Granted Patent B2

US 12,253,900 · App. 18/078,260 · Granted Mar 18, 2025

Inventors: Jean-Olivier Gerphagnon (Les Clayes-sous-Bois, FR); Jean-François Berree (Gieres, FR)

Assignees: BULL SAS; LE COMMISSARIAT À L'ÉNERGIE ATOMIQUE ET AUX ÉNERGIES ALTERNATIVES

G06F1/329

Abstract

Claims (43)

1. A method for optimizing energy consumption of a computing infrastructure comprising an operating system at an operating system level and a plurality of compute nodes, each compute node of the plurality of compute nodes being associated with a plurality of jobs being run or waiting to be run on said each compute node, wherein the method comprises:

receiving an energy consumption reduction request comprising an energy consumption reduction objective;

determining a set of jobs to be suspended from the plurality of jobs associated with said each compute node, the set of jobs being determined based on the energy consumption reduction objective that is received;

sending, to a runtime environment of said each compute node associated with at least one job comprised in the set of jobs that is determined, a suspension request comprising said at least one job;

wherein said runtime environment is at an application level of said each compute node of said plurality of compute nodes;

upon receiving the suspension request, suspending the at least one job comprised in the suspension request that is received, the suspending being carried out via a set of non-native instructions implemented at the runtime environment of the at least one job on said each compute node of the computing infrastructure associated with said at least one job;

wherein said set of non-native instructions are not received by said operating system level of the computing infrastructure, and

wherein said suspending said at least one job is not performed by said operating system;

active waiting by the at least one job that is suspended for a suspension stop message comprising said at least one job that is suspended, the active waiting being carried out via the set of non-native instructions implemented at the runtime environment of the at least one job on said each compute node of the computing infrastructure associated with said at least one job;

wherein said method is performed independent of said operating system.

2. The method according to claim 1 , wherein the set of non-native instructions is based on a Message Passing Interface standard.

3. The method according to claim 1 , wherein each job of said plurality of jobs is represented by one or more processes on one or more compute nodes of the plurality of compute nodes of the computing infrastructure, and the suspending said at least one job comprises synchronously suspending the one or more processes running said at least one job.

4. The method according to claim 1 , wherein said plurality of jobs are distributed across one or more compute nodes and are associated with one and a same application, and the suspending the at least one job comprises suspending all jobs of the plurality of jobs associated with said same application.

5. The method according to claim 1 , further comprising, upon the runtime environment of the at least one job of the each compute node receiving the suspension stop message, stopping the suspending of the at least one job comprised in the suspension stop message that is received, the stopping being performed via the set of non-native instructions implemented at the runtime environment of the at least one job on said each compute node of the computing infrastructure associated with said at least one job.

6. The method according to claim 1 , wherein the energy consumption reduction request further comprises an item of reduction-removal information, the method further comprising sending the suspension stop message to the runtime environment of the at least one job on the each compute node, the sending depending on said item of reduction-removal information.

7. The method according to claim 1 , wherein the energy consumption reduction request further comprises one or more of

reduction date and time information, and

reduction duration information.

8. The method according to claim 1 , wherein each job of the plurality of jobs is associated with at least one item of job information, and the set of jobs is further determined based on the job information associated with said each job.

9. A computing infrastructure comprising an operating system at an operating system level and a plurality of compute nodes, each compute node of the plurality of compute nodes being associated with a plurality of jobs being run or waiting to be run on said each compute node, wherein the computing infrastructure comprises:

a receiver configured to receive an energy consumption reduction request comprising an energy consumption reduction objective;

a processor configured to

determine a set of jobs to be suspended from the plurality of jobs associated with said each compute node, the set of jobs being determined based on the energy consumption reduction objective that is received;

send, to a runtime environment of said each compute node associated with at least one job comprised in the set of jobs that is determined, a suspension request comprising said at least one job;

wherein said runtime environment is at an application level of said each compute node of said plurality of compute nodes;

wherein said each computer node of the plurality of compute nodes are configured to, at the runtime environment

upon receipt of the suspension request, suspend the at least one job comprised in the suspension request that is received;

wherein said suspend is carried out via a set of non-native instructions implemented at the runtime environment of the at least one job on said each compute node of the computing infrastructure associated with said at least one job;

wherein said set of non-native instructions are not received by said operating system level of the computing infrastructure, and

wherein said suspending said at least one job is not performed by said operating system;

actively wait for a suspension stop message comprising a suspended job;

wherein said actively waiting is carried out via the set of non-native instructions implemented at the runtime environment of the at least one job on said each compute node of the computing infrastructure associated with said at least one job;

wherein said receive, said determine, said send, said suspend and said actively wait are performed independent of said operating system.

10. A computer program-product comprising instructions that, when executed on a computer, lead the computer to implement a method for optimizing energy consumption of a computing infrastructure comprising an operating system at an operating system level and a plurality of compute nodes, each compute node of the plurality of compute nodes being associated with a plurality of jobs being run or waiting to be run on said each compute node, wherein the method comprises:

receiving an energy consumption reduction request comprising an energy consumption reduction objective;

determining a set of jobs to be suspended from the plurality of jobs associated with said each compute node, the set of jobs being determined based on the energy consumption reduction objective that is received;

sending, to a runtime environment of said each compute node associated with at least one job comprised in the set of jobs that is determined, a suspension request comprising said at least one job;

wherein said runtime environment is at an application level of said each compute node of said plurality of compute nodes;

upon receiving the suspension request, suspending the at least one job comprised in the suspension request that is received, the suspending being carried out via a set of non-native instructions implemented at the runtime environment of the at least one job on said each compute node of the computing infrastructure associated with said at least one job;

wherein said set of non-native instructions are not received by said operating system level of the computing infrastructure, and

wherein said suspending said at least one job is not performed by said operating system;

active waiting by the at least one job that is suspended for a suspension stop message comprising said at least one job that is suspended, the active waiting being carried out via the set of non-native instructions implemented at the runtime environment of the at least one job on said each compute node of the computing infrastructure associated with said at least one job;

wherein said method is performed independent of said operating system.

Assignments (2)

Priority Claims (1)

Continuity (1)

References Cited (38)