Application-level scheduling on distributed heterogeneous networks

TLDR

Heterogeneous networks increasingly host resource‑intensive distributed parallel applications, where task scheduling is critical and often must be performed by the user because resources cannot be centrally controlled. The study aims to enable users to achieve optimal performance by incorporating both application‑specific and dynamic system information into their scheduling decisions. The authors propose principles for application‑level scheduling and present a work‑in‑progress framework, AppLeS, that implements these principles. Applying AppLeS to a distributed 2D Jacobi application on two production heterogeneous platforms demonstrates the approach’s effectiveness through detailed results.

Abstract

Heterogeneous networks are increasingly being used as platforms for resource-intensive distributed parallel applications. A critical contributor to the performance of such applications is the scheduling of constituent application tasks on the network. Since often the distributed resources cannot be brought under the control of a single global scheduler, the application must be scheduled by the user. To obtain the best performance, the user must take into account both application-specific and dynamic system information in developing a schedule which meets his or her performance criteria. In this paper, we define a set of principles underlying application-level scheduling and describe our work-in-progress building AppLeS (application-level scheduling) agents. We illustrate the application-level scheduling approach with a detailed description and results for a distributed 2D Jacobi application on two production heterogeneous platforms.