PScheD: Political scheduling on the CRAY T3E

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Volumn 1291, Issue , 1997, Pages 117-138

  Lagerstrom, Richard N ^a   Gipp, Stephan K ^a  

^a HEWLETT PACKARD ENTERPRISE   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

ARTIFICIAL INTELLIGENCE; COMPUTER SCIENCE; COMPUTERS;

HIGH UTILIZATIONS; PARALLEL PROCESSING; PRODUCTION WORK; TIME DEPENDENT;

SCHEDULING;

EID: 84948964129     PISSN: 03029743     EISSN: 16113349     Source Type: Book Series    
DOI: 10.1007/3-540-63574-2_19     Document Type: Conference Paper

References (15)

1. Additional PEs may be present to support operating system needs.
2. The number is determined by the size of the machine and the type of workload.
3. Generally, the command region has no restrictive attributes.
4. Migration is managed by the political scheduler.
5. Batch and interactive job initiators, for example, have different service provider types.
6. An early attempt to globally manage memory draaxlatically clarified this issue.
7. The design assumption is that the machine is to be used more for multi-PE applications than single-PE work.
8. A command region and a single application region.
9. The policy() system call.
10. See Section 5.1.
11. See MUSE, Section 6.
12. The three factors are quite different so typically only one of them has a dominant weight.
13. This could have devastating consequences with large multi-PE applications.
14. Either share-by-UID or share-by-account may be selected. Share-by-UID is assumed here.
15. Not necessarily fastest turn-around, though, since the MUSE factor does not indicate the performance capability of a machine.