Gossip-Based Workload Prediction and Process Model for Composite Workflow Service

In this paper, we propose to predict the workloads of the service components within the composite workflow based on the communication of the queue condition of service nodes. With this information, we actively discard the requests that has high probability to be dropped in the later stages of the workflow. The benefit of our approach is the efficient saving of limited system resource as well as the SLA-satisfied system performance. We present mechanisms for four basic workflow patterns and evaluate our mechanism through simulation. The experiment results show that our mechanism can help to successfully serve more requests than the normal mechanism. In addition, our mechanism can maintain more stable response time than the normal one.