Parallel computer system and method for controlling parallel computer system

• US 9,942,309 B2
• Filed: 05/21/2015
• Issued: 04/10/2018
• Est. Priority Date: 05/22/2014
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Nakashima, Makoto
• Primary Examiner(s): Winder, Patrice L
• Assistant Examiner(s): SOLOMON, RUTH S.
• Application Number:
  US14/718,132
• Publication Number:
  US 20150341424A1
• Time in Patent Office: 1,055 Days
• CPC Class Codes: G06F 11/0715 in a system implementing multitasking multitasking per se G06F9/46
• CPC Class Codes: G06F 11/0751 Error or fault detection not based on redundancy power supply failures G06F1/30; network fault management H04L41/06
• CPC Class Codes: G06F 11/0793 Remedial or corrective actions recovery from an exception in an instruction pipeline G06F9/3861; by retry G06F11/1402; for recovering from a failure of a protocol instance or entity H04L69/40
• CPC Class Codes: G06F 11/302 where the computing system component is a software system
• CPC Class Codes: G06F 2201/865 Monitoring of software
• CPC Class Codes: G06F 2209/5017 Task decomposition
• CPC Class Codes: G06F 9/5061 Partitioning or combining of resources
• CPC Class Codes: G06F 9/5083 Techniques for rebalancing the load in a distributed system
• CPC Class Codes: H04L 67/10 in which an application is distributed across nodes in the network software deployment G06F8/60; multiprogramming arrangements G06F9/46

Abstract

A parallel computer system includes computing nodes to execute a parallel program for generating computing processes that perform computations and monitoring processes and forming a monitoring hierarchical structure, each monitoring process monitors a monitoring process and a computing process arranged immediately lower than the monitoring process, each of the computing nodes operating as the computing process when the computing process is allocated to the computing node and operating as the monitoring process when the monitoring process is allocated to the computing node. Each of the computing nodes allocated to the monitoring process performs processing that changes the hierarchical structure based on a first target value serving as a target value for the total number of subordinate computing processes of the allocated monitoring process and a second target value serving as a target value for the number of monitoring processes and computing processes arranged immediately lower than the allocated monitoring process.

Claims

1. A parallel computer system, comprising:

a plurality of computing nodes capable of executing a parallel program for generating a plurality of computing processes that perform respective predetermined computations and a plurality of monitoring processes,

the plurality of computing processes and the plurality of monitoring processes are allocated to the plurality of computing nodes,

the plurality of computing nodes form a monitoring hierarchical structure having two or more layers, each of the plurality of monitoring processes being capable of monitoring predetermined number of subordinate processes that are in a layer immediately lower than a layer that each of the plurality of monitoring processes exists, the subordinate processes including at least one of a monitoring process among the plurality of monitoring processes and a computing process among the plurality of computing processes,

each node among the plurality of computing nodes operating as each of the plurality of monitoring processes performs processing that changes the monitoring hierarchical structure based on a first target value and a second target value, the first target value serving as a target value for total number of subordinate computing processes of each of the plurality of monitoring processes, the subordinate computing processes being in layers lower than a layer that a corresponding monitoring process among the plurality of monitoring processes exists and the subordinate computing processes connecting to the corresponding processes directly or indirectly, the second value serving as a target value for number of the subordinate processes of each of the plurality of monitoring processes, the second value being equal to the predetermined number, and

the first target value is calculated using a formula of “

total number of computing processes in the monitoring hierarchical structure/(the second target value)ⁿ”

, and the exponent “

n”

in the formula indicates a value of a layer that each of the plurality of monitoring processes exists in the monitoring hierarchical structure and the second target value is constant.

2. The parallel computer system according to claim 1, wherein each node among the plurality of computing nodes operating as each of the plurality of monitoring processes controls generation and disappearance of a monitoring process included in the subordinate processes and rearrangement of a computing process associated with the generation and the disappearance of the monitoring process included in the subordinate processes, as the processing that changes the monitoring hierarchical structure.

3. The parallel computer system according to claim 1, wherein the plurality of computing nodes include a first computing node operating as a first monitoring process among the plurality of monitoring processes, the first monitoring process has the subordinate computing processes exceeding the first target value, and the first computing node determines to change an arrangement position of an excess of a computing process over the first target value so that the excess becomes subordinate processes of a monitoring process that is a high-order process of the first monitoring process.

4. The parallel computer system according to claim 1, wherein the plurality of computing nodes include a first computing node operating as a first monitoring process among the plurality of monitoring processes, and when the number of the first monitoring process has subordinate processes including a computing process, the number of subordinate processes of the first monitoring process exceeds the second target value, and a second monitoring process existing a layer immediately lower than a layer that the first monitoring process exists has subordinate processes of which falls below the second target value, the first computing node determines to change a computing process included in the subordinate processes of the first monitoring process into the subordinate processes of the second monitoring process so that the number of subordinate processes of the second monitoring process does not exceed the second target value.

5. The parallel computer system according to claim 1, wherein the plurality of computing nodes include a first computing node operating as a first monitoring process among the plurality of monitoring processes, and when number of monitoring processes included in subordinate processes of the first monitoring process falls below the second target value and number of subordinate processes of a second monitoring process among the monitoring processes included in the subordinate processes of the first monitoring process exceeds the second target value, the first computing node generates a new monitoring process included in the subordinate processes of the first monitoring process and determines to change a part or all of computing processes included in the subordinate processes of the second monitoring process into subordinate processes of the new monitoring process.

6. The parallel computer system according to claim 1, wherein the plurality of computing nodes include a first computing node operating as a first monitoring process among the plurality of monitoring processes, and when number of subordinate processes of a second monitoring process existing a layer immediately lower than a layer that the first monitoring process exists does not exceed the second target value even after changing all computing processes included in the subordinate processes of the second monitoring process into subordinate processes of the first monitoring process, the first computing node determines to terminate the second monitoring process and changes all of the computing processes of the second monitoring process into the subordinate processes of the first monitoring process.

7. The parallel computer system according to claim 1, wherein the plurality of computing nodes include a first computing node operating as a first monitoring process among the plurality of monitoring processes determines on a value obtained by dividing the first target value by the second target value supplied to the first monitoring process as the second target value of a second monitoring process existing a layer immediately lower than a layer that the first monitoring process exists.

8. The parallel computer system according to claim 1, wherein the plurality of computing nodes include a first computing node operating as a first monitoring process among the plurality of monitoring processes, and when number of monitoring processes included in subordinate processes of the first monitoring process exceeds the second target value, the first computing node determines to terminate at least one of the monitoring processes included in the subordinate processes of the first monitoring process such that the number of monitoring processes included in the subordinate processes of the first monitoring process is not more than the second target value.

9. A non-transitory computer-readable recording medium having stored therein a program for causing a plurality of computing nodes to execute a process comprising:

generating a plurality of computing processes that perform respective predetermined computations and a plurality of monitoring processes, the plurality of computing processes and the plurality of monitoring processes are allocated to the plurality of computing nodes; and

forming a monitoring hierarchical structure having two or more layers, each of the plurality of monitoring processes being capable of monitoring predetermined number of subordinate processes that are in a layer immediately lower than a layer that each of the plurality of monitoring processes exists, the subordinate processes including at least one of a monitoring process among the plurality of monitoring processes and a computing process among the plurality of computing processes,

the program causing each node among the plurality of computing nodes operating as each of the plurality of monitoring processes to execute processing including

performing processing that changes the monitoring hierarchical structure based on a first target value and a second target value, the first target value serving as a target value for total number of subordinate computing processes corresponding to each of the plurality of monitoring processes, the subordinate computing processes being in layers lower than a layer that a corresponding monitoring process among the plurality of monitoring processes exists and connecting to the corresponding monitoring process directly or indirectly, the second value serving as a target value for number of the subordinate processes of each of the plurality of monitoring processes, the second value being equal to the predetermined number,

the first target value is calculated using a formula of “

total number of computing processes in the monitoring hierarchical structure/(the second target value)ⁿ”

, and the exponent “

n”

in the formula indicates a value of the layer that each of the plurality of monitoring processes exists in the monitoring hierarchical structure and the second target value is constant.

10. A method of controlling a parallel computer system including a plurality of computing nodes the method comprising:

the plurality of computing nodes generating a plurality of computing processes that perform respective predetermined computations and a plurality of monitoring processes by executing a parallel program, the plurality of computing processes and the plurality of monitoring processes are allocated to the plurality of computing nodes;

the plurality of computing nodes forming a monitoring hierarchical structure having two or more layers, each of the plurality of monitoring processes being capable of monitoring predetermined number of subordinate processes that are in a layer immediately lower than a layer that each of the plurality of monitoring processes exists, the subordinate processes including at least one of a monitoring process among the plurality of monitoring processes and a computing process among the plurality of computing processes; and

each node among the plurality of computing nodes operating as each of the plurality of monitoring processes performing processing that changes the monitoring hierarchical structure based on a first target value and a second target value, the first target value serving as a target value for total number of subordinate computing processes corresponding to each of the plurality of monitoring processes, the subordinate computing processes being in layers lower than the layer that a corresponding monitoring process among the plurality of monitoring processes exists and the subordinate computing processes connecting to the corresponding monitoring process directly or indirectly, the second value serving as a target value for the number of subordinate processes of each of the plurality of monitoring processes, the second value being equal to the predetermined number,

the first target value is calculated using a formula of “

total number of computing processes in the monitoring hierarchical structure/(the second target value)ⁿ”

, and the exponent “

n”

in the formula indicates a value of a layer that each of the plurality of monitoring processes exists in the monitoring hierarchical structure and the second target value is constant.