A workflow-based dynamic scheduling approach for Web services platforms

摘要

Several techniques have been proposed to deal with task assignment in workflow systems. An "appropriate" service provider is selected using specific details about tasks (such as task type and price). The matching of the specific requirements of a task is performed and the task is then assigned to an appropriate provider. These solutions work well, however they do not produce good performance in terms of the mean flow time and the mean slow down. Existing solutions also do not deal with the issues of reliability (i.e. failure of the service provider), server load (i.e. the amount of tasks in the queue) and processing capacity of the service provider. This paper addresses these limitations by assigning tasks to the solution provider with lower load, higher processing capacity and better reliability. We introduce the concept of "delay factor", which is computed for each server to characterise the reliability of a service provider. For a given server with a specific load and processing capacity, we measure the mean flow time of tasks. The dispatcher then assigns tasks to the service provider that has the lower mean flow time and the smallest delay factor. With extensive performance testing, we demonstrate that proposed approach out-performs existing load distribution strategies.