JP2011191901A - Signal processor and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To efficiently process a received signal, in a signal processor. <P>SOLUTION: The signal processor processing a signal received from the outside includes: a means accumulating the received signals; and a means starting a plurality of signal processing threads for processing the accumulated received signals when the number of the accumulated received signals reaches a predetermined number or when a predetermined period lapses, allocating the accumulated received signals to the signal processing threads, and processing subthreads of a part or all processes of subthreads of the processes constituting each signal processing thread, in parallel, by use of a parallel processing processor between the plurality of signal processing threads. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a signal processing apparatus and a program, and can be applied to, for example, processing of a SIP (Session Initiation Protocol) signal.

  Conventionally, as a signal processing device (SIP server) for processing a SIP signal, there is a SIP server disclosed in Patent Document 1.

  In a conventional SIP server such as the SIP server described in Patent Document 1, generally, as shown in FIG. 5, a plurality of SIP signals are transmitted and received between a plurality of SIP terminals at the same time, and a plurality of SIP signals are transmitted between SIP servers. There are cases where data is transmitted and received simultaneously.

  For this reason, the conventional SIP server program generally has a structure in which a thread for processing a plurality of SIP signals is activated to simultaneously process transmission / reception of a plurality of SIP signals.

  A conventional SIP server program includes, for example, one reception thread and a plurality of execution threads, and each thread operates in parallel.

  The time required for the reception thread to receive one signal and activate the execution thread takes only a negligible time compared to the time during which one execution thread operates.

  For this reason, when a plurality of signals are continuously received, a plurality of execution threads are activated, and the activated plurality of threads execute processing for each received signal in parallel.

  However, the number of threads that can be physically executed by one CPU in parallel is usually an order of magnitude, not a very large number.

  Therefore, as the number of threads that are activated simultaneously is increased, the processing of individual threads becomes increasingly slower, and there is an upper limit on the number of execution threads that can be processed within a certain time.

  Conventionally, as a solution to this problem, for example, there is a case where the number of execution threads on a SIP server is limited so as not to be activated.

JP 2009-124572 A

  However, a signal processing apparatus such as a SIP server is more essential in providing a service than a service process that performs a process corresponding to a received signal, which is an analysis process (lexical analysis, syntax analysis) of the content of the received signal. Etc.) and the assembly process of the transmission signal may take more time.

  Further, the analysis process and the assembly process described above are CPU bound processes in which a block for input / output or the like does not occur in the middle and the CPU is continuously used. As a result, the processing capability of a signal processing device such as a SIP server has a structure in which it is difficult to improve the processing capability because the analysis processing and assembly processing described above become a bottleneck.

  Therefore, a signal processing apparatus and program that can efficiently process received signals are desired.

  According to a first aspect of the present invention, in a signal processing apparatus for processing a received signal received from the outside, (1) a parallel processing processor for processing a plurality of data in parallel, (2) storage means for storing received signals, 3) When the number of received signals accumulated by the accumulating means reaches a predetermined number, or when a predetermined period has elapsed, a plurality of signal processing threads for processing the received signals accumulated by the accumulating means are activated, The reception signal accumulated by the accumulation means is assigned to each signal processing thread, and a plurality of signal processing threads are provided for some or all of the process sub-threads constituting each signal processing thread. Thread control means for performing parallel processing using the parallel processing processor.

  According to a second aspect of the present invention, in the signal processing program, (1) a computer provided with a parallel processing processor for processing a plurality of data in parallel, mounted on a signal processing device for processing a received signal received from outside (2) ) Storage means for storing received signals; and (3) received signals stored by the storage means when the number of received signals stored by the storage means reaches a predetermined number or when a predetermined period has elapsed. A plurality of signal processing threads are started, and the reception signals accumulated by the storage means are allocated to each signal processing thread, and some or all of the sub-threads of each process constituting each signal processing thread Process sub-threads function as thread control means for processing in parallel using a parallel processor among a plurality of signal processing threads. And wherein the door.

  According to the present invention, a received signal can be efficiently processed in a signal processing device.

It is the block diagram shown about the functional structure of the SIP server which concerns on embodiment. It is the flowchart shown about the signal reception process in the SIP server which concerns on embodiment. It is the flowchart (1) shown about the execution control process in the SIP server which concerns on embodiment. It is the flowchart (2) shown about the execution control process in the SIP server which concerns on embodiment. It is the block diagram shown about the connection structure between the conventional SIP server and a SIP terminal.

(A) Main Embodiment Hereinafter, an embodiment of a signal processing device and a program according to the present invention will be described in detail with reference to the drawings. In this embodiment, an example in which the signal processing apparatus of the present invention is applied to a SIP server is described.

(A-1) Configuration of Embodiment FIG. 1 is a block diagram showing a functional configuration of the SIP server 1 of this embodiment.

  The SIP server 1 receives a reception signal SR from a transmission source (another SIP server, SIP terminal, or the like), performs a predetermined process based on the content of the reception signal SR, and processes the transmission signal SS corresponding to the process. To a transmission destination (another SIP server, SIP terminal, etc.) according to The transmission of the transmission signal SS is performed as necessary from the processing content of the reception signal SR, and may not be transmitted depending on the processing content. Examples of the SIP server 1 include a device that performs signaling such as call control between SIP terminals.

  The SIP server 1 installs the signal processing program of the embodiment in an information processing apparatus including a SIMD (Single Instruction Multiple Data) processor such as a GPU (Graphics Processing Unit) and a normal CPU, for example. Even in this case, the functional configuration can be shown as shown in FIG.

  The SIMD type processor has a higher calculation performance than a normal CPU by processing a plurality of data simultaneously for one instruction. The SIP server 1 is configured to efficiently perform processing of the received SIP signal (particularly analysis processing of the received signal SR and assembly processing of the transmission signal SS) by using this SIMD type processor.

  The SIMD type processor was originally used to realize a supercomputer, but recently it has been applied to a GPU for 3D graphic display mounted on a PC or the like. Since the number of threads that can be processed simultaneously by the SIMD type processor is two or more digits larger than that of the CPU, any process that can be easily applied to the SIMD type processor can be processed with much higher performance than the CPU. However, since the processing capability can be improved by using the SIMD type processor when a plurality of data can be processed at the same time by one instruction, the same processing should be executed as simultaneously as possible.

  FIG. 1 shows the structure of a signal processing program (thread) in the SIP server 1 for improving the processing capability of the SIP server using a SIMD type processor such as a GPU. In addition, as the above-described information processing apparatus constituting the SIP server 1, for example, a PC equipped with an existing GPU can be applied.

  As shown in FIG. 1, the SIP server 1 includes one receiving thread 10, a plurality of execution threads 20, a plurality of analysis threads 30 (30-1, 30-2), and a plurality of processing threads 40 (40-1, 40). -2), a plurality of assembly threads 50 (50-1, 50-2) and a plurality of transmission threads 60 (60-1, 60-2), each thread operating in parallel.

  In FIG. 1, one execution thread 20, one analysis thread 30, processing thread 40, assembly thread 50, and transmission thread 60 are described as two configurations, but the number of these threads activated simultaneously. Is not limited.

  Among these threads, threads that are executed by a normal CPU are a reception thread 10, an execution thread 20, a processing thread 40, and a transmission thread 60, and threads that are executed by a SIMD type processor are an analysis thread 30 and an assembly thread. 50.

  The reception thread 10 is a thread for receiving a reception signal SR from outside the SIP server 1 (another SIP server or SIP terminal). In the reception thread 10, a signal reception process 11 operates.

  Each time the signal reception process 11 receives a received signal SR from another SIP terminal, SIP server, or the like, the received signal text D1 obtained by converting the received signal SR into a text format is converted into a corresponding memory area (hereinafter referred to as “area”). ) And after receiving a predetermined number of reception signals SR, the execution thread 20 is activated.

  In FIG. 1, as an example, the reception thread 10 receives two reception signals SR-1 and SR-2, and converts the reception signals SR-1 and SR-2 into texts, respectively. 1 and D1-2.

  The execution thread 20 is a thread that controls execution of a series of processes for each of a plurality of received signals. In the execution thread 20, an execution control process 21 operates.

  In the execution control process 21, in order to process a plurality of signals at once, the analysis thread 30, the processing thread 40, the assembly thread 50, and the transmission thread 60 are activated and assigned by the number of received reception signals SR. Take control.

  At this time, it is assumed that the analysis thread 30, the processing thread 40, the assembly thread 50, and the transmission thread 60 are activated one by one for one received signal SR.

  In the analysis threads 30-1 and 30-2, the lexical analysis processes 31-1 and 31-2 and the syntax analysis processes 32-1 and 32-2 are sequentially executed.

  In the lexical analysis processes 31-1 and 31-2, text-format received signals are input from the areas of the received signal text D1-1 and D1-2, respectively, and they are converted into lexical characters such as character strings, numerical values, IP addresses, and FQDNs. The result of the conversion is output to the areas of the lexical analysis results D2-1 and D2-2.

  The parsing processes 32-1 and 32-2 input lexical sequences for received signals from the areas of the lexical analysis results D2-1 and D2-2, extract received signal information such as methods and parameters, and the internal format The received signal is output to the areas of received signal data D3-1 and D3-2.

  In the processing threads 40-1 and 40-2, service processes 41-1 and 41-2 are executed, respectively.

  The service processes 41-1 and 41-2 receive the received signal in the internal format from the areas of the received signal data D 3-1 and D 3-2 and execute processes corresponding to the signals. When the service processing 41-1 and 41-2 transmit a signal during the execution of the processing, the information of the transmission signal in the internal format is output to the areas of the transmission signal data D4-1 and D4-2.

  In the assembly threads 50-1 and 50-2, signal assembly processes 51-1 and 51-2 are executed, respectively.

  The signal assembling processes 51-1 and 51-2 receive the transmission signal information in the internal format from the areas of the transmission signal data D 4-1 and D 4-2, convert it into the text format, and transmit the text format transmission signal. Are output to the areas of the transmission signal texts D5-1 and D5-2.

  In the transmission threads 60-1 and 60-2, signal transmission processing 61-1 and 61-2 are executed, respectively.

  In the signal transmission processing 61-1 and 61-2, a transmission signal in text format is input from the areas of the transmission signal texts D5-1 and D5-2, and transmitted to the outside of the SIP server (such as another SIP server or SIP terminal). Signals SS-1 and SS-2 are transmitted.

  The processing contents in each of the analysis thread 30, the processing thread 40, the assembly thread 50, and the transmission thread 60 can be applied to the processing contents similar to those of the existing SIP server. .

  In the SIP server 1, the upper limit of the number of threads executed simultaneously is the upper limit value N. N is an arbitrary number, and may be determined according to the scale of processors (ordinary CPU and MIMD type processor) mounted on the SIP server 1, for example.

(A-2) Operation | movement of embodiment Next, operation | movement of the SIP server 1 of this embodiment which has the above structures is demonstrated.

  First, the operation of the reception thread 10 (signal reception processing 11) will be described.

  FIG. 2 is a flowchart showing the operation of the reception thread 10 (signal reception processing 11).

  The signal reception process 11 first collects the number of running execution threads (S101), and determines whether the number of execution threads is equal to or greater than the upper limit value N (S102).

  When the number of execution threads is equal to or greater than the upper limit value N in the determination in step S102 described above, the signal reception process 11 waits for a while (waiting for a predetermined time) to wait for the end of the execution thread (S103). The process returns to step S101 described above.

  On the other hand, if the number of execution threads is less than the upper limit value N in the determination in step S102 described above, the signal reception process 11 secures M areas of the reception signal text D1 (S104) and receives the reception signal SR (S105). ). Note that M is an arbitrary number. For example, M is determined according to the scale of processors (ordinary CPU and MIMD type processor) installed in the SIP server 1 and the distribution situation of the timing at which the received signal SR arrives. Also good.

  At this time, in the above-described step S105, the process is blocked (waiting) until the first reception signal SR is received. When the first reception signal SR is received, the first of the areas of the M reception signal texts D1, The reception signal SR received first is output (S106), and the current time is set as the start time TS (S107).

  Next, it is determined whether the number of outputs of the received signal SR to the area of the received signal text D1 is less than M and the difference between the current time and the start time TS is less than the upper limit value TM (arbitrary time) (S108). ).

  If the number of outputs of the reception signal SR to the area of the reception signal text D1 is less than M and the difference between the current time and the start time TS is less than the upper limit value TM in the determination in step S108, signal reception processing is performed. 11 waits until the next reception signal SR is received (S109).

  In step S109 described above, a setting is made to time out if the next reception signal SR does not arrive even after a predetermined time has elapsed, and after step S109 described above, it is determined whether timeout or signal reception has occurred (S110). .

  If it is determined in step S110 that a timeout has occurred, the process returns to step S108.

  When the reception signal SR is newly received in the determination in step S110 described above, the reception signal SR is output in order in the second and subsequent areas out of the M reception signal text D1 areas (S111). Return to S108.

  On the other hand, if the number of outputs of the received signal SR to the area of the received signal text D1 is M or more or the difference between the current time and the start time TS is greater than or equal to the upper limit value TM in the determination in step S108 described above, It starts (S112) and returns to step S101.

  As described above, in the process of FIG. 2, the signal reception process 11 activates an execution thread and receives an execution thread on the SIP server 1 when M signals are received or when the time TM has elapsed since the first signal reception. Is limited to start up to a maximum of N.

  Next, the process of the execution thread 20 (execution control process 21) will be described.

  3 and 4 are flowcharts showing the process of the execution thread 20 (execution control process 21).

  3 and 4, “processing thread activation number s1” and “transmission thread activation number s2” to be described later are used as variables.

  In the execution control process 21, first, the processing thread activation number s1 is set to 0 (S201), and the M analysis threads 30 are activated (S202).

  As a result, the i-th analysis thread 30 among 1 to M inputs data from the area of the i-th received signal text D1, and outputs the result to the area of the i-th received signal data D3.

  The execution control process 21 collects the number of activated analysis threads 30 after the process of step S202 described above (S203), and determines whether the number of activated analysis threads 30 is greater than 0 or 0 (S204). ).

  If the number of activated analysis threads 30 is greater than 0 in the determination in step S204 described above, the value of (M-number of analysis threads 30-number of processing thread activations s1) is obtained, and the number of processing threads 40 corresponding to that number is obtained. Start up (S205).

  Thereafter, a value of (the number of M-analysis threads 30) is set as the processing thread activation number s1 (S206), and in order to wait for the end of the analysis thread 30 being executed, a little time is allowed (waiting for a predetermined time). From (S207), the process returns to the above-described step S203.

  If the number of activated analysis threads 30 is 0 in step S204 described above, the value of (M-processing thread activation number s1) is obtained, and the corresponding number of processing threads 40 are activated (S208).

  In the above-described step S205 or the above-described step S208, when the i-th analysis thread 30 in 1 to M is terminated and the i-th processing thread 40 is not activated, the i-th processing thread. 40 is activated. As a result, the i-th processing thread 40 in 1 to M inputs data from the i-th received signal data D3 area, and outputs the result to the i-th transmission signal data D4 area. .

  The execution control process 21 collects the number of active process threads 40 after the process of step S208 described above (S209), and determines whether the number of active process threads 40 is greater than 0 or 0 (S210). ).

  If the number of active processing threads is greater than 0 in the determination in step S210 described above, after waiting for a predetermined time (waiting for a predetermined time) to wait for the end of the processing threads (S211), The process returns to step S209.

  If the number of active processing threads is 0 in the determination in step S210 described above, 0 is set as the transmission thread activation number s2 (S212), and M assembly threads 50 are activated (S213).

  As a result, the i-th assembly thread among 1 to M inputs from the area of the i-th transmission signal data D4, and outputs the result to the area of the i-th transmission signal text D5.

  The execution control process 21 collects the number of activated assembly threads after the process of step S213 described above (S214), and determines whether the number of activated assembly threads is greater than 0 or 0 (S215).

  If it is determined in step S215 above that the number of active assembly threads is greater than 0, a value of (M-number of assembly threads 50-transmission thread activation number s2) is obtained, and the transmission threads 60 corresponding to that number are activated. (S216).

  Thereafter, a value of (M-number of assembly threads 50) is set as the number of activations of the transmission thread s2 (S217), and after a while (waiting for a predetermined time) to wait for the end of the assembly thread 50 (S218). ), The process returns to step S214.

  On the other hand, when the number of active assembly threads 50 is 0 in the determination in step SS215, the value of (M-transmission thread activation number s2) is obtained, and the transmission threads 60 corresponding to that number are activated (S219). .

  In the process of step SS216 or step S219 described above, when the i-th assembly thread 50 in 1 to M ends and the i-th transmission thread 60 is not activated, the i-th assembly thread 50 is not activated. The transmission thread 60 is activated. As a result, the i-th transmission thread among 1 to M inputs data from the area of the i-th transmission signal text D5, and generates and transmits a corresponding transmission signal SS.

  The execution control process 21 collects the number of active transmission threads 60 after the process of step S219 described above (S220), and determines whether the number of active transmission threads 60 is greater than 0 or 0 (S220). S221).

  If it is determined in step S221 that the number of active transmission threads is greater than 0, after waiting for a predetermined time (waiting for a predetermined time) to wait for the end of the transmission thread 60 (S222), the above-described steps The process returns to S220.

  On the other hand, if the number of active transmission threads is 0 in the determination in step S221 described above, the execution control process 21 ends.

  As shown in FIG. 3 and FIG. 4 described above, the execution control process 21 first activates the M analysis threads 30 and waits for one of the analysis threads 30 to end. When one of the analysis threads 30 is terminated, the corresponding processing threads 40 are sequentially activated. After the activation of up to M processing threads 40, the end of all the processing threads 40 is awaited. Further, when all the M processing threads 40 are finished, the M assembly threads 50 are activated, and the end of any one of the assembly threads 50 is awaited. When any one of the assembly threads 50 is terminated, the corresponding transmission threads 60 are sequentially activated. After the activation of up to M transmission threads 60, the completion of all the transmission threads 60 is awaited. When all the M transmission threads 60 are finished, the execution control process 21 is also finished.

(A-3) Effects of Embodiment According to this embodiment, the following effects can be achieved.

  As shown in FIGS. 3 and 4 described above, the execution control process 21 simultaneously activates M analysis threads 30 and assembly threads 50 to be executed for the SIMD type processor. Thereby, the same lexical analysis process 31, the same syntax analysis process 32, and the same signal assembly process 51 for a plurality of signals are performed in parallel by the SIMD type processor. In the SIMD type processor, since the number of threads that can be executed in parallel with the CPU is much larger, the processing for M signals can be performed in a shorter time than the CPU.

  Further, by executing the CPU bound lexical analysis processing 31, the syntax analysis processing 32, and the signal assembly processing 51 by the SIMD type processor, the load on the CPU itself is reduced. As a result, the number of execution threads 20, processing threads 40, and transmission threads 60 executed on the CPU side can be increased simultaneously. As a result, the processing on the received signal can be executed in parallel in a shorter time, so that the processing capacity of the SIP server 1 can be improved.

  Furthermore, since the GPU in the existing PC can be used as the SIMD type processor mounted on the SIP server 1, it can be constructed at low cost.

(A-4) Modification of Embodiment In the SIMD type processor, a plurality of data is processed in parallel with one instruction, but usually the number of threads for executing the same instruction at the same time is determined.

  Basically, threads executed simultaneously execute the same instructions in the process in order.

  However, if there is a branch in the process and threads that are executed at the same time include threads that branch in different directions, it takes an execution time to execute both branches.

  In this way, if a branch is mixed in a thread that is executed at the same time, the processing time of one thread becomes longer. Therefore, it is difficult to mix branches in a thread that is executed at the same time. It becomes a point to draw out.

  Therefore, in the SIP server 1 of the above-described embodiment, a modified embodiment that reduces branching in threads that are simultaneously executed for the lexical analysis processing 31, the syntax analysis processing 32, and the signal assembly processing 51 executed by the SIMD type processor. Is described below.

(A-4-1) Modified Example Regarding Lexical Analysis Processing In order to efficiently execute the lexical analysis of the SIP signal, a method of realizing the processing as a finite automaton that receives the character to be analyzed is often used.

  When this method is used, the processing of the lexical analysis processing 31 is realized as a finite automaton that inputs characters one by one from the beginning of the received signal text D1 and outputs the next state and the lexical analysis result D2 for each character.

  In the lexical analysis processing, if the processing for determining the next state and output is programmed with conditional branching for one input character, the processing time may become long when executed by a SIMD processor. is there.

  For this reason, the analysis thread 30 (lexical analysis processing 31) draws a table using the inputted one character as an index, and sets the next state and output from the values of the table (hereinafter referred to as “method 1A”). ) And branching can be reduced during processing, and can be efficiently executed by a SIMD type processor.

  Further, the number of times the state transition of the finite automaton is repeated for lexical analysis depends on the size of the input received signal. When threads having different received signal sizes are mixed among threads executed simultaneously, it takes time to process a received signal having the longest size among them. Therefore, by assigning threads to the analysis thread 30 (lexical analysis process 31) in the order of the size of the received signal (hereinafter referred to as “method 1B”), the variation in the received signal size within the simultaneously executed threads is reduced, and the thread Overall processing time can be reduced. That is, with respect to the analysis thread 30 (lexical analysis process 31), by assigning a group of threads to be processed simultaneously using a SIMD type processor according to the size of the received signal, a branch is mixed in the threads executed simultaneously. The possibility can be reduced.

(A-4-2) Modified Example Regarding Parsing Process In the parsing of the SIP signal, an algorithm called shift reduction parsing that uses the lexical to be analyzed as an input is often used in order to efficiently execute the parsing.

  When this method is used, the parsing process 32 executes a shift process for inputting a lexical phrase one by one from the beginning of the lexical analysis result D2 and putting it in the stack for each lexical phrase, or a reduction process for extracting from the stack. You may do it. When the reduction process is executed, the syntactical handling of the lexical phrase is determined, and the reception signal data D3 is output.

  In the analysis thread 30 (syntax analysis process 32), if a process for determining a value to be input / output to / from the stack at the time of a shift process or a reduction process for one input lexical word is programmed as a conditional branch, the SIMD processor When it is executed, the processing time becomes long. For this reason, the analysis thread 30 (syntax analysis process 32) draws a table using one input lexical character as an index, and determines a value to be input / output to / from the stack based on the value of the table (hereinafter, “method 2A”). ")", Branching can be reduced during processing, and efficient execution can be performed by a SIMD type processor.

  However, since it is unavoidable that the analysis thread 30 (syntax analysis process 32) determines whether to perform shift processing or reduction processing and branches the processing, branching cannot be eliminated at all. However, compared with the reduction process, the shift process is simple and most of the processing time is required for the reduction process. Therefore, even when the shift process and the reduction process branch, There is no significant reduction in processing efficiency.

  Even if there is a branch in the processing of the analysis thread 30 (syntactic analysis process 32), if there is no branch in the threads that are executed at the same time, the processing does not deteriorate. As a method of reducing, there are methods described below.

  In the case of a SIP signal, the type of signal is determined by the first word of the received signal. However, if the signals are of the same type, it can be expected that there is less variation in syntax than signals of different types. Therefore, with respect to the analysis thread 30 (syntactic analysis process 32), target data and threads are assigned to threads that are executed simultaneously so that only the same type of signal is targeted (hereinafter referred to as “method 2B”). You may do it.

  In addition, it is expected that there is little variation in syntax for signals of the same type received from the same transmission source (SIP server or SIP terminal). Therefore, with respect to the analysis thread 30 (lexical analysis process 31), the target data and the thread are allocated to the threads that are executed at the same time so that only the same type of signal received from the same SIP server is targeted (hereinafter referred to as the target thread). , “Method 2C”). Similarly, the target data and thread are assigned to the simultaneously executed threads so as to target only signals of the same type received from the SIP terminal group (hereinafter referred to as “method 2D”). May be.

(A-4-3) Modified Example Regarding Signal Assembly Processing The signal assembly processing 51 needs to change the syntax corresponding to the signal assembly method according to the signal type and the number of parameters set in the transmission signal data D4. And it is inherently difficult to reduce branching.

  However, there is a method described below as a method for reducing the possibility of mixing branches.

  If signals of the same type are used, it can be expected that there is less variation in syntax than signals of different types. Therefore, with respect to the assembly thread 50 (signal assembly processing 51), assignment of target data and threads is determined so that only threads of the same type are targeted for simultaneously executed threads (hereinafter referred to as “method 3A”). You may do it.

  Further, it can be expected that the same type of signal transmitted to the same destination (SIP server or SIP terminal) has little syntax variation. Therefore, it is also possible to determine the allocation of target data and threads so that only threads of the same type transmitted to the same server are targeted for threads executed simultaneously (hereinafter referred to as “method 3B”). good. Similarly, for the simultaneously executed threads, allocation of target data and threads is determined so that only signals of the same type transmitted to the SIP terminal group are targeted (hereinafter referred to as “method 3C”). You may do it.

  Furthermore, in the service processing 41, if the signals are of the same type output by the same service, it can be expected that there is less variation in syntax than signals output by different services. Therefore, the allocation of the target data and the thread is determined so that only the same type of signal output from the same service is targeted for the threads executed simultaneously (hereinafter referred to as “method 3D”). Also good.

(A-4-4) In the lexical analysis process 31, the branch itself can be reduced from the process by the above-described method 1A, and further, the possibility of a mixture of branches can be reduced by the above-described method 1B.

  In the parsing process 32, the branch itself can be reduced from the process by the above-described method 2A, and further, the possibility of a mixture of branches can be reduced by the above-described method 2B, method 2C, and method 2D.

  Further, in the signal assembling process 51, the possibility of mixing branches can be reduced by the above-described methods 3A, 3B, 3C, and 3D.

  With these methods, the SIP server 1 can reduce the occurrence of a mixture of branches in threads that are executed simultaneously, and can improve the processing efficiency of the SIMD type processor.

  However, the above-described method 1A and method 2A are not used for improving the processing performance of the SIMD type processor, but are used for the purpose of simplifying the program.

  Further, the above-described methods 1A, 1B, 2A to 2D, 3A to 3D may be applied to the SIP server 1 or only a part thereof.

(B) Other Embodiments The present invention is not limited to the above-described embodiments, and may include modified embodiments as exemplified below.

(B-1) In the above embodiment, the case where the lexical analysis processing 31, the syntax analysis processing 32, and the signal assembly processing 51 are processed by the SIMD type processor has been described as an example. However, the processing content of the service processing 41 is simple. For example, the service processing 41 may be processed by the SIMD type processor.

  In the above embodiment, the lexical analysis process 31, the syntax analysis process 32, and the signal assembly process 51 have been described as being processed by the SIMD type processor. However, some processes may be performed on the CPU side. .

(B-2) In the above embodiment, the example in which the signal processing apparatus of the present invention is applied to the SIP server has been described. However, the signal processing apparatus may be applied to an apparatus that processes signals of other protocols. For example, H.M. In a communication system using other than SIP such as H.248, it may be applied to an apparatus that performs signal processing (for example, call control signal processing). In addition, the present invention may be applied to a signal processing apparatus that performs a protocol process in which the signal content is text-based. In that case, each processing performed in each thread shown in FIG. 1 described above is replaced with processing contents in the signal processing apparatus.

(B-3) In addition to processing a signal, a processing thread (for example, a thread that compiles a program language or translates a natural language) that requires lexical analysis or syntax analysis of a large number of texts at the same time is executed. The device to be used may be assigned to the SIMD type processor in the same manner as the analysis thread 30, the processing thread 40, and the assembly thread 50 in the SIP server 1 described above.

(B-4) In the above embodiment, an example in which a SIMD type processor is applied as a processor that performs data processing in parallel in the SIP server 1 has been described. However, a MIMD (Multiple Instruction stream, Multiple Data stream) type processor is used. For example, a processor that performs other parallel processing may be applied.

  DESCRIPTION OF SYMBOLS 1 ... SIP server, SR, SR-1, SR-2 ... Reception signal, 10 ... Reception thread, 11 ... Signal reception process, 20 ... Execution thread, 21 ... Execution control process, D1, D1-1, D1-2 ... Received signal text, 30, 30-1, 30-2 ... analysis thread, 31, 31-1, 31-2 ... lexical analysis processing, D2, D2-1, D2-2 ... lexical analysis result, 32, 32-1 , 32-2 ... parsing process, D3, D3-1, D3-2 ... received signal data, 40, 40-1, 40-2 ... processing thread, 41, 41-1 and 41-2 ... service process, D4 , D4-1, D4-2 ... transmission signal data, 50, 50-1, 50-2 ... assembly thread, 51, 51-1, 51-2 ... signal assembly processing, D4, D4-1, D4-2 ... Send signal text, 60, 60-1, 60-2 ... send thread, 1,61-1,61-2 ... signal transmission processing, SS, SS-1, SS-2 ... transmission signal.

Claims (8)

In a signal processing device for processing a received signal received from the outside,
A parallel processor that processes multiple data in parallel;
Storage means for storing received signals;
When the number of received signals accumulated by the accumulating means reaches a predetermined number or when a predetermined period has elapsed, a plurality of signal processing threads for processing the received signals accumulated by the accumulating means are activated, The reception signal accumulated by the accumulation means is allocated to the signal processing thread, and among the sub-threads of each process constituting each signal processing thread, a part or all of the process sub-threads are between the plurality of signal processing threads. And a thread control unit configured to perform parallel processing using the parallel processing processor.   The thread control means causes the parallel processing processor to process at least a sub thread of a received signal analysis processing step among sub threads of each step constituting each signal processing thread. Item 2. The signal processing device according to Item 1.   The thread control means includes at least a sub thread of an assembly processing step of a transmission signal to be transmitted to a transmission destination according to a processing content based on a reception signal among sub threads of each step constituting each signal processing thread. The signal processing apparatus according to claim 1, wherein the signal processing apparatus performs processing using a parallel processor.   4. The signal processing according to claim 2, wherein the sub-thread of the analysis processing step includes at least lexical analysis processing of the received signal and syntax analysis processing for the processing result of the lexical analysis processing. apparatus.   The thread control means is a lexical analysis process performed by a sub-thread of an analysis processing step that constitutes each signal processing thread, and a signal that is simultaneously processed using the parallel processing processor according to the size of a reception signal to be processed. The signal processing apparatus according to claim 4, wherein a group of processing threads is assigned.   The thread control means is a syntactic analysis process performed by a sub-thread of an analysis processing step that constitutes each signal processing thread, and simultaneously performs the parallel processing processor according to the type or transmission source of the received signal related to the data to be processed. The signal processing device according to claim 4, wherein a group of signal processing threads to be processed is assigned.   The thread control means is a sub thread of the assembly processing step that constitutes each signal processing thread, and simultaneously uses the parallel processing processor according to the type, transmission destination, or service content related to the transmission signal to be processed. 4. The signal processing apparatus according to claim 3, wherein a group of signal processing threads to be processed is assigned. A computer equipped with a parallel processing processor for processing a plurality of data in parallel, mounted on a signal processing device for processing a received signal received from the outside,
Storage means for storing received signals;
When the number of received signals accumulated by the accumulating means reaches a predetermined number or when a predetermined period has elapsed, a plurality of signal processing threads for processing the received signals accumulated by the accumulating means are activated, The reception signal accumulated by the accumulation means is allocated to the signal processing thread, and among the sub-threads of each process constituting each signal processing thread, a part or all of the process sub-threads are between the plurality of signal processing threads. A signal processing program that functions as thread control means for performing parallel processing using the parallel processor.

Images