Autonomous Delay Regulation for Multi-Threaded Internet Servers

How to properly assign system resources to satisfy the Service Level Agreement (SLA) is a challenging research problem. Previous approaches include queuing model based feedback control strategies with queuing predictor for feed-forward delay prediction. However, ignorance of the multi-threaded nature can induce large model errors. To compensate this, previous approaches typically perform offline identification, thus making the system dependent on a particular workload and a specific Internet application. In this paper, we propose a novel framework for autonomous delay regulation for multi-threaded Internet servers. We formulate a processor-sharing queuing model for the multi-threaded server architecture to precisely predict service rate for worker threads. In addition, the proposed scheme uses the sleep actuator to properly assign resources based on the calculated service rate. We evaluate our techniques experimentally using an Apache web server test-bed. We demonstrate that the proposed strategy performs better than the previous approaches under a realistic workload.