Sem conflito no cache: observações de uma plataforma de comércio eletrônico multi-tenant.

Abstract

Caching is a classic technique for boosting system performance by reducing client-perceived latency and server load. However, carefully designing and configuring the cache is a challenging task. It includes choosing capacity size and eviction policy, among others. This task is more challenging in multi-tenant systems, where each tenant operates independently and exhibits different demands over time. For this reason, effective cache configuration and management require an understanding of workload characteristics, including load levels, patterns of access, and temporal locality. This paper focuses on the workload characterization of a multi-tenant cache of a large ecommerce platform. We find a significant diversity of tenants regarding load patterns, cache usability, and temporal locality. Based on this, we highlight strategies to optimize the management of multi-tenant cache systems by adopting admission policies and dynamic capacity adjustment (scaling). Implementing these strategies within a production environment will constitute a subsequent phase in this investigation.