Autonomous Delay Regulation for Multi-Threaded Internet Servers

摘要

How to properly assign system resources to satisfy the ServicernLevel Agreement (SLA) is a challenging research problem.rnPrevious approaches include queuing model based feedbackrncontrol strategies with queuing predictor for feed-forward delayrnprediction. However, ignorance of the multi-threaded nature canrninduce large model errors. To compensate this, previous approachesrntypically perform offline identification, thus making thernsystem dependent on a particular workload and a specific Internetrnapplication.rnIn this paper, we propose a novel framework for autonomousrndelay regulation for multi-threaded Internet servers. We formulaterna processor-sharing queuing model for the multi-threadedrnserver architecture to precisely predict service rate for workerrnthreads. In addition, the proposed scheme uses the sleep actuatorrnto properly assign resources based on the calculated servicernrate. We evaluate our techniques experimentally using anrnApache web server test-bed. We demonstrate that the proposedrnstrategy performs better than the previous approaches under arnrealistic workload.