Distributed and dynamic self‐scheduling of parallel MPI Grid applications

Abstract

SUMMARYThe execution of distributed applications on the Grid is already a reality. However, as both the number of applications grow and Grids increase in scale, the efficient utilization of the available but shared heterogeneous resources will become increasingly essential to the Grid's successful maturity. Furthermore, it is unclear whether existing Grid management systems are capable of meeting this challenge. The EasyGrid middleware is a hierarchically distributed application management system (AMS) that is embedded into MPI applications to autonomously orchestrate their execution efficiently in computational Grids. The overhead of employing a distinct AMS to make each application system aware brings at least two benefits. First, the adopted policies can be tailored to the specific needs of each application, leading to improved performance. Second, distributing the management effort among executing applications makes Grid management more scalable. This article focuses on scheduling policies of an AMS for a particular class of application, describing a low intrusion implementation of a hybrid scheduling strategy designed to elicit good performance even in dynamic environments such as Grids. Using application-specific scheduling policies, near-optimal runtimes highlight the advantages of self-scheduling when executing one or more system aware applications on a Grid.

An SOA approach to high performance scientific computing: Early experiences

“…It allows applications to be autonomous, which is not a property of pure MPI applications with the help of a Resource Management System (RMS) [16] who implements self-scheduling policies. EasyGrid has given positive results according to [17] and [15]. However, it is not clear that EasyGrid addresses the problems associated with task granularity and high communication costs expected in scientific applications.…”

A strategy for scheduling tightly coupled parallel applications on clusters

“…Building task graphs expressing precedence constraints for complex parallel applications is non-trivial and hence cannot be adopted for a large number of parallel applications. Some efforts use performance models that the predict execution times for parallel applications [4,23,31]. These performance models are used by optimization algorithms that search the space of candidate schedules and choose the schedule with the minimum predicted execution time.…”

Process Rescheduling in High Performance Computing Environments

“…Certainly, this idea is time consuming and new applications and resources require a new series of tests. After developing the first version of MigBSP, we have observed the promotion of intelligent scheduling systems which adjust their parameters on the fly and hide intrinsic complexity and optimization decisions from users (Ding et al, 2009;Nascimento et al, 2007;Sanjay & Vadhiyar, 2009). In this context, we developed a new heuristic named AutoMig that selects one or more candidates for migration automatically.…”