(Paper #53)
Current shared-memory multiprocessor CC-NUMA architectures provide a global address space to applications by hardware. However, even thought the memory is virtually shared, it is physically distributed. Since memory nodes are distributed across the system, the cost of the memory accesses depends on the distance between nodes. To reduce the impact of a bad initial memory placement, some operating systems offer a dynamic memory migration mechanism. In this paper, we want to demonstrate that memory migration mechanisms are a useful approach, but that their performance depends more on related issues, such as the processor scheduling, than on the mechanism itself. To show that, we evaluate the case of the automatic memory migration mechanism provided by IRIX, in Origin systems. We have evaluated several workloads of OpenMP applications. We focus on the effects of the page migration mechanism on the CPU time consumed, the processor allocation, and the speedup, when applying performance-driven scheduling policies. Results show that, if the scheduler maintains as much as possible the system stable, the automatic memory page migration mechanism provided by IRIX will improve the execution time of the applications. Experiments also show that a good combination of a performance-driven policy and the memory management results in a system which can be automatically self-evaluated and self-configured.
Keywords:
Scheduling Algorithms and Analysis
Scheduling
Operating Systems
Case Study/Experience Report