Evaluation of cache usage on ecommerce catalogs of multitenant platforms.

Abstract

A common requirement for ecommerce platforms is the management of product catalogs. On large, multitenant, catalogs, the managing latencies is a challenge, and caching is usually employed to enhance performance. By observing a catalog solution of a large ecommerce company, we identified what engineers thought was a sub-optimal cache system. Based on previous experience, the engineers of the company believe that the characteristics of the workload impact cache performance. This work aims to verify this belief. We use traces collected from production incoming requests to analyze and understand the workload characteristics, as well as relate them with the cache behavior.