Method of executing a job in a computer system, a resource manager and a high performance computer system

A resource manager managing computing resources in a computer system to execute computing jobs submitted for execution, each job comprising one or more executable applications, wherein the resource manager comprises:

• a memory and a processor implementing:
• a control module arranged to assign jobs to resources and to control operation of the computer system to produce different hardware configurations thereof, the control module also being arranged to identify one or more representative samples of the job, the one or more representative samples being a part of the job and to allow execution of a job in two phases, an experimental phase in which the job is partially executed and a completion phase in which a remainder of the job is executed each representative sample is executed with each different allowed hardware configuration of the computer system;
• an energy optimizer arranged to identify one or more representative samples of the job, the one or more representative samples being a part of the job, the energy optimizer also being arranged to store the different hardware configurations of the computer system, and to request the control module to assign the job to resources in the computer system to allow execution of the one or more representative samples in the experimental phase, the one or more representative samples being executed each with a different hardware configuration of the computer system;
• and
• a monitoring module arranged to collect execution data including speed of execution and energy used in each hardware configuration during execution of the one or more representative samples,
• wherein the energy optimizer is arranged to select a suitable hardware configuration for the job taking speed of execution and energy use into account, and to request the control module to execute a remainder of the submitted job in the completion phase with the suitable hardware configuration.

A high performance computer (HPC) system comprising nodes coupled together by an interconnect to form a cluster and a resource manager managing computing resources in a computer system to execute computing jobs submitted for execution, each job comprising one or more executable applications wherein the resource manager comprises:

• a memory and a processor implementing:
• a control module arranged to assign jobs to resources and to control operation of the computer system to produce different hardware configurations thereof, the control module also being arranged to identify one or more representative samples of the job, the one or more representative samples being a part of the job and to allow execution of a job in two phases, an experimental phase in which the job is partially executed and a completion phase in which a remainder of the job is executed;
• each representative sample is executed with each different allowed hardware configuration of the computer system
• an energy optimizer arranged to identify one or more representative samples of the job, the one or more representative samples being a part of the job, the energy optimizer also being arranged to store the different hardware configurations of the computer system, and to request the control module to assign the job to resources in the computer system to allow execution of the one or more representative samples in the experimental phase, the one or more representative samples being executed each with a different hardware configuration of the computer system;
• and
• a monitoring module arranged to collect execution data including speed of execution and energy used in each hardware configuration during execution of the one or more representative samples;
• and
• wherein the energy optimizer is arranged to select a suitable hardware configuration for the job taking speed of execution and energy use into account and to request the control module to execute a remainder of the submitted job in the completion phase with the suitable hardware configuration.