holding unit to the matching pattern is written in the main storage unit when the matching pattern is detected by the pattern detection unit. Network communication system. The detection pattern includes a start pattern and an end pattern indicating the start and end of the command, respectively.
The network interface unit
A received message holding unit for holding the reconstructed message stream;
A pattern detection unit that checks whether there is a pattern that matches each of the start pattern and end pattern in the message stream held by the received message holding unit,
When the pattern detection unit detects a pattern that matches each of the start pattern and the end pattern, from the pattern that matches the start pattern to the pattern that matches the end pattern in the message stream held by the received message holding unit. The network communication system according to claim 1, wherein the data is written in the main storage unit. An internal flag that is enabled when a pattern that matches the start pattern is detected, and disabled when a pattern that matches the end pattern is detected;
The network communication system according to claim 3, wherein the pattern detection unit determines data from a pattern that matches the start pattern to a pattern that matches the end pattern with reference to the internal flag.   5. The network communication system according to claim 1, wherein the detection pattern is notified to the pattern detection unit by the application execution unit when communication with the opposite host is started. 6.   The network communication system according to claim 1, wherein the detection pattern is an arbitrary pattern. The detection pattern is a pattern represented by a regular expression that represents a character string in a predetermined format.
The network communication system according to claim 1, wherein the pattern detection unit includes a regular expression search logic capable of detecting a pattern with the regular expression. The detection pattern further includes a fixed pattern,
The pattern detection unit further includes a fixed pattern search logic capable of pattern detection with the fixed pattern,
The network communication system according to claim 7, wherein when the detection pattern notified from the application execution unit is the fixed pattern, the regular expression search logic is dynamically switched to the fixed pattern search logic. An application notification method performed in a server device having a main storage unit, a network interface unit and a CPU,
A first step in which the network interface unit receives a packet sent from the opposite host, reconstructs a message stream based on the received packet, and writes the reconstructed message stream in the main storage unit;
A second step in which an application running on the CPU interprets a message stream written in the main storage unit as a command and executes a communication process with the opposite host;
In the first step, when the network interface unit includes a pattern that matches a detection pattern registered in advance in the reconstructed message stream, the matching pattern is determined as a delimiter of the command, An application notification method including a step of writing the reconstructed message stream to the main storage unit for each command.   In the first step, the network interface unit holds the reconstructed message stream, checks whether or not the matching pattern exists in the held message stream, and if the matching pattern exists, The application notification method according to claim 9, further comprising a step of writing data from a head of the retained message stream to the matching pattern in the main storage unit. The detection pattern includes a start pattern and an end pattern indicating the start and end of the command, respectively.
In the first step, the network interface unit holds the reconstructed message stream, checks whether there is a pattern that matches the start pattern and the end pattern in the held message stream, and When there is a pattern that matches each of the start pattern and the end pattern, the step of writing data from the pattern that matches the start pattern to the pattern that matches the end pattern of the held message stream in the main storage unit is further included. The application notification method according to claim 9, further comprising: An internal flag is provided in the network interface unit that is enabled when a pattern that matches the start pattern is detected and disabled when a pattern that matches the end pattern is detected,
The first step further includes a step in which the network interface unit determines data from a pattern matching the start pattern to a pattern matching the end pattern with reference to the internal flag. Application notification method described in 1.   The application notification method according to any one of claims 9 to 12, further comprising a step in which the application notifies the network interface unit of the detection pattern when communication with the opposite host is started. .   The application notification method according to claim 9, wherein the detection pattern is an arbitrary pattern.   The application notification method according to claim 9, wherein the detection pattern is a pattern represented by a regular expression that represents a character string in a predetermined format. The detection pattern further includes a fixed pattern,
In the first step, when the detection pattern notified from the application is the fixed pattern, pattern detection with the fixed pattern is possible from a regular expression search logic capable of pattern detection with the regular expression. The application notification method according to claim 15, further comprising a step of dynamically switching to a fixed pattern search logic. Receiving a packet sent from the opposite host, reconstructing a message stream based on the received packet, causing the network interface card to execute a first process of writing the reconstructed message stream to the main storage unit;
A program that causes a CPU to execute a second process of interpreting a message stream written in the main storage unit as a command and executing a communication process with the opposite host,
In the first process, when there is a pattern that matches a detection pattern registered in advance in the reconstructed message stream, the conforming pattern is determined as a delimiter of the command, and the reconstructed message stream Including a process of writing to the main storage unit for each command. A network interface card mounted on a server device having a main storage unit and an application execution unit that interprets a message stream written in the main storage unit as a command and executes a communication process with an opposite host,
A packet receiver for receiving a packet sent from the opposite host;
A message reconstruction unit for reconstructing a message stream based on the packet received by the packet reception unit;
A received message holding unit for holding the message stream reconstructed by the message reconstructing unit;
A pattern detection unit that checks whether there is a matching pattern that matches a detection pattern registered in advance in the message stream held in the received message holding unit,
When the matching pattern is detected by the pattern detection unit, data from the head of the message stream held by the received message holding unit to the matching pattern is written to the main storage unit.

Images