Effects of Job and Task Placement on Parallel Scientific Applications Performance

Abstract

Abstract-this paper studies the influence that task placement may have on the performance of applications, mainly due to the relationship between communication locality and overhead. This impact is studied for torus and fat-tree topologies. A simulation-based performance study is carried out, using traces of applications and application kernels, to measure the time taken to complete one or several concurrent instances of a given workload. As the purpose of the paper is not to offer a miraculous task placement strategy, but to measure the impact that placement have on performance, we selected simple strategies, including random placement. The quantitative results of these experiments show that different workloads present different degrees of responsiveness to placement. Furthermore, both the number of concurrent parallel jobs sharing a machine and the size of its network has a clear impact on the time to complete a given workload. We conclude that the efficient exploitation of a parallel computer requires the utilization of scheduling policies aware of application behavior and network topology.

Effects of Topology-Aware Allocation Policies on Scheduling Performance

“…Regarding the quasi-contiguous allocation, we can appreciate that performance is always good, being only 30-50% higher to that obtained with purely contiguous allocation. These results confirm our expectations: a good allocation strategy can substantially reduce the execution time of a set of applications sharing a parallel computer as stated in [8]. Now we will asses the real cost of contiguity on scheduling.…”

“…However, our work differs from them in several important aspects. Previous research work shows that, depending on the communication pattern of the application, contiguous allocation provides remarkable performance improvements [8]. Therefore, we do not make extensive use of non-contiguity to increase system utilization; instead, we incorporate backfilling scheduling policy into the scheduler.…”

“…In [8], the authors evaluate the possible benefits of contiguity for a collection of parallel applications. These benefits are highly dependent on the communication patterns of the applications.…”

APHiD: Hierarchical Task Placement to Enable a Tapered Fat Tree Topology for Lower Power and Cost in HPC Networks

“…Practically, task placement must rely on heuristics [8]- [11]. Past work confirms the performance and cost benefits of locality-and network-aware placements on a mesh [12], [13], torus [13], [14], fat tree [14], and dragonfly [15]. In the case of a mesh, the worst-to-best ratio of application performance can reach 4× [14].…”

Efficient and Contention-Free Virtualisation of Fat-Trees

“…Recently, Pascual et al [4] studied the effect of job and task placement on application performance and found that maximising processor locality had a significant impact on application performance. The authors also presented a simple allocation strategy which does not consider routing containment.…”