| Article ID | Journal | Published Year | Pages | File Type |
|---|---|---|---|---|
| 424731 | Future Generation Computer Systems | 2011 | 7 Pages |
The development of workflow management systems (WfMSs) for the effective and efficient management of workflows in wide-area infrastructures has received a lot of attention in recent years. Existing WfMSs provide tools that simplify the workflow composition and enactment actions, while they support the execution of complex tasks on remote computational resources usually through calls to web services (WSs). Nowadays, an increasing number of WfMSs employ pipelining during the workflow execution. In this work, we focus on improving the performance of long-running workflows consisting of multiple pipelined calls to remote WSs when the execution takes place in a totally decentralized manner. The novelty of our algorithm lies in the fact that it considers the network heterogeneity, and although the optimization problem becomes more complex, it is capable of finding an optimal solution in a short time. Our proposal is evaluated through a real prototype deployed on PlanetLab, and the experimental results are particularly encouraging.
Research highlights► In this work, we present an algorithm for finding the optimal ordering of pipelined services when the services communicate directly with each other through links characterized by different transmission times. ► The optimization goal is to detect the ordering that minimizes the query response time, which, due to parallelism, depends on the bottleneck service. ► Experiments conducted in a world-scale infrastructure (PlanetLab) provide strong insights into the actual performance benefits of the proposed algorithm. The main conclusion is that, in real cases, our algorithm can lead to considerable performance benefits compared to approaches that do not consider the network heterogeneity.
