Method-Based Caching in Multi-tiered Server Applications

Abstract

In recent years, application server technology has become very popular for building complex but mission-critical systems such as Web-based E-Commerce applications. However, the resulting solutions tend to suffer from serious performance and scalability bottlenecks, because of their distributed nature and their various software layers. This paper deals with the problem by presenting an approach about transparently caching results of a service interface's read-only methods on the client side. Cache consistency is provided by a descriptive cache invalidation model which may be specified by an application programmer. As the cache layer is transparent to the server as well as to the client code, it can be integrated with relatively low effort even in systems that have already been implemented. Experimental results show that the approach is very effective in improving a server's response times and its transactional throughput. Roughly speaking, the overhead for cache maintenance is small when compared to the cost for method invocations on the server side. The cache's performance improvements are dominated by the fraction of read method invocations and the cache hit rate. Our experiments are based on a realistic four-tier E-commerce Web site scenario and site user behaviour is emulated in an authentic way. By inserting our cache, the maximum user request throughput of the web application could be more than doubled while its response time (such as perceived by a web client) was kept at a very low level. Moreover, the cache can be smoothly integrated with traditional caching strategies acting on other system tiers (e.g. caching of dynamic Web pages on a Web server). The presented approach as well as the related implementation are not restricted to application server scenarios but may be applied to any kind of interface-based software layers. © Springer-Verlag Berlin Heidelberg 2003.