Intelligent Dynamic Caching Framework

Abstract

Application-level caches can effectively improve the performance of any I/O bound applications. However, what needs to be cached in the application caches, should be decided by the developers. This holds true for enterprise level application cache implementations such as Redis and Memcached. Deciding what to be cached, would not be a straightforward task as it might require the knowledge of the system or the business domain, system specifications and possible workloads. In this thesis, a lightweight, simple to integrate and Intelligent application-level cache framework is proposed, and with the use of the framework, application logic can be decoupled from the cache logic. The proposed framework can be used in the general case, without limiting to any database centric caches as it mainly considers frequency and data size which are readily available in every application considered. The framework uses a Support Vector Machine (SVM) classifier model to predict what to cache, hence removing much burden from the developers. The framework sits between the application and the cache implementation, and handles the extra processing asynchronously and automatically, avoiding any overhead added by the extra processing used for the caching decision. Since the framework acts as a transparent layer between the application and the underline cache, for existing applications which already use a cache, this caching framework can be integrated seamlessly. The framework is written in Python and uses a Redis cache as underneath cache implementation but can be extended to support any type of cache implementation. The proposed framework was evaluated against uniform workloads and non-uniform workloads, with different cache eviction methods as well as with different time to live values. The experimental results show that the performance of the application can be improved up to 17% with the use of the proposed model when the specified cache size is limited compared to the total size of all the possible cacheable data.