Towards automatic adaptation of resource requirements to actual workload for task-based programming models
2012 (English)Conference paper (Refereed)
Given the considerate amounts of research on task-based parallel programming models for maximizing performance, this project tries to go one step further and investigate mul- tiprogrammed many-core systems. We provide a mathemat- ical model for a parallel runtime that automatically adapts the need for resources to the workload actually produced. The programming model developed is based on other well established models like Cilk++ and Wool.
The paper concludes with two conditions that guarantee over-utilization and under-utilization of of the existing work- ers without having to capture eciency metrics periodically.
The work presented here is the rst step towards a com- plete framework for the system-wide scheduling and load balancing of multiprogrammed many-core system, assuming a variety of workload types and guaranteeing at least average execution for each running program.
Place, publisher, year, edition, pages
Scheduling, parallel programming, multicore, manycore
Research subject SRA - ICT
IdentifiersURN: urn:nbn:se:kth:diva-107671OAI: oai:DiVA.org:kth-107671DiVA: diva2:577058
MCC-2012. Fifth Swedish Workshop on Multicore Computing. KTH University. Stockholm, Sweden. November 22-23, 2012
FunderSwedish e‐Science Research Center
QC 201301142013-01-142012-12-142013-09-10Bibliographically approved