JPH1185707A - Selection method/device for job input computer for parallel computer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To consider the reliability of a node anti to schedule a job at the time of selecting the node being the objet of job input from among plural nodes (computers) constituting a parallel computer. SOLUTION: A reliability data registration part 15 collects reliability data of the respective nodes and a node rank registration part 11 divides data into reliability ranks in accordance with the degree of reliability and registers them in a node rank table 12. A job rank decision part 10 decides the ranks of the jobs from the priority of the inputted job and the ranks of the jobs registered in a job rank table 16. A node selection part 13 refers to the node rank table 12 and selects the node having the same reliability rank as the priority rank of the job as the computer inputting the job. A job supply part 14 supplies the job to the selected node.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a method for selecting a computer to which a job is to be submitted in a parallel computer, and more particularly to a method for determining a job submission computer based on the reliability of the computer.

[0002]

2. Description of the Related Art A conventional scheduling method for selecting a computer to which a job is to be submitted from among a plurality of computers constituting a parallel computer is disclosed in, for example, Japanese Patent Laid-Open No. 5-1202.
There is known a method of selecting a computer with the smallest computer load as disclosed in JP-A-43-43.

[0003]

The parallel computer system is composed of computers called a plurality of nodes, each of which has a processor, and performs processing in parallel. The capacity of the parallel computer can be improved by increasing the number of nodes, and nodes can be added relatively easily. Therefore, the number of nodes is increased according to the need for processing capacity. As a result, nodes with an old manufacturing time (particularly, processors) and new nodes may be mixed. The fact that the nodes are manufactured at different times means that the reliability of the nodes is different. Therefore, when scheduling the job submission, it is necessary to consider the reliability of the node.

An object of the present invention is to provide a job scheduling method that takes into account the reliability of a node.

[0005]

SUMMARY OF THE INVENTION According to the present invention, reliability data is collected for each of the computers constituting a parallel computer, classified into one of a plurality of reliability ranks according to the degree of reliability, and information about a job is obtained. Is input, the job is classified into any one of the priority ranks having the same rank number as the reliability rank according to the priority level of the job, and the computer having the same reliability rank as the job priority rank is selected as the computer to which the job is input. It is characterized by a method of selecting a job input computer.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a configuration diagram of a parallel computer system according to the present embodiment. The system includes a node 1 that is a computer that controls the submission of a job, and nodes 31, 32,... 33 that are connected to the node 1 via a transmission line 20 and are a plurality of computers that execute the job. . Here, the node is a computer including a processor and an input / output device connected to the processor, a storage device, a communication control device, and the like. Each of the nodes 31, 32,... 33 is a computer constituting a parallel computer. The node 1 may be one of the computers constituting the parallel computer, or may be another independent computer. The input device 2 is an external storage device that is connected to the node 1 and stores information (JCL) on a job. The storage device of the node 1 stores a node rank table 12 and a job rank table 16. The node rank table 12 includes nodes 31, 32,
.. Is a table for ranking each of the 33 nodes from the viewpoint of reliability. The job rank table 16 is a table for ranking jobs from the viewpoint of priority.
Each program of the node rank registration unit 11, the reliability data registration unit 15, the job rank determination unit 10, the node selection unit 13, and the job submission unit 14 is stored and executed in the main storage device of the node 1. These programs are generally included in a job management program that is a part of an operating system (OS). Reliability data registration unit 15
33 periodically collects reliability data from each of the nodes 31, 32,... 33, and calculates the MTBF (mean failure interval according to JIS standard X0014) of the node from the reliability data.
Alternatively, the failure rate is calculated and registered in the node rank table 12. The node rank registration unit 11 is started by the reliability data registration unit 15, reads out the MTBF or failure rate of each node from the node rank table 12, obtains a node rank, and registers the node rank in the node rank table 12.
The job rank determining unit 10 receives job information from the input device 2 and determines the rank of the job based on the designated priority. If no priority is specified in the job information and the job rank is registered in the job rank table 16, the registered rank is adopted. The node selection unit 13 selects a node having a reliability rank equal to the job priority rank determined with reference to the node rank table 12.
The job submission unit 14 submits a job to the selected node. Each of the nodes 31, 32,...
S, the job management program receives the JCL of the input job and starts executing the job. In the following description, for simplicity, all nodes 31, 3
2,..., 33 have the same processor performance. Each node has the programs required to execute the job,
It is assumed that resources such as a storage device and an input / output device are provided. Further, the program in the node 1 can be stored in a storage medium, read into a main storage device of the node 1 via a driving device connected to the node 1, and executed.

FIG. 2 is a diagram showing a data configuration of the node rank table 12 and the job rank table 16.
The node rank table 12 stores a node identifier, an MTBF, a rank of reliability, and a busy flag for each node. MTBF is the latest MTBF value of the node (or the processor constituting the node). The rank is the rank of the node determined from the MTBF.
C is ranked. When a node is unavailable, it is ranked D. The failure rate of the node may be used as an index of the reliability instead of the MTBF. At that time, the rank is a rank of reliability determined from the failure rate. The busy flag is a flag indicating whether or not the node is being used.

The job rank table 16 stores a job name and a priority rank corresponding to each job.
The rank is ranked as any of AC. Instead of registering the rank of the job, only the job name, user name, or program name may be registered.

FIG. 3 is a flowchart showing the processing flow of the reliability data registration unit 15 and the node rank registration unit 11. The reliability data registration unit 15 and the node rank registration unit 11 are started and executed periodically. The reliability data registration unit 15 inquires the nodes 31, 32,... 33 to collect operation time data of each node (step 41). The operation time is the time during which the processors constituting the nodes operate continuously without any failure, and there is a continuous operation time for each node corresponding to the number of failures. A node in which the reliability data registration unit 15 cannot collect the operation time is regarded as an unusable node. Alternatively, the availability of the node may be checked by separately inquiring. Next, an MTBF is calculated from at least one continuous operation time for each node (step 42) and stored in the column of the node in the node rank table 12 (step 43). If each node records the accumulated operation time and the number of failures instead of the continuous operation time, collect these data, calculate the failure rate (failure probability per unit time) instead of MTBF, and calculate the node rank. Stored in table 12. For a node where the number of failures is 0, for example, an average MTBF or failure rate can be assumed.

Next, if the end of the node rank table 12 has not been reached (NO in step 44), the node rank registration unit 11 selects the next node (step 45) and, based on the report from the reliability data registration unit 15, It is determined whether or not is available (step 46). If the node is available (step 46 YES), the MTB of the node
Rank from F to A to C (Step 4)
7). As a method of ranking, for example, the MTB of the node
Assuming that F is distributed according to a normal distribution, M
There is a method in which the range of the average ± variance of the TBF is set to rank B, the MTBF above this range is set to rank A, and the MTBF below this range is set to rank C. Regarding the failure rate ranking, A to C can be similarly ranked for the reciprocal of the failure rate. Next, the determined rank is stored in the column of the node in the node rank table 12 (step 48). If the current node is not usable (step 46 NO), the rank is set to D and the node rank table 12
(Step 48). Rank nodes with MTBF or failure rate worse than certain criteria
May be ranked. When the end of the node rank table 12 has been reached and processing of all nodes has been completed (step 44 YES), the processing ends.

FIG. 4 is a flowchart showing the flow of the process of the job rank determining section 10. The job rank determination unit 10 determines whether each job, that is, the J
Input CL (job control language). First, the rank of the job priority is set to C (step 51). J of the job
If the CL has a priority designation (step 52 YES),
The rank is changed according to the designated priority (step 54). If the job priority levels described in the JCL do not match the ranks A to C, the job priority is converted to ranks A to C. If the job is a job registered in the job rank table 16 (step 53YE
S), the registered rank is set as the job rank (step 54). Although no priority is specified, if a user name or job name is registered as an important user or job (step 55 YES), rank C is increased by one to B (step 56). If the job information is a program in which the program name designated to be executed is registered (step 57 YES), the rank of the job is increased by one (step 58). The registered program is, for example, a program that cannot be re-executed when the execution is interrupted during the execution of the program.

FIG. 5 is a flowchart showing the processing flow of the node selection unit 13 and the job submission unit 14. The node selecting unit 13 refers to the node rank table 12 to search for a node whose busy flag is off and which has a rank equal to the job rank (step 61). If there is a corresponding node (step 62 YES), the procedure goes to step 64. If there is no corresponding node (step 62N
O), select a node of higher rank (step 6)
3). Next, the busy flag of the selected node in the node rank table 12 is turned on (step 64). The job submission unit 14 submits a job to the selected node (Step 65). If there is no corresponding node of the higher rank, the job submission is stopped. After this, the jobs are
Executed on selected node out of 33. When the node reports the end of the job, the node selecting unit 13 turns off the busy flag of the node in the node rank table 12. Although the unused nodes are selected in the above embodiment, the number of used jobs is counted instead of the in-use flag, and when selecting a node having a rank equal to the job rank, the number of used jobs is minimum and A node having a rank equal to the rank of the job may be selected. In this case, when the number of used jobs has reached the predetermined number and there is no node with the smallest number of used jobs, a node with a higher rank is selected.

[0014]

According to the present invention, the node to which a job is input is determined by matching the reliability rank of each node with the job priority rank. Therefore, the reliability of the job execution environment is determined according to the job priority. Can be secured.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a parallel computer system according to an embodiment.

FIG. 2 is a diagram illustrating a data configuration of a node rank table 12 and a job rank table 16 according to the embodiment.

FIG. 3 is a flowchart illustrating a processing flow of a reliability data registration unit 15 and a node rank registration unit 11 according to the embodiment.

FIG. 4 is a flowchart illustrating a processing procedure of a job rank determination unit according to the embodiment.

FIG. 5 is a flowchart illustrating a processing flow of a node selection unit 13 and a job submission unit 14 according to the embodiment.

[Explanation of symbols]

1: node, 10: job rank determination unit, 11: node rank registration unit, 12: node rank table, 13: node selection unit, 14: job input unit, 15: reliability data registration unit, 16: job rank table, 31, 32, 3
3: Node

Claims (5)

[Claims] 1. A method for selecting a computer to which a job is to be submitted from among a plurality of computers constituting a parallel computer, comprising: collecting reliability data for each of the computers constituting the parallel computer; Divide into any of a plurality of reliability ranks, input information about the job, and classify into any of the priority ranks having the same number of ranks as the reliability rank according to the degree of priority of the job. A method for selecting a job input computer, wherein a computer having the same reliability rank as the rank is selected as a computer for inputting a job. 2. The method according to claim 1, wherein said reliability rank is classified based on a failure rate of each computer. 3. The method for selecting a job input computer according to claim 1, wherein said reliability rank is classified based on an MTBF (mean failure interval) of each computer. 4. A computer for selecting a computer to which a job is to be submitted from among a plurality of computers constituting a parallel computer, collecting reliability data for each of the computers constituting the parallel computer and according to the degree of reliability. Means for classifying into any of a plurality of reliability ranks, means for inputting information about a job, and classifying into any of priority ranks having the same rank number as the reliability rank according to the priority level of the job; Means for selecting a computer having the same reliability rank as the job priority rank as a computer to which a job is to be submitted. 5. A computer program which is embodied on a computer-readable storage medium and selects a computer to which a job is to be submitted from among a plurality of computers constituting a parallel computer, the program comprising the following steps: Including: (a) collecting reliability data for each of the computers constituting the parallel computer, classifying the data into one of a plurality of reliability ranks according to the degree of reliability, and (b) inputting information about the job and According to the degree of priority, the job is classified into any one of the priority ranks having the same rank number as the reliability rank, and (c) a computer having the same reliability rank as the job priority rank is selected as a computer to which a job is input.