Adaptive pull-based policies for wide area data delivery
ACM Transactions on Database Systems (TODS)
On the use of optimal stopping theory for improving cache consistency
WISE'12 Proceedings of the 13th international conference on Web Information Systems Engineering
| Hi-index | 0.00 |
Web caching is an important technique for acceleratingweb applications and reducing the load on the webserver and the network through local cache accesses. Asin the traditional data caching, web caching poses the well-recognizedproblem of maintaining cache consistency. Webcaching, however, has the leeway of delaying the refreshmentof caches when the web server updates the originaldata, i.e., web caching tries to get better performance allowingtolerable inconsistency. This weak consistency requirementintroduced the concept of time-to-live (TTL: thetime during which the cached data item is expected to bevalid) in the face of future updates. Subsequently, a numberof methods have been invented to have the cache server estimatethe TTL. However, the two well-known TTL estimationmethods-the fixed TTL method and the heuristic method-do not allow intuitive understanding of the estimation processesand lack theoretical reasoning behind them, disallowingadministrators from configuring the cache serverby their intention. To mend these deficiencies, we proposethe update-risk based TTL estimation method. This methoduses a formal, yet intuitive, approach based on probabilisticanalysis. In the proposed method, users provide the updaterisk as the probability that the original data will be updatedwithin the estimated TTL. Then, based on our model, thecache server calculates the value of the TTL using the updaterisk. The results of our experiments, performed usingthe logs of a real cache server, show experimentally that themeasured update risk closely matches the one used to estimatethe TTL. Moreover, the notion of the update risk isclear in its intention and semantics. These confirm the superiorityof our method to the conventional ones. We alsoshow the impact of update risk on performance and consistencyin order to help administrators select an appropri-atevalue for update risk to obtain performance and consistencydesired. In addition, we reilluminate the two aforementionedconventional methods in light of our method.