Enhanced multicore load balancer with core-to-task ratio based trigger level
This work described here then proceeds to evaluate (1) the cost of migration using an industry standard micro-benchmark ‘lmbench’ on a uniquely configured piece of hardware to ensure fair comparison, (2) the number of migrations on a representative operating system used for departmental HPC (dominate...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Thesis |
| 出版: |
2015
|
| 主題: | |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 總結: | This work described here then proceeds to evaluate (1) the cost of migration using an industry standard micro-benchmark ‘lmbench’ on a uniquely configured piece of hardware to ensure fair comparison, (2) the number of migrations on a representative operating system used for departmental HPC (dominated by the GNU Linux operating system) using a GNU Linux Scheduler simulator, and finally (3) proposes, implements and evaluates a dynamic load balancer triggering, based on the number of tasks in the system and the number of processing cores for a most effective utilisation of the processing cores. The benefits is illustrated using synthetic benchmarks and standard HPC benchmarks, the NASA Parallel Benchmarks. In this work, the proposed method achieved up to a 27% improvement. |
|---|