A Comparative Study of Application-level Caching Recommendations at the Method Level

TL;DR

This paper empirically compares two automatic application-level caching recommendation approaches, APLCache and MemoizeIt, across seven web applications to evaluate their effectiveness and provide insights for future improvements.

Contribution

It provides a comparative analysis of existing caching recommendation methods, highlighting their strengths, weaknesses, and application-specific performance, along with lessons for future development.

Findings

Effectiveness varies by application and configuration.

Invalid recommendations impact caching performance.

Insights lead to seven lessons for future approaches.

Abstract

Performance and scalability requirements have a fundamental role in most large-scale software applications. To satisfy such requirements, caching is often used at various levels and infrastructure layers. Application-level caching -- or memoization -- is an increasingly used form of caching within the application boundaries, which consists of storing the results of computations in memory to avoid re-computing them. This is typically manually done by developers, who identify caching opportunities in the code and write additional code to manage the cache content. The task of identifying caching opportunities is a challenge because it requires the analysis of workloads and code locations where it is feasible and beneficial to cache objects. To aid developers in this task, there are approaches that automatically identify cacheable methods. Although such approaches have been individually evaluated, their effectiveness has not been compared. We thus in this paper present an empirical evaluation to compare the method recommendations made by the two existing application-level caching approaches at the method level, namely APLCache and MemoizeIt, using seven open-source web applications. We analyse the recommendations made by each approach as well as the hits, misses and throughput achieved with their valid caching recommendations. Our results show that the effectiveness of both approaches largely depends on the specific application, the presence of invalid recommendations and additional configurations, such as the time-to-live. By inspecting the obtained results, we observed in which cases the recommendations of each approach fail and succeed, which allowed us to derive a set of seven lessons learned that give directions for future approaches to support developers in the adoption of this type of caching.