面向多租户Web应用的性能隔离方法

Abstract

Multi-tenancy enables the sharing of resources and costs across a large pool of users, thus allowing for centralization of infrastructure in locations with lower costs. Performance isolation is a key requirement for multi-tenant Web application hosting environments. This paper proposes a performance isolation approach for application-level sharing multi-tenancy Web applications. A fine-grained application-level resource management model based on the transaction-processing chain is first defined, as well as its concurrency control algorithm based on the semaphore mechanism. This model helps to manage resources usage in a transaction-level and phase-targeted way. It reuses one thread within a transaction process, and therefore, avoids the reengineering cost generated by system reconstruction. The end result is an approach that extends the current Web application hosting environments to multi-tenancy aware hosting environments. A detailed set of experiments were performed under different settings for changing workloads in the controlled TPC-W application to evaluate the effectiveness and performance overhead of our approach. Compared to the traditional coarse-grained strategy, experimental results show that the fine-grained strategy based on our approach is more efficient and agile in avoiding performance interference between multiple hosted tenants and CPU overload. Our results also show that the performance overhead of our approach is small.