JPS62189548A - Load distribution control mechanism - Google Patents

Abstract

PURPOSE:To attain efficient system operation by specifying identification information on a computer center site instead of processor identification information comprising a rough joint multiprocessor system at a corresponding computer center site in case the terminal of one computer site utilizes another computer center site. CONSTITUTION:If a local processor 1 is selected, a regional processor 2 reports the effect to the local processor 1 through a line l6. The local processor 1 can be connected to a terminal 50 through lines l8, l7, l5, l2 and l1. In that case an application program APi operates under a time sharing system control program. In case the terminal of one computer center site used the other computer site, it merely specifies identification information on the other computer center site instead of the processor indentification information comprising the rough joint mltiprocessor system at the corresponding computer center site.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a multiprocessor system, and in particular, when each processor is configured in a loosely coupled state, a terminal can be operated without specifying a group of multiple processors. The present invention relates to a load distribution method suitable for smoothly processing connection requests from .

[Conventional technology]

Various methods have been proposed and put into practical use to improve the processing performance of electronic computer systems by combining multiple processors. Typical configuration examples include (1) a tightly coupled multiprocessor system in which memory is shared between processors, and (2) a loosely coupled multiprocessor system in which each processor has its own main memory. The system is generally well known.

In the above two typical multiprocessor system configurations, the tightly coupled multiprocessor system (1) is effective in improving the processing performance of real-time processing (known as online real-time processing). The loosely coupled multiprocessor system (2) is effective in improving batch processing performance, that is, improving throughput.

Recently, from the viewpoint of improving the reliability of electronic computer systems, the loosely coupled multiprocessor system configuration (2) is becoming popular. This is a loosely coupled multiprocessor
In the case of a multiprocessor system, even if one of the processors in a group of processors that makes up the system fails and the processor stops, the processor is not allowed to perform work (job execution). This is probably because if it is controlled so that it does not occur, the service of the entire computer system will not stop.

By the way, with the remarkable development of the use of computer systems in recent years, in addition to batch processing, time sharing systems (Time Sharing Systems) have become popular.
There is a high demand for online processing such as (abbreviated as TSS). Therefore, under a loosely coupled multiprocessor system configuration aimed at improving batch processing performance, TSS
This means that there is a need to improve the performance of online processing, typified by processing.

This results in realizing a control method that balances the loads applied to the plurality of processors in a multiprocessor configuration as much as possible.

One of the conventional techniques related to a load distribution method for a multiprocessor system in online processing such as TSS processing is described in Japanese Patent Laid-Open No. 157778/1983. The load distribution method disclosed in this publication can only be realized in the configuration of a tightly coupled multiprocessor system, and is not disclosed as a control method for a loosely coupled multiprocessor system.

Furthermore, the control method according to the above-mentioned publication can be realized by providing a common memory between each processor.

On the other hand, with the development of computer networks, it has become natural to use a computer system in another region from a terminal of a computer system in a remote location.

Figure 1 shows three computer centers, namely site A (S
This shows a network configuration in which ITE A), Site B, and SITE C are connected to each other by a line.

When using the computer system of site A from the terminal 50 of the site, the user of the terminal 50 must know the configuration of the computer system of site A and declare which processor it will be used under. There is a need to. Specifically, TSS is used under processor 2, and application program j (Application Pr
ogrdm j :hereinafter abbreviated as Apj), LOGON TSS2 (user identifier>/<
password> data stream.

That is, the above key word "TSS2" specifies the TSS control program running under processor 2.

[Problem that the invention seeks to solve]

According to such a method, even in a computer system that constitutes a loosely coupled multiprocessor system, the problem that the load is biased on one processor depending on the method of specifying end users is solved. arise. Furthermore, it is desirable for non-users to be able to use the computer system without being aware of the processor configurations of other computing centers, and it is also desirable to be able to use the computer system within the same site without being aware of the processors.

An object of the present invention is to provide a means for appropriately distributing processor loads and automatically selecting an appropriate processor without specifying a specific processor between the plurality of processors in a loosely coupled multiprocessor system configuration. It's about doing.

[Means for solving problems]

The load distribution method of the loosely coupled multiprocessor system of the present invention includes, in the main memory unique to each processor, an indicator indicating the acceptance state of the processor, an indicator indicating the load state of the processor, and an indicator connectable to the processor. The device includes means for storing the upper limit of the number of terminals and the number of currently connected terminals.

Furthermore, means is provided for storing a corresponding pair of terminal identification information and processor identification information corresponding to the terminal.

In addition, a supervising processor (Global P) in a loosely coupled multiprocessor system
The processor (rocessor) is provided with means for storing all load states constituting the loosely coupled multiprocessor system.

[Effect]

Based on this component, loosely coupled multiprocessor
When using the own calculation center system that constitutes the system, the user specifies the site name of the calculation center without specifying the processor. Each processor within the computing center system corresponding to the site name first examines the acceptance status indicator. If acceptable, processor identification information is obtained from a control table of corresponding pairs of terminal identification information that can be processed by the processor and processor identification information corresponding to the terminal. If this processing processor identification information is equal to the own processor identification information, the process proceeds to the next load distribution control process.

In load distribution control processing, the maximum number of terminals that can be serviced is set based on the constantly monitored load status of the processor. Next, compare this maximum value with the number indicating the number of terminals currently in service, and if the number of terminals in service is smaller than the maximum value above, add 1 to the number of terminals, and then remove the terminal from the terminal. This means that the processor has processed the service request from the terminal. Note that when the processing of the service request from the end terminal is completed, the numerical value indicating the number of terminals in service is reduced by 1, and then the connection relationship with the terminal is severed.

On the other hand, if the processor identification information corresponding to the terminal is not equal to its own processor identification information, or if the number of terminals in service is equal to or greater than the upper limit of the number of connectable terminals, the processor The service request cannot be processed and this fact is reported to the supervising processor.

Upon receiving the report that the processor identification information is not equal, the supervising processor issues a command to the processor corresponding to the processor identification information in the processor identification information table to process the service request from the terminal. The processor that receives the command processes the service request from the terminal based on the designation of the supervising processor.

Furthermore, if the supervising processor receives a report from the processor that the number of terminals that can be connected is equal to or greater than the upper limit, or a report that is "unacceptable", the supervising processor will store the information in its own processor. Check the load status indicators for each processor.

Select the least loaded processor. A command is issued to the selected processor to process the service request from the terminal.

The processor that receives the command processes the service request from the terminal based on the command from the supervising processor.

As a result of the above, it is possible to properly distribute the load from the terminals of other computing center systems among the processor groups that make up the loosely coupled multi-processor system of the own computing center, and it is possible to select an appropriate processor without specifying a specific processor. .

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to the drawings.

Figure 1 shows an example in which calculation centers are connected to each other in a network configuration. In Figure 1, reference numeral 110 is a calculation center site (SITE-A
111 is a calculation center site (denoted as SITE-B) consisting of one processor, and 112 is a calculation center site (SITE-B) consisting of two processors.
ITE-C). Each processor is numbered 1-3. and 11 to 13.

Each processor has a main memory dedicated to the processor, and the main memory contains an operating system, a TSS control program, an application program Aj, and a load status indicator that realizes the load distribution method of the present invention. Terminal identification information/processor identification information, control programs, etc. are stored.

Here, taking the 5ITE-AIIO calculation center as an example, the supervising processor is processor 2. The local processors are processor 1 and processor 3. The mechanism of the load distribution system of the present invention is shown in FIG.
00 and 20 for the supervisory processor.
A load distribution control mechanism 0 and a load distribution control mechanism 100 are incorporated in the local processor.

Here, the difference between the load distribution control mechanism in the central processor and the load distribution control mechanism in the local processors is that the load distribution control mechanism in the central processor has a load distribution control mechanism for local processors as well as a There is a control block that stores the load of each processor in the combined multiprocessor system and a processing program that distributes the load to each processor.

Referring again to FIG. 1, reference numerals 15 to 18 are communication control devices, and within the communication control devices, a network control program (Network Control Program) is executed.
m:NCP) operates. Reference numerals 50 to 59 are terminal groups connected to the computer system of each site, and reference numeral 5 is a terminal group connected to the computer system of each site.
Terminal groups 0 to 55 are terminal groups connected to the communication control device, and 56 to 59 are terminal groups directly connected to each processor.

Communication control device at each site e.g. 5ITE-A 11
0 communication control device 15.16 and 5ITE-C112 communication control device 18. The communication control device 17 of 5ITE-B 111 and the like are connected to all the processors in each site. Furthermore, processors within each calculation center are also connected to each other.

The operation of load distribution between processors will be explained. Here, the terminal 5 of the calculation center 5ITE-C112 shown in FIG.
The operation when using the calculation center 5 ITEA from 0 will be explained by taking as an example.

FIG. 2 shows an input data stream from a terminal when utilizing the load distribution control method of the present invention. In FIG. 2, data 20 in the data street is a keyword for establishing a line. The data 20 uses the load distribution control method of the present invention, instead of specifying the name of the TSS control program that runs under a designated processor in the past,
Indicates that the name of the calculation center is specified. data 2
2 is an argument for the TSS control program, which in this case specifies user identification information and password.

When the data stream shown in FIG. 2 is input from the terminal 50, the control program 100 in the processor 13 examines the data 21 and determines that the work is not handled by the own calculation center site 112. The data stream shown and the terminal identification information of the terminal 50 are sent to the computing center site 5ITE-A 110 through lines Q1 and Q2.

The network control program 19 in the communication control device 16 of the calculation center site 5ITE-A 110 selects a processing processor based on the connection priority between the processor and the processor 3 (the processing procedure is not shown). Assuming that processor 3 is selected, the data stream shown in FIG. 2 is passed to load distribution control mechanism 100 in processor 3 via line Q3. Load distribution control mechanism 1
00 first determines whether or not its own processor can process a connection request from the terminal. If the connection request from the terminal 50 can be processed, the line Q3. A connection can be established between the terminal 50 and the processor 3 via Q2゜kl, and the TS
The application program API operates under the S control program 3.

On the other hand, if the processor 3 is unable to process the connection request from the terminal 50, it reports this to the supervising processor 2 via the line Q4. The load distribution controller 41ff 200 in the central processor 2 examines the load status of each processor making up the loosely coupled multiprocessor system, and selects the processor with the lightest load. As a result, if supervisory processor 2 is selected, line Q5. tIAΩ2. Line Q1
A connection can be established between the terminal 50 and the central processor 2 via the TSS control program 2, and the application program A
Pj works. On the other hand, if local processor 1 is selected, supervisory processor 2 indicates the selection via line Ω6.
to local processor 1 via. Local processor l is connected to IIAQ 8° line Q7. Line Q5. Line Q
2. A connection can be established with the terminal 50 via line Q1. In this case, the application program Ai operates under the TSS control program.

As a result of the above, when using another computing center/site from the terminal of another computing center/site, you can do so without specifying the processor identification information that constitutes the loosely coupled multi-processor system of the corresponding computing center/site. It is only necessary to specify the identification information of the calculation center site, and within the calculation center site, a processor with an appropriately distributed load can be automatically selected and a connection relationship can be established with the terminal that has requested connection. Note that in the example of FIG.
This means that you can connect to 110.

Next, the operation of the load distribution control mechanism of the present invention will be explained in detail with reference to FIG. 3 and subsequent figures.

FIG. 3 shows control blocks used by the load distribution control mechanisms 100 and 200 shown in FIG. 1, and these blocks exist in the main memory of each processor.

Reference numeral 25 in FIG. 3 is a system control block, and an indicator PSTAT indicates whether or not the processor can accept a connection request from a terminal.

Identifier S of the calculation center site to which the processor belongs
YSTEMID, identification indicator PROCI of the processor
D.

Maximum number of terminals that can be connected to this processor TRMMAX
.

The current number of connected terminals TRMNO and address values in the main memory where other control blocks exist are stored.

Reference numeral 26 is an indicator indicating the load state of the processor. As indicators of processor load status, it is possible to use the generally known CPU usage rate, number of input/output operations per unit time, and main memory usage rate, or to use a value that is a mixture of these. It's okay. It is sufficient if the load state of the processor can be expressed as a result.

Reference numeral 27 is a terminal control block, and the control block 2
The first entry number 7 indicates the number ENTNO of corresponding pairs of terminal identification information and corresponding processor identification information. The second entry and subsequent entries indicate terminal identification information TRMNO. From the second entry onward, a corresponding pair of terminal identification information TRMID and processing processor identification information PROCID # is stored.

Reference numeral 28 is a control block LDTBL that stores the load of each processor in the loosely coupled multiprocessor system, the first entry is the load state of the dominant processor, and the second and subsequent entries are the load states of the other processors in the loosely coupled multiprocessor system. (local processor) load status. In addition, the control block LI) TBL 2 B
exists only in the supervising processor (global processor). Further, the control blocks shown in FIG. 3 exist in the main memory of each processor shown in FIG. The load distribution control mechanism 200 shown in FIG. 1 includes the control block LDTBL 28.
The load distribution control mechanism 100 does not have 2B.

In addition, in FIG. 1, when a loosely coupled multiprocessor system is configured at each calculation center site, the load distribution control mechanisms 100 and 200 are shown redundantly, but this does not mean that they have the same function. It shows.

FIG. 4 is a flowchart showing the operation of the load distribution control mechanism of the present invention.

Line Ql from the terminal 50 of another calculation center/site.

@Q2. The operation when a terminal connection request is issued to the local processor 3 of the own calculation center site via Q3 will be described with reference to FIG.

Here, the data stream is as shown in Figure 2.
It is assumed that the data 21 specifies "5ITEA". Therefore, the terminal 50 shown in FIG. 1 represents a request to connect to any processor within the computing center site 110.

Referring to FIG. 4, the load distribution control mechanism 100 of the local processor 3 checks in process 31 whether there is a connection request to its own system. This is the data 2 in Figure 2 and the SYSTE in the system control block 25 shown in Figure 3.
Compare with MID field.

If the result of the comparison is equal, processes 39 to 42 are executed. This process will be explained later.

If the results of the comparison are not equal, it is possible that the data 21 (1) specifies another calculation center/site or (2) directly specifies a processor, so the process 32 and subsequent steps are executed.

In process 32, processor configuration information PROCONFx to PR3CONF in the system control block 25 in FIG.
,.

and data 21 are sequentially compared. As a result, if there is no match, processing 33 is executed to retrieve the previously received data. The stream is sent again to another calculation center site and the process ends.

If there is a match between the processor configuration information and the data 21, it is checked in step 34 whether it is a connection request to the own processor. This process compares the data 21 with the PROCID in the system control block 25 of FIG. If they do not match as a result of the comparison, this means that another processor within the own calculation center site has been designated, and the processing is entrusted to the central processor in step 37. If the comparison results in a match, the processing is performed by the own processor, and connection processing of the own processor is attempted in process 35. In this process 35, load distribution control between processors is performed.

Before explaining the operation of this process 35, processes 39 to 42
Explain the operation.

In process 39, since the terminal identification information of the terminal 50 in FIG. 3 has been sent to this load distribution control mechanism 100, this terminal identification information and the terminal control block RMID BL 2
7 is compared with the terminal identification information TRMID for the number ENTNO. As a result of the comparison, if there is no equality in determination process 40, process 43 is executed. On the other hand, if there is equal terminal identification information, a determination process is executed.

If the determination process 41 determines that the request from the terminal should be processed by the own processor, the process 35 is executed. In step 42, it is determined that the processing is to be performed by another processor within the own system, so this is notified to the supervising processor.

In process 35, a process for load distribution control is executed. First, the processor displays the load status indicator PLD36 in FIG.
From the value ea of system control block 5YSCTL 25
(7) Set the TRMMAX value.

At this time, the value of sa is reported to the overall processor 2.

The supervising processor modifies the value of the corresponding entry in the load display information LDTBL 28. The value from ea to TRMMAX is set by proportional calculation.

Next, the system control block 5YSCT shown in FIG.
TRM in L 25? Compare IAX and TRMNO.

If TRMNO is smaller than TRMMAX, the processor can process the connection request from the terminal 50;
After setting TRMNOR to +1, the corresponding program A P t etc. is activated in the processor. This determination process is the determination process 36 in FIG. 4, and the process 38 is the startup process of the program A Pa. Note that when a program for disconnection processing (for example, logoff processing) from the corresponding terminal is executed, the terminal is logically disconnected and TRMNO is set to -1.

On the other hand, if TRMNO is equal to or larger than TRMMAX, it means that the processor 3 cannot process the process, and in process 43, the process is entrusted to the supervising processor. This is reported to the supervisory processor 2 via line Q4. Note that when processing 37 and processing 42 report to the supervising processor, they are carried out via the 6th line Q4.

When the supervising processor 2 receives the report from the local processor 3 in the state of process 43 in FIG. 4, it searches each entry in the load display information block LDTBL 28 of each processor and selects the processor corresponding to the entry with the smallest value. do. For example, if another local processor 1 is selected, this is reported to local processor 1 via line Q6. As a result, line Q8. Line Q7. Line Q
5. Line Q2. A connection relationship with terminal 50 is established via line Q1.

On the other hand, process 37 in FIG. If a report is received in the state of process 42, the processor to be processed is specified, so the report is immediately sent to that processor. Note that the processor that receives the instruction from the global processor determines whether or not it can perform the processing using the same method as described above. Furthermore, it goes without saying that the processors to be selected also include the supervising processor.

Terminal group 56- connected to communication control device 15-]8
The method of allocating processors to 59 is also the same as described above.

〔Effect of the invention〕

As described above, according to the present invention, when using another computing center site from a terminal of another computing center site, it is possible to specify the processor identification information constituting the loosely coupled multiprocessor system of the corresponding computing center site. Instead, it is sufficient to specify the identification information of the calculation center site, and within the calculation center site, the loads of the plurality of processors in the loosely coupled multiprocessor system are balanced, and the multiprocessor system can be operated efficiently.

[Brief explanation of drawings]

Figure 1 is a diagram illustrating a state in which calculation centers are connected to each other in a network configuration and a state in which the load distribution control mechanism of the present invention is incorporated, and Figure 2 is an example of a data street for a connection request from a terminal. FIG. 3 is a diagram showing the relationship between control blocks used in the load distribution control mechanism, and FIG. 4 is a flow chart showing the operation of the load distribution control mechanism. 100... a-local processor load distribution control mechanism, 200... supervising processor load distribution control mechanism, 2
5... System control block, 26... Load status indicator, 27... Control block storing a corresponding pair of terminal identification information and processor identification information, 28...
A load status indicator for each processor that resides in the main memory of the supervising processor. 1st fast talking 2nd day

Claims (1)

[Claims] 1. A computational sensor site and a loosely coupled multiprocessor system configured as a loosely coupled multiprocessor system in which each processor has a dedicated main memory and a plurality of processor groups are loosely coupled. configuration, tightly coupled multiprocessor system configuration, or single system configuration, when a network configuration is formed with other computing center sites in a loosely coupled multiprocessor system configuration, two processors in each computing center site in a loosely coupled multiprocessor system configuration a first storage means for storing a name of a calculation center site and processing processor identification information; and a second storage means for storing a plurality of pairs of terminal identification information and processing processor identification information; - When using a calculation center site in the system configuration from a terminal, if you specify the calculation center site name without specifying the processing processor, the processor that first receives a connection request from the terminal will be the calculation site in the first storage means. A load characterized by comprising means for comparing a name with a designated calculation center/site name, and means for searching a second storage means and selecting a processor based on terminal identification information of a connection request. Distribution control mechanism. 2. Each processor includes a third storage means for always storing the load state, a means for storing the upper limit of the number of terminals that can be connected to the processor, a means for storing the number of terminals currently connected, and The global processor that centrally manages the combined multiprocessor system is provided with a fourth storage means for storing the load status of all the processor groups constituting the multiprocessor system, and after the processing processor is determined, the processing processor is transferred to the terminal. setting an upper limit value for the number of terminals connectable to the processor based on the third storage means when a service request for program startup is made from the processor;
A control means that compares the number of terminals currently in service, and as a result of the comparison, if the upper limit of the number of terminals is equal to or smaller than the number of terminals in service, it reports to the global processor that the processor cannot process requests from the terminal. means to
Claims characterized in that a control means is provided for searching the fourth storage means in the global processor that has received the report and instructing the selected processor to process a service request for starting a program from the terminal. The load distribution control mechanism according to item 1. 3. Claim No. 3, characterized in that, when the terminal directly specifies a processing processor instead of the name of the calculation center site, means is provided for instructing the processor to process a service request from the terminal. The load distribution control mechanism according to item 1.