Auto-tune design and evaluation on staged event-driven architecture
Zhanwen Li, David Levy, Shiping Chen, John Žic · 2006
SEDA (Staged Event-Driven Architecture) is a middleware architecture designed to support massive concurrency demands of internet services. However, managing the resources manually to achieve high performance in such a computing system has proved difficult, time-consuming, error-prone and non-QoS-guaranteed. In this paper, we propose an adaptive control approach to automatic resource management and performance control for SEDA-based applications. This approach is based on a combination of a load balancing strategy and feedback auto-tune stages for global optimal performance. In addition, our control algorithms are able to automatically optimize the control parameters at runtime. The design has been built into a SEDA-based web sever and validated by benchmarking this web server. The experimental results demonstrate that our auto-tune design is able to yield superior adaptation performance for SEDA applications in dynamic working environments, achieving desired performance targets with simple control algorithms and automatic parameter tuning.