JPS60134370A - Distributed processing system - Google Patents

Abstract

PURPOSE:To decrease the cost and the response time for each proseccing request and to improve the processing efficiency of the whole distribution system by allotting the request to the computer which minimizes either of respective parameter values or both values. CONSTITUTION:A retriever 19 reads the processing requirement contents of an analyzer 17, and transfers the discrimination code of a computer 8i to a waiting time storage device 13. The waiting time storage device 13 reads the waiting time predicted value, and transfers it to an optimum computer decision device 20. The retriever 19 transfers the discrimination code of the computer 8i and the estimated value of the computer resource using amount of the processing request for a computer processing cost storage device 27. The storage device 27 calculates the estimated value of the computer processing cost in the case that the processing request is processed by a computer 8i from the above-mentioned information, the processing cost of the computer 8i and the parameter value which are stored in the storage device 27, and transfers the results to the optimum computer decision device 20.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a distributed processing system, and particularly to a distributed processing system that allocates a load to one of the computers in consideration of computer processing cost, communication cost, etc.

[Background of the invention]

Conventionally, in distributed computing centers and the like, the load tends to concentrate on specific distributed processing devices, which has been a problem. Therefore, by connecting multiple computers and terminal devices with communication lines and having them communicate with each other and share processing among them, overall processing efficiency and availability (processing in case of partial failure) can be improved. ability to continue)
A distributed processing system aimed at increasing the performance of the system was established as a subsidiary system.

In this system, for each processing request issued from a terminal device or computer, the name of the program or file necessary for its execution is analyzed, and a computer equipped with these (hereinafter, including intelligent terminals) is Assign processing requests.

Here, there may be several computers equipped with necessary programs and files. When deciding which of these calculation methods to allocate the above processing request to, it will affect not only the response time, processing cost and communication cost for the processing request, but also the processing efficiency and cost of the distributed processing system as a whole. become.

When allocating a processing request to one of a plurality of computers, the applicant proposed a system in 1981 that takes the above conditions into account and decided on the processing waiting time of the Tani computer and the communication delay between each point in the distributed processing system. Distributed processing that periodically collects time-related information and allocates each processing request to the computer that minimizes the sum of processing wait time and communication delay time from among the computers that can be allocated to each processing request. An application was filed for the invention of a system (Utility Application No. 56-2793).

According to this distributed processing system, it is possible to minimize the response time to each processing request, equalize the load among computers, and improve the operating rate.

However, in the above system, only the time delay is taken into consideration, and the costs (costs, fees, etc.) required for communication and processing of the total number of devices are not taken into consideration.

- Despite the fact that many users are concerned only with cost or both cost and time delay,
Conventional systems cannot meet these demands.

[Purpose of the invention]

The purpose of the present invention is to improve such conventional deficiencies,
An object of the present invention is to provide a distributed processing system that can reduce the cost and response time for each processing request, and improve the processing efficiency of the entire distributed system.

[Summary of the invention]

In order to achieve the above object, the distributed processing system of the present invention uses a plurality of computers or terminal devices, processing requests from any of the computers or terminal devices, or files and programs necessary during the execution of the processing requests. and a processing allocation device that determines which computer to allocate a request for. The present invention is characterized by storing the above-mentioned requests, and allocating the above-mentioned requests to the plan where one or both of the above-mentioned respective parameter values are the minimum. Another feature is that the above-mentioned request is assigned to the computer that minimizes the processing cost, communication cost, processing waiting time and communication delay time of each computer as variables.

[Embodiments of the invention]

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

FIG. 1 is a configuration diagram of a distributed processing system of the present invention.

In FIG. 1, 1 is a processing allocation device, 2 is a communication line network, and 3 is a delay time management unit of the communication line network, which includes all computers, terminal devices, and processing allocation device 1 at each point in time.
A parameter value related to communication delay time (for example, delay time per unit data transfer length) is stored and updated. 5 is a transmitting device, and 5 is a receiving device. A communication delay time storage device 6 stores and updates the same information as the delay time management section 3. 8 is a calculator, and any one of the several boards provided is represented by 8(i). A queue length counting device 9 counts the number of processing requests (usually called a job queue or transaction queue) waiting to be processed for each computer 8(i).

Reference numeral 10 denotes a throughput counter, which counts the number of processing cases per unit time (ie, throughput) of the computer 8(i) at each point in time. 11 is a transmitting device attached to each 51n machine 8(1), and 12 is a receiving device. 13
is a latency storage device, and each computer 8 at each point in time
(i) Store the expected value of the waiting time. Reference numeral 15 denotes a terminal device, which includes not only terminals for data communication, but also office computers, small-sized classification processing systems, word processors, small-sized search systems, and the like. 1
A buffer memory 6 temporarily stores processing request information when it is received from the receiving device 22. Reference numeral 17 denotes a processing request content analysis/storage device, which decodes items necessary for determining computer allocation, such as program names and file names necessary for executing each processing request. 18 is a directory, which is a computer that holds each program and each file (in this case,
There may be more than one identification symbol. Reference numeral 19 indicates a search device for searching for assignable computers, and 20 indicates an optimal computer determination device, which determines the computer for which the weighted sum of expected waiting time, communication delay time, computer processing cost, and communication cost in the searched computer is the minimum. Determine. Reference numeral 21 denotes a destination editing device, which adds an identification symbol of a computer designated as a destination to a processing request. 22 is a receiving device, which receives a processing request from the communication line network 2, computer 8 (iL or terminal 15), and transfers it to the buffer 16. 23 is a transmitting device, which receives a processing request to which an identification symbol of the optimum computer has been added. is transmitted to the computer via the communication line network 2. 26 is a communication cost storage device, and all computers 8 and terminal devices 1
5 and the processing allocation device 1 is stored and updated. 27 is a computer processing cost storage device;
Store and update parameter values related to processing costs (e.g., charges per processing step, charges per issue of input/output to external storage device) for all computers.2
δ is an input device for registering or correcting the stored contents of the surface storage devices 26 and 27. Among the above explanations, the newly installed device according to the present invention is the cost-related storage device 26.
．． 27, an input device 28, a storage device 6 related to delay time and transmission, a receiving device 5, a storage device 13 related to waiting time and transmission, receiving devices 11 and 12, and a directory 18 and search device 19 related to computer specification. This is a determination device 20.

Other devices are already installed in conventional distributed processing systems. In particular, the communication line network delay time management section 3
In order to allocate each processing request to a route that minimizes the communication delay time at that time, the communication delay time between each element in the communication line network 2 is recorded in 1. C. It is something to be updated.

Note that an arbitrary number of processing allocation devices f l can be arranged within the distributed processing system, but each terminal device 15
There is a problem as to which processing allocation device 1 (J) a processing request issued from computer 8 (i) should be directed to.

This can be solved by determining the allocation in advance, or by giving an instruction from the issuing operator each time a processing request is issued.

In addition, liberalization of online data communication and data communication,
Alternatively, with the practical application of the INS system, it is possible that high-speed communication and lower communication line costs will be possible, but in that case, the transmitting devices 4 and 11 and the receiving device
12, the communication device 23, and the receiving device 22 can be located at remote locations, and one processing allocation device 1 can be provided in common in distributed processing systems in various places, and it can also be made to function as a composite distributed processing system.

FIG. 2 is an operational flowchart of the system of FIG. 1.

The delay time management unit 3 of the communication line network transfers information about the portion that has been changed within each fixed period to the communication delay time storage device 6 via the transmitting device 4 and the receiving device 5. The storage device 6 receives the transferred information and updates the corresponding portion of the stored report.

Throughput counting device 10 connected to each computer 8(i)
(i) calculates the throughput value Nit in the period t by dividing the number of processing requests completed within an arbitrary period Jt by the period t, and calculates the throughput value Nit in the period t, Report to 1). Queue length counter 9(i)
At the end of the period t, the receptionist of the computer 8(1) counts the number of processing requests waiting in the queue and reports it to the computer 8(1). Calculator 8(i) has both counting devices g(
When both reports of i) and lIi) are received, the total expected waiting time Tit at period t is calculated using the following formula.

Tit "wit ' Nit """"' (1) Next, the calculation 468 (i) calculates the expected waiting time value 'r11 to the transmitted value [11 (1), and stores the waiting time via the receiving device 12 (i). Transfer to device 13.

On the other hand, the system operator can control all the computers 8, terminal devices 15, and processing allocation devices via the input device 28.
Parameter values related to the communication cost unit price and the processing cost for all computers 8 are registered in the communication cost storage device 26 and the computer processing cost storage device 1ffi 27, respectively, and when the registered contents are changed, , correction is made from the input device 28.

Next, operations for each processing request will be explained with reference to FIG. Note that the first number in parentheses in the block of FIG. 2 indicates the symbol of the device in FIG. 1, and indicates the operation of the device 6.

As described above, a processing request issued from any terminal device 15 or computer 6(i) is directed to a specific processing assignment device 1 based on a predetermined association or an operator's finger. The processing allocation device 1 first temporarily stores the arrived processing request in the buffer memory 16. The processing request content analysis device 17 reads the content of the processing request with the highest processing priority from the buffer memory 16, and analyzes the program name, file name, computer resource usage (central processing unit, The number of running steps, the number of inputs and outputs to an external storage device, etc.), the estimated data transfer length, and the identification code of the terminal device 15 or computer 8(1) to which the processing results are to be output are deciphered and stored (
Step 31.32.33).

Here, the processing priority is determined from information regarding the urgency of the processing given in advance to each processing request, the processing time length of the processing request in the buffer memory 16, etc. Next, the analysis device 17 transfers the program name and file name necessary for executing the processing request, and the identification symbol of the destination to the assignable computer search device 19 (
Step 33).

The retrieval device 19 searches the total number of devices 8 with all the program names and file names transferred from the directory 18.
Search for the identification symbol (g) (step 34). If there is no meter that satisfies the above conditions, a message to that effect is sent to the destination editing device 21, which sends the processing request to the terminal device 15 or the computer 8 (i.e. ) is given as the destination, and the transmission value e2
3 (steps 35 and 37). Next, the destination editing device 21 erases the storage contents related to the processing request from the buffer memory 16 and the analysis device 17 (step 60). Also, if the calculator 8 that satisfies the above conditions
If there is only one (1), the retrieval device 19 transfers the identification symbol of the computer 8 (in-season) to the destination editing device 21 (steps 35 and 36). The stored contents of the processing request are retrieved, the identification code of calculation 4-18(i) is assigned as the destination, and the data is sent from the sender (g device 23) (steps 39, 59).

Subsequently, the memory contents related to the processing request are erased from the buffer memory 16 and the analysis device 17 (step 6).
0). If there are multiple calculation blocks 8 that meet the above conditions, calculate the weighted sum of the communication delay time, expected waiting time at the computer, communication cost, and computer processing cost using the following procedure, and calculate this value. The computer 8 with the smallest value is determined (step 35°38.4O-57).

That is, the processing allocation device 1 first sets the value S1t in the optimal computer determination device 2O to an extremely large value of 1, and then sequentially performs the following operations on all the computer groups that satisfy the condition (step 38).

One assignable computer 80 is selected and designated as 8(1) (step 39). The search device 19 reads the contents of the analysis device 17 and stores the waiting time storage device 13 in the computer 8 (i.e.
) is transferred (step 41).

The waiting time storage device 13 reads the expected waiting time value of the total g(fi8(i) and transfers it to the optimal computer determination device 2O (step 45).The search device 19 also stores the computer processing cost storage device 27, the calculator 8(i
) and the estimated amount of computer resource usage (central processing unit, number of running steps, number of inputs/outputs to external storage devices, etc.) of the processing request are transferred (step 42). The storage device 27 stores the above information, the total processing cost of A-[8(i) stored in the storage device 27, and the Harameter value (fee per processing step, per input/output issue to the external storage device). fees, etc.), and send the processing request to that calculator 8.
Estimated value C of computer processing cost when processing in (i)
11 is calculated using, for example, the following equation, and the result is transferred to the @appropriate computer determination device 20 (step 46).

Cp=R-C+H,'・C・It t I t I (2) Here, R1 is the estimated value of the number of processing request 10 running steps, R
1 is the estimated value of the number of input/output issues of the external storage device for the processing request t, Ci is the interest rate per processing step of the computer 80),
C1 is the charge per issue of input/output to the computer 8 (currently an external storage device).

Further, the verification device 19 transfers the computer and destination identification symbols and the estimated value of the data transfer length to the communication delay time storage device 6 (step 43).

The storage device 6 refers to the stored contents and stores information from the processing allocation device 1 to the computer 80) and the computer 8 (
+) to the destination (for example, delay time per unit data transfer length), and then based on these and the transfer information from the search device 19, calculate the communication delay time C1. , for example, calculated using the following formula,
Transfer the result to the determination device 20 (step 47.5
0).

T, 1== vx, i −11+'vi, , ・lj
-=・(3) Here, vl,i is the delay time per unit data length from the device 1 to its computer 8(I), V i
, Dl is the delay time per unit data length from the computer 8(+) to the destination Di, and lt is the device 1 with processing request t.
The data transfer length from to computer processing, lt is the processing request t
This is the data transfer length from computer processing to destination Dt.

Further, the search device 19 transfers the identification code of the computer and destination, and the estimated value of the data transfer length to the communication cost storage device 26 (step 44).

The communication cost storage device 2G refers to the stored contents, and
From the processing allocation device 1 to its computer δ(i), and its total
9. Based on the transfer information, the customer's cost C1t is calculated using the following formula, for example, and the result is sent to the computer 14.
o (step 48.51).・, (: C1t- to 1.l-11+KI, Dt-11...
・(4) Here, K+, i is the b1°3I from device 1
- Communication up to similar 8(1) 3-stroke single 1'lIl8 Ki, n
l is the communication cost unit price from the bI calculator 8(i) to the destination Dt, and for lt and d't, refer to the previous equation (3).

Next, the optimal calculation-like determination device b''j,' 20 is expressed as 6'1
．． Q. Based on the transfer 'tolerance' from devices 6.13 and 26.27, we can calculate the weighted sum S11 of the delay and cost of the processing request t divided into i11 calculation 8(1), for example. A calculation is performed using the following formula (step 52).

S11 = α, °'f'7 (10α2C1t+αB
'I'4 (+α4C71...-(5) Here, α1 to α4 (these are real numbers arbitrarily set in advance by the system operator).

Next, the determination device 2O compares the value of Sit in the memory contents with the new Sit, and if S1t<Sit, replaces *S. with S11, and identifies 1 of the divider 8(i). The symbol is memorized (steps 53 and 54). Also, Si
If t>Sit, the determination device 2o reports to the search device 19 that the operation regarding the computer 8(1) has ended (steps 53 and 55). Next, if there are still computers that can be allocated, the search device 19 selects the next computer by 8 (
I) and repeat the above operation (steps 56.41 to 5).
5). If a total of n machines that can be assigned no longer exist, the identification symbol of the computer source l*) corresponding to Srt is read from the determination device 2O and transferred to the destination editing device 21 (
Step 57). The editing device 21 extracts the stored contents of the processing request from the buffer 16, adds thereto the identification symbol of calculation a8(i*) to which the processing request should be assigned as a destination, and transmits it from the transmitting device 23. (Steps 58.59). Now, the processing in the processing allocation device 1 regarding the processing request t has been completed, and the device l is transferred to the buffer memory 16. The memory contents related to the processing request t are erased from the analysis device 17 (step 60).

As a result, the processing request t sent from the transmitting device 23 is attached with the identification symbol of the computer 6(1) that should process it, so that the processing request t sent out from the transmitting device 23 is specified by the information transfer function provided in the communication line yJ2. Total p: Transferred to the machine.

Next, two modified examples of the present invention will be described.

(i) In the above embodiment, the processing request content analysis device 17
The system is based on a method in which the programmer looks at the contents of each processing request and analyzes the file name and program name necessary for execution. However, there is an FNQQ method other than this method.

That is, with the execution of each processing request, an independent request (this is called a task request) is issued for any necessary files or programs. In this case, the file name or program name required for each processing request is not known in advance, so even if the processing request is assigned to a certain computer, if a task request is issued during execution and the file or program name required for this execution is If the computer does not have one, the execution of the task request must be transferred to another computer.

For this purpose, the above embodiment is modified and operated as follows. First, processing requests are assigned to the total FJ 46 on the assumption that no task requests are issued. next,
If a task request is issued in the process request and cannot be executed by the computer, the computer sends the task request to the buffer memory 16 of the process allocator l. The processing request content analysis device 17 includes a buffer memory 1
In the case of a processing request, the same operation as in the embodiment is performed, and in the case of a task request, the content analysis device 17 extracts the program name and file name necessary for executing the task request. decipher. Next, the search device 19 assigns the task request to a running computer if both the program and the file are present in that computer. If the computer is not equipped, the computer to which it should be allocated is determined in the same manner as in the embodiment, the task request is allocated to it, and the task request is sent. - After the processing of this task request is completed, the processing is returned to the original computer, and subsequent processing requests are executed on the original computer.

(11) This method is almost the same as the above modification (1), but differs only in the last part. That is, modification (1)
At the end, after the processing of the task request is completed on the al" computer to which the requested task is assigned, the subsequent processing is continued on that calculation 4a. This method does not return the processing to the original computer.

In addition, in the method of the present invention, if a failure suddenly occurs in some total h:, the total -i! 4'M reports this to the waiting time storage device 13, and by setting an extremely high value to the expected waiting time value of the corresponding divider in the storage device 13, the other computers that can be assigned, Processing or task requests are automatically assigned.

In addition, when a failure occurs in a part of the communication line network, the delay time management unit 3 of the communication line network reports this to the communication delay time storage device 6, and an extremely large value is stored in the corresponding part of the storage device 6. By setting , assignments to computers that have to go through a faulty location are automatically avoided.

〔Effect of the invention〕

As explained above, according to the present invention, in response to any processing request or task request issued from a computer or terminal device, if there are multiple computers equipped with all the files and programs necessary for executing the request, Among these, the processing request or task request is assigned to the computer that minimizes the waiting time (time), processing cost on the computer, communication delay time, and function value with communication cost as variables. It is possible to reduce the response time and cost of each processing request, and to equalize the load among multiple computers and improve the operating rate.In addition, if a failure occurs in part of computer 2 or the communication line network, automatic Since these points can be avoided, it is also effective in improving the reliability of the distributed processing system.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a distributed processing system showing an embodiment of the present invention, and FIG. 2 is an operational flow chart of FIG. 1: Processing allocation device, 2: Communication line network, 3: Communication line network delay time management section, 4: Transmitting device, 5: Receiving device, 6: Communication delay time storage device, 8: Computer, 9: Attached to computer 8 10: Nt number device attached to calculation PA8, 11: Transmission device attached to calculation) fAa, 12: Receiving device, 13: Latency storage device,
15: Terminal device, 16: Buffer, 17: Processing request content analysis device 11δ: Directory, 19: Allocatable computer search device, 20: Optimal computer determination device, 21: Destination editing device, 22: Receiving device t, 23 : Transmitting device, 26
: Communication cost storage device, 27: Total 4 machine processing cost storage device, 28: Human power device. Patent applicant Masatoshi Isomura, patent attorney representing Hitachi, Ltd.

Claims (3)

[Claims] (1) Determine which computer to allocate processing requests from multiple computers or terminal devices, or requests for files and programs necessary during the execution of the processing requests. The processing allocation device stores one or both of parameter values related to processing cost in each computer and parameter values related to communication cost between each point, and performs processing in response to each of the above requests. A distributed processing system characterized in that the request is assigned to a total of q machines in which one or both of the parameter values are minimum. (2) Determine which computer to allocate processing requests from multiple computers or terminal devices, or requests for files and programs necessary during the execution of the processing requests. The processing allocation device has a processing allocation device that calculates one or both of the parameter values related to the processing cost at each total of n machines, the parameter values related to the communication cost between each point, the processing waiting time of each computer, and the parameter value related to the communication cost between each point. Information regarding communication delay time between points is periodically memorized, and for each of the above requests, the processing cost, communication cost, processing waiting time at each computer, and communication delay time at each computer are set as variables. A distributed processing system characterized in that the above request is assigned to a computer with a minimum function value. (3) The processing allocation device initially assumes that there is no request for files or programs necessary for execution,
After assigning a processing request to a computer, if the above request is issued during execution, the above request will be assigned to another computer equipped with the relevant file or program, and after the processing of the request is completed, the original accrual will be returned. The distributed processing system according to claim 1 or 2, characterized in that the processing request is returned to Hajime, or the processing is continued as is. (The processing allocation device calculates the weighted sum of processing wait time, processing cost, communication delay time, and communication cost in the computer for each processing request or request for an file or program, and selects the one with the smallest value. is compared with an already stored value, and if the stored value is smaller than the stored value, the stored value is replaced with a new value. processing system.