公开(公告)号：GB2605922A

公开(公告)日：2022-10-19

申请号：GB202209994

申请日：2020-12-14

Applicant: IBM

Inventor： LIANA FONG ,  SEETHARAMI SEELAM ,  GANESH VENKATARAMAN ,  DEBASHISH SAHA ,  PUNLEUK OUM ,  ARCHIT VERMA ,  PRABHAT REDDY

IPC: G06F9/50 ,  G06N3/08 ,  G06N5/02 ,  G06N20/00

Abstract: A system, program product, and method for supporting elastic execution of a machine learning (ML) workload using application based profiling. A joint profile comprised of both ML application execution and resource usage data is generated. One or more feature(s) and signature(s) from the joint profile are identified, and a ML execution model for ML application execution and resource usage is built. The ML execution model leverages the feature(s) and signature(s) and is applied to provide one or more directives to subsequent application execution. The application of the ML execution model supports and enables the ML execution to elastically allocate and request one or more resources from a resource management component, with the elastic allocation supporting application execution.

公开(公告)号：GB2545068B

公开(公告)日：2020-05-27

申请号：GB201617773

申请日：2016-10-20

Applicant: IBM

Inventor： BULENT ABALI ,  SEETHARAMI SEELAM ,  CHUNG-SHENG LI ,  HUBERTUS FRANKE

IPC: G06F12/0862 ,  G06F12/0882

Abstract: A method for reducing reactivation time of services that includes examining page faults that occur during processing of a service after the service has been inactive to provide a plurality of prefetch groups, and formulating a prefetch decision tree from page fault data in the prefetch groups. Pages from an initial page table for the service following a reactivated service request are then compared with the prefetched pages in the resident memory in accordance with the prefetch decision tree. Pages in the page table that are not included in said prefetched pages are paged in. A process to provide to provide the service is executed using the page table. Executing the process substantially avoids page faults.

公开(公告)号：GB2605922B

公开(公告)日：2023-04-26

申请号：GB202209994

申请日：2020-12-14

Applicant: IBM

Inventor： LIANA FONG ,  SEETHARAMI SEELAM ,  GANESH VENKATARAMAN ,  DEBASHISH SAHA ,  PUNLEUK OUM ,  ARCHIT VERMA ,  PRABHAT REDDY

IPC: G06F9/50 ,  G06N3/08 ,  G06N5/02 ,  G06N20/00

Abstract: Embodiments relate to a system, program product, and method for supporting elastic execution of a machine learning (ML) workload using application based profiling. A joint profile comprised of both ML application execution and resource usage data is generated. One or more feature(s) and signature(s) from the joint profile are identified, and a ML execution model for ML application execution and resource usage is built. The ML execution model leverages the feature(s) and signature(s) and is applied to provide one or more directives to subsequent application execution. The application of the ML execution model supports and enables the ML execution to elastically allocate and request one or more resources from a resource management component, with the elastic allocation supporting application execution.

公开(公告)号：GB2545068A

公开(公告)日：2017-06-07

申请号：GB201617773

申请日：2016-10-20

Applicant: IBM

Inventor： BULENT ABALI ,  SEETHARAMI SEELAM ,  CHUNG-SHENG LI ,  HUBERTUS FRANKE

IPC: G06F12/0862 ,  G06F12/0882

Abstract: A method for reducing reactivation time of services (e.g Cloud Based Services) comprises examining page faults occurring during processing of a service after the service has been inactive including analysing the address that caused the page fault, to provide a plurality of prefetch groups by logging addresses that caused page faults; formulating a prefetch decision tree from page fault data in the prefetch groups, including edges providing a probability that a page will be touched during the service request; where pages from an initial page table for the service following a reactivated service request are then compared with the prefetched pages in the resident memory in accordance with the prefetch decision tree, and pages in the page table that are not included in said prefetched pages are paged in. A process to provide to provide the service is executed using the page table. Executing the process substantially avoids page faults. Preferably, previously paged out pages are inserted in an in parameter/out parameter vector for bulk restore. Examining page faults preferably includes a reactivation time reducing system, that is switched from a non-participating mode to a learning mode by a reactivation event.