JPH01100657A - Simulator - Google Patents

Abstract

PURPOSE:To attain a high speed by a simple and small scale parallel processing by defining a part for executing a basic function to be simulated as a slave process and executing and controlling by a master process. CONSTITUTION:Constitution defining information is inputted to a network constitution definition file 4 according to the input menu display 2 of a constitution defining input part 1. The change, addition of the file and the assurance of a consistency between information in the file are executed by a network constitution input process 3 to input the constitution defining information by a program 5. After the completion of this input, the part 1 interrupts an SM executing process 11 executing the simulation (SM) to change and add the constitution and request for the continuation of the SM. The process 11 maintains a packet transfer between the slave processes for executing the basic function of the SM and a time relation on a communication process. The output of a SM executing result is requested according to the interruption of the master process to maintain the time relation in this case.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to simulation technology, and particularly to a simulation execution device suitable for queuing problems.

[Conventional technology]

Conventionally, GPS is a typical discrete system simulator that handles queuing problems using general-purpose computers.
S (General Purpose Sim
system) and S IMSCRI
PT (SIMulationS CRI P To
r), etc., and are frequently used in large-scale computers.

For example, in the case of GPSS, the path that a transaction takes, that is, the processing process that a transaction undergoes, is described in what is called a block, and each block has several parameters, and by specifying these parameters, the user can Define the function of each block.

Even when writing a complex function such as a conditional two-branch condition that allows exiting from the queue, it is necessary to execute it by manipulating parameters. in this way.

With GPSS, users need to define blocks unrelated to the original model in order to obtain simulation timing (synchronization), and knowledge of the internal mechanisms of GPSS is essential. Furthermore, as can be seen from the above block, creating a model when the simulation target is complex becomes rapidly difficult.

Therefore, a simulator that performs simulation execution control (for example, the above synchronization) by itself, reduces the user's development of the simulation execution control part, and facilitates the user's model creation work.
Alternatively, developing a simulation language is especially necessary when the simulation target is complex.

As such a simulator, IEICE Transactions D, Vol. J70-D, No. 5
"Logical Language Dedicated to Queuing Network Simulation Using R-Time Objects", pp. 896-906. In the above languages, a type of macro instruction called a time object is prepared to express the time relationship between processes, and a model is created by combining these.

This method of using time objects makes it much easier to describe simulation execution control than GPSS.

However, since the sequential processing Prolog is used as a processing system, the processing speed is extremely slow, and parallel processing on multiple processors is essential for practical use. Therefore, multi-processor control such as job distribution becomes even more necessary, and the scale of the processing system becomes larger.

[Problem that the invention seeks to solve]

The above-mentioned conventional technology does not give sufficient consideration to high-speed processing of simulations, and there is a problem in that increasing the speed complicates simulation execution control and increases the size of the simulator.

Furthermore, in network simulations, when the network configuration is complex, it is difficult to understand the network configuration using time object descriptions, making it difficult to use the system as a design support tool for configuration optimization and the like.

An object of the present invention is to realize a simulation apparatus that is easy to control when speeding up processing by parallel processing using multiprocessors and the like, and the scale of the simulator itself is small.

Another objective is to make it easy to understand and change the configuration when simulating a network or the like.

[Means for solving problems]

The above purpose is achieved by making the part that executes the basic functions of the simulation target a child process, and controlling the execution of the simulation target by one parent process controlling the execution of the plurality of child processes. The relationship is established using a central clock managed by the parent process, and the child process is started and monitored by the parent process to ensure that the time relationship is maintained. Therefore, the child process performs parallel processing under the control of the parent process.

Furthermore, by defining the relationship between the child processes in advance by the third process and communicating this to the parent process, the child processes do not need to know the relationship between each other and can operate completely under the control of the parent process. Therefore, even if the number of child processes increases due to the complexity of the simulation target, the scale of the simulator will not increase because the child processes themselves do not control the execution of the simulation. Further, configuration changes can be realized by increasing or decreasing the number of processes or changing the number of startups.

[Effect]

The simulation method of the present invention performs the basic functions of the simulation target by multi-process operation. Since a process is a processing unit that is completely protected from other processes in terms of software processing, its execution is not targeted by other processes and is processed completely independently and in parallel. Also,
By managing how many times each child process is started within one unit of time of the central clock, the parent process protects the time relationship between each child process, so a specific process may be delayed or preceded. This prevents simulation results from becoming inconsistent with the actual situation due to differences between the actual system and the simulation system.

〔Example〕

Hereinafter, one embodiment of the present invention will be explained in order starting from FIG. FIG. 1 is an overall configuration diagram of a simulation processing system of the present invention. The overall structure consists of three main parts. The first
This is the part that defines the configuration to be simulated, the types and internal attributes of the components, such as functions, processing speed, processing constraints, etc., and inputs that information into the configuration definition file. Equivalent to. The second part is the part that executes the simulation, and this part is from 10 to 10 in Figure 1.
It consists of the parts shown in 17 and performs parallel processing using a multi-process function. The third part is a part that analyzes and displays simulation execution results. FIG. 1 shows the configuration when these parts are applied to network simulation. The above three parts have an interface using files so that they can be executed by separate processes (the execution part is further divided into multiple parent and child processes). Therefore, independent parallel processing of these three parts is possible. The multi-process function described above is a basic function possessed by a general-purpose multi-process operating system.

Further, in principle, speeding up by parallel processing by multiple computers such as multiprocessors can be achieved without changing the processing configuration. That is, this is achieved by having each computer perform the above-mentioned process-by-process processing, and by performing an interface between each process by data transfer on a processor-coupled bus.

Next, the simulation execution procedure and simulation operation will be explained using FIGS. 1 and 5.

First, define the network configuration as a simulation target. For example, when evaluating the average delay time, throughput, line usage rate, etc. in a packet switching network according to the amount of traffic entering the network, input the following configuration definition.

The amount of traffic entering the network, the processing capacity of each switch (number of packets processed per second), the length of each queue within the switch, the routing method within each switch, processing delay time, line connection information, transmission speed and bit error rate of each line. .

distance, etc. These are input into the network configuration definition file 4 according to the network configuration definition input menu display 2. Definition files include connection relationship files and switch attribute files.

A plurality of files such as a line attribute file are prepared, and the network configuration definition input process 3 changes and adds each file and guarantees consistency between the information in the files. In addition, a program 5 for changing information into the configuration definition format of the present invention is provided so that configuration definition information can be input from another management program.

When the above configuration definition input is completed, this configuration definition input part 1 is running in an asynchronous process independent of the simulation execution and the output of the results.
An interrupt is made to the simulation execution parent process 11 that is currently executing the simulation, and since there have been changes or additions to the configuration, a request is made to read the changes or additions to the configuration and continue the simulation. At this time, parent process 11
handles interrupts so that the time relationships in simulation execution control are not disturbed by asynchronous interrupts. In other words, the time relationships in packet transfer and communication processing between the child processes that execute the basic functions to be simulated (in this case, packet switching equipment, traffic generating subscriber terminals, etc.) are maintained, and the A process forwards packets or generates traffic in advance so that it does not differ from the temporal context in the actual network. To do this, as shown in Figure 5 (1), when an interrupt occurs, the parent process must keep a record of the occurrence of the interrupt using something like a flag, and once the child process has started one cycle 74, it can be simulated. At time point 75, when one unit of time of the central clock provided to control the execution of the process has elapsed, the actual configuration change process will be performed.

Here, the above-mentioned central clock owned by the parent process is a virtual clock provided for synchronization that maintains the temporal relationship of processing between simulation targets, and its minimum time (1 unit)
The simulation is performed using discrete events that are an integer multiple of .

The configuration change by the parent process is shown in Figure 5 (1).

This will be explained below with reference to FIG. 5(2).

(1) The contents of the network configuration definition file 4 are read 76 and compared 77 with the contents of each file before the configuration change to determine 78 where the configuration has been changed.

(2) Message formation 79 is performed at the configuration change location, and a message is sent 80 to the child process affected by the configuration change.

(3) The child process that receives the configuration change message changes 97 various internal tables. After the change is completed, a termination message is notified 98 to the parent process.

(4) When the parent process finishes sending configuration change messages to related child processes and receives termination messages from all of the child processes, the configuration change is completed 81.

(5) Copy 82 the network configuration definition file in the new configuration to another file and use it as the file before the configuration change when the next configuration change occurs.

(6) The parent process re-executes the simulation 14 (returns to 71) from the time of the central clock where it was interrupted.

Execution, configuration changes, output of execution results, etc. as described above are
This is done by the parent process instructing a specific child process in a message format as shown in FIG. For this purpose, the message includes the command type 100 and destination child process name 10.
Write down 1 etc. As shown in Figures 5 and 2, the child process is normally in a state of waiting for a message from the parent process91, and when it receives an instruction from the parent process, it is in a parallel state independent of other child processes. process,
After the process is finished, the parent is notified of the completion 98 and the process returns to a message waiting state 91. Child processes also increase or decrease themselves due to configuration changes from the parent process. Therefore, when starting a simulation, a parent process is first created, and then a necessary number of child processes are created in accordance with the configuration definition given at that time.

In the actual execution of the simulation, data is exchanged between child processes such as intra-network packet transfer of a packet switch, so the inter-process communication function of the operating system 10 is used for data exchange 17 in FIG. This function is also used in the message communication 16 between the parent process and the child process described above. This feature is typically supported by general purpose operating systems.

Next, as explained in Figure 5.2, the output of the simulation execution results is requested by the user via an interrupt to the parent process during or after the simulation, such as when a configuration change request is made. . In this case as well, the time relationship is maintained, and a result output request message is sent to the actual child process when one unit of time of the central clock has elapsed.

This message is sent to the Kaneko process, each child process outputs the simulation execution result data 24 to the simulation result output file 23 specified by the message 103 in FIG. 6, and a termination notification is sent back to the parent process. When the output of the simulation execution results of the process is completed, the parent process resumes execution from the time of the central clock where it was interrupted. In this way,
By outputting simulation results to a designated file only when required by the user, the number of outputs to files that take a long processing time is reduced and processing speed is increased.

Alternatively, output request messages can be sent from the parent process to the child process at regular intervals. In this way, when a certain amount of simulation results are accumulated in the file,
Simulation result analysis/output process 22 in Figure 1
is started by a user's request (keyboard interrupt) or a parent process's request (program processing), and displays the results on the screen or printer. This simulation result analysis/output process 22 is also processed independently and in parallel with other processes in the execution part such as the other configuration definition input processes 3, parent processes, and child processes.

FIG. 2 is an example of a network configuration, and FIG. 2.1 is a model of a packet switching network. packet is 30 or 3
A child process simulating the traffic generating subscriber terminal of No. 5 generates traffic determined by the configuration definition each time the parent process starts. Packet switch 31, 32, 33
．． 34 internally queues packets transferred from another child process, distributes them to transfer destinations, and transfers them.

The above-mentioned queues of the packet switch are set based on configuration definitions for a child process that simulates the switch, and queue operations are performed within the child process. How many packets does one switch process in one parent process startup?
The difference in processing speed of each exchange is simulated based on how many times a child process simulating the exchange is activated in one hour of the central clock.

FIG. 2 (2) is an example of a queuing model representing a production line in a factory. A transaction is generated by a child process 40 that simulates the amount of material to be shipped, and child processes 41, 42 .
43, it arrives at a child process 44 that simulates the amount of product shipments.

Based on the simulation execution results of each model, Figure 2.1
Figure 2 (2
) allows inventory forecasting and production planning.

FIG. 3 shows an example of a simulation execution screen. As shown in FIG. 1, the network configuration definition input menu 51 serves as the unit of parallel processing of the simulation.
, a screen 52 that displays the network configuration during simulation execution, and a simulation execution result display screen 53 are displayed in a multi-window format that can be used on a workstation or the like. This allows the user to see the parallel processing with his or her own eyes.

Figure 4 shows that the parent process (Figure 1 11) that controls the execution of the simulation is connected to the child process (Figure 1 12.13,
This is an execution control table used when starting 14.15) with a message. There are at most 64 tables for the number of Kaneko processes, and this table is created when a configuration definition is input and at the same time as a child process is generated. In the execution control table,
Name of the child process to which the message is sent (process number assigned by the operating system) 60, Network node type 61 that distinguishes the type of function that the child process simulates, such as a packet switch or subscriber terminal, etc., The startup circuit of each child process is centrally located. The number of executions expressed within one unit of clock time to maintain the relative processing performance and traffic generation amount of the child process 62, and the network node type 6
It consists of node attributes 63 that hold parameters such as the internal processing and functions of the node indicated by 1 and the queue length. The parent process controls the execution of the simulation shown in FIG. 16 by looking at this execution control table and transmitting a message to the child process at 73 in Section 5.1 of FIG.

〔Effect of the invention〕

According to the present invention, even if there is a change in the configuration of the simulation target, the simulation can be easily executed by increasing or decreasing the number of processes that execute the basic functions of the target or by changing the number of times the processes are activated. ,Also,
Since processes can be processed in parallel for any configuration, high-speed execution is possible.

[Brief explanation of the drawing]

Figure 1 is an overall configuration diagram of a simulation processing system according to an embodiment of the present invention, Figure 2 is network configuration examples 1 and 2 in network simulation, Figure 3 is an example of a simulation execution screen in network simulation, and Figure 4 is a network configuration example. ga<2) Day 5 1 shows the execution procedure of the child process, and FIG. 6 is a message configuration diagram between the parent process and the child process.

Claims (1)

[Claims] In a computer system with multi-process functionality,
a processing means consisting of a plurality of first processes for simulating the basic functions of the simulation target and a second process for starting and controlling them;
Execution means for the second process to execute the simulation with discrete events that are an integral multiple of the virtual minimum time (1 unit) on the simulation in order to maintain the mutual temporal relationship of parallel processing in the simulation of the process; a control means for controlling the relative difference in mutual processing capacity of the first processes by the number of times the second process starts the first process; and a change for changing the number of the first processes according to the configuration to be simulated. and communication means for performing data communication between the second process and the first process in order to start and control the first process.