An adaptive admission control policy for geographically distributed web systems
Proceedings of the 2nd international conference on Scalable information systems
| Hi-index | 0.00 |
Web servers nowadays have to cope with unprecedented amounts of workload, due to increasing popularity and complexity; in particular, dynamically generated content becomes the standard, hence the term Web application. Providing enough resources to sustain these workloads is a grand challenge, thus the application's runtime environment is often trusted to third-party hosting providers. To optimize resource utilization, hosting providers tend to share server machines between several applications. In such context, it is desirable to make sure that hosted applications are guaranteed a predictable level of performance, in other words implement service-level agreements (SLAs).In this paper, we present an approach to Web application SLA, based on profiling and geared at black-box server components, which allows us to express SLAs using application-level metrics, such as request rates. We conducted experiments in the context of a two-tiers server architecture shared by two e-commerce applications, and were able to enforce performance guarantees expressed as a number of concurrent user sessions.