Dynamic Thread Count Adaptation for Multiple Services in SMP Environments

We propose a dynamic mechanism, thread count adaptation, that adjusts the thread counts that are allocated to services for adapting to CPU requirement variations in SMP environments. Our goal is to increase the maximum throughput available on a system that has multiple dynamic content services while meeting different service time criteria for these services in dynamic workloads. Our challenge is to significantly improve response times for dynamic content on a busy well-tuned thread-pool-based system without prioritizing any specific services. Our experiments demonstrate that a prototype using our approach on J2EE middleware quickly (around every 20 ms) adjusted the thread counts for the services and that it improved the average 90th-percentile response times by up to 27% (and 22% on average) for the SPECjAppServer2004 benchmark.