Configuration decision making using simulation-generated data

Abstract

As service-oriented systems grow larger and more complex, so does the challenge of configuring the underlying hardware infrastructure on which their consitituent services are deployed. With more configuration options (virtualized systems, cloud-based systems, etc.), the challenge grows more difficult. Configuring service-oriented systems involves balancing a competing set of priorities and choosing trade-offs to achieve a satisfactory state. To address this problem, we present a simulation-based methodology for supporting administrators in making these decisions by providing them with relevant information obtained using inexpensive simulation-generated data. Our services-aware simulation framework enables the generation of lengthy simulation traces of the system's behavior, characterized by a variety of performance metrics, under different configuration and load conditions. One can design a variety of experiments, tailored to answer specific system-configuration questions, such as, "what is the optimal distribution of services across multiple servers" for example. We relate a general methodology for assisting administrators in balancing trade-offs using our framework and we present results establishing benchmarks for the cost and performance improvements we can expect from run-time configuration adaptation for this application. © 2011 Springer-Verlag.