Web Database Server Online View Of The Dynamic Selection Problem

The rapid development of Internet generates a staggering amount of web content,especially dynamic web content. However, dynamic content has significantly higher resource demands than static web pages (at least one order of magnitude) and creates a huge scalability problem at web servers.Dynamic web caching has been proposed to solve this scalability issue. The biggest probem of employing caching techniques for dynamic web content is the coupling of serving access requests and handling of updates, since an update that invalidates a cached object will result in the object being recomputed on the next access request. For example, imagine a cache that can store dynamically generated web pages. During normal operation we get an 80% hit rate (which means that only 20% of the pages will need to be recomputed). If we get a small surge in the update stream, a big percentage of the cached pages could be invalidated, and the hit rate will drop significantly. A sudden drop in hit rate leads to a sudden increase in the average response time and possibly to server saturation. View materialization can solve this problem, since it decouples the serving of access requeste from the handling of the updates.Selecting which views to materialize, the view selection problem, has been studied extensively in the context of data warehouses. However, unlike data warehouses, which are off-line during updates, most web servers maintain their back-end databases online and perform updates concurrently with user accesses. Therefore, in the context of the Web, selecting which views to materialize must be decided dynamically and needs to consider both performance and data freshness.In this paper, we present OVDS(θ), an adaptive algorithm for the Online View Selection problem. OVDS(θ) acts as a knob in the system, determining at runtime which views should be materialized (cached and refreshed immediately on updates) and which ones should just be cached and re-used as necessary. Parameter θ corresponds to the level of data freshness that is considered acceptable for the current application.