Implementing a Distributed Execution Service for a Grid Broker

Abstract

Grid middleware such as OurGrid offer solutions for executing parallel tasks on a grid system. In such systems, users submit their applications for executions through a client broker. MyGrid is the client broker used for the OurGrid system; it is in charge of managing task executions that a user has submitted. Although the broker is able to detect task failures and reschedule them, MyGrid itself constitutes a single point of failure from the user perspective. If it fails, all knowledge of task executions is lost. Moreover, MyGrid is also a bottleneck, since hundreds, or even thousands, of executions could potentially be spawned by an application and need to be managed at the same time by a single broker. In this paper we present the design and implementation of a fault-tolerant distributed execution service that allows for load balancing and improves MyGrid performance. A checkpointing mechanism is used to ease the implementation of the service and to further increase system reliability.