Runtime performance management of multi-tier systems under simultaneous multithreading

Abstract

Recent years have witnessed the deployment of enterprise applications on modern pro­cessors that employ technologies such as Simultaneous Multi-threading (SMT) to exploit the parallelism in such applications. By studying a TPC-W testbed we observe that the performance of SMT for a multi-tier application strongly depends on the workload in execution in the system, thus prompting the need for smart management policies to decide when to enable or disable it. Towards this objective, we propose the Runtime Performance Optimizer (RPO), a module for the Apache \Neb server that automates SMT activation and deactivation at runtime. Experimental results obtained for a variety of workload scenarios indicate that an RPO-enabled system can gain up to 40% in transaction response times compared to the best static SMT selection policy, at the expense of a small overhead of 0.6% per core utilization. To the best of our knowledge, this is the first study that investigates multi-tier server dynamic management issues arising from SMT. 'vVe believe that the methodology followed as well as the results from the specific system we consider would be of interest to others interested in managing enterprise applications.