Scheduling Parallel Task Graphs on (Almost) Homogeneous Multicluster Platforms

Abstract

Abstract-Applications structured as parallel task graphs exhibit both data and task parallelism, and arise in many domains. Scheduling these applications efficiently on parallel platforms has been a long-standing challenge. In the case of a single homogeneous platform, such as a cluster, results have been obtained both in theory, i.e., guaranteed algorithms, and in practice, i.e., pragmatic heuristics. Due to task parallelism these applications are well suited for execution on distributed platforms that span multiple clusters possibly in multiple institutions. However, the only available results in this context are non-guaranteed heuristics. In this paper we develop a scheduling algorithm, MCGAS, which is applicable to multi-cluster platforms that are almost homogeneous. Such platforms are often found as large subsets of multi-cluster platforms. Our novel contribution is that MCGAS computes task allocations so that a (tunable) performance guarantee is provided. Since a performance guarantee does not necessarily imply good average performance in practice, we also compare MCGAS with a recently proposed non-guaranteed algorithm. Using simulation over a wide range of experimental scenarios, we find that MCGAS leads to better average application makespans than its competitor.Index Terms-ixed parallelism, parallel task graph scheduling, performance guarantee, multi-cluster platform ixed parallelism, parallel task graph scheduling, performance guarantee, multi-cluster platform M ✦

On cluster resource allocation for multiple parallel task graphs

“…Weaknesses in both HCPA (and thus in CPA) and M-HEFT were identified and remedied in [22], which performs a thorough comparison of both improved algorithms. In the case of a homogeneous multi-cluster platform, an algorithm to schedule a single PTG was recently proposed in [13]. This algorithm is based on a resource allocation algorithm that provides a performance guarantee, but has high computational complexity.…”

A Bi-criteria Algorithm for Scheduling Parallel Task Graphs on Clusters

“…This result was improved in [11], leading to a ∼4.73 performance ratio in the general case. The algorithm proposed in [10] was implemented and compared to HCPA in [12]. It was shown that non-guaranteed algorithms were competitive with the guaranteed one on the average but with tremendously shorter scheduling times.…”

“…Note that this inner loop actually corresponds to an interval of iterations of the seminal allocation procedure of CPA, as shown in Figure 1. Each time T CP ≤ T ′ A , the current allocation is stored for each task (lines [11][12][13]. At the end of this procedure, P different allocations are associated with each task in the PTG.…”

Scheduling Workflows in Multi-cluster Environments

“…Research in this area has not considered such parallel tasks as part of a workflow containing task dependencies. For workflow scheduling, existing solutions restrict the execution of parallel tasks to a single cluster [8] [3]. This is due to the support of high interconnection speed, high bandwidth and low latency networks such as Infiniband [9], and the support of Network File System (NFS) which provides easy and transparent access to data from any of the parallel process.…”