基于Chunk Folding的多租户数据库缓存管理机制

摘要

Chunk Folding is one of the common storage architectures in SaaS, which employs shared databases and shared architectures to store tenants' data, so as to gain benefits of economies of scale. However, Chunk Folding based on traditional databases, in lack of multi-tenant properties in memory management, leads to performance degradation and consequently violates tenants' SLA. We propose a Self-Adaptive Multi-Tenant Memory Management (SAMTMM) to consistently achieve tenant's SLA requirement while the memory consumption is minimized, which dynamically generates a series of cache replacement units according to the current access model and computes the corresponding I/O yield, and then adopts a greedy algorithm to select the corresponding replacement units for each tenant. The effectiveness of our multi-tenant memory management is evaluated by our experiments.%Chunk Folding是SaaS模式下常用的存储架构之一,通过共享数据库共享架构来存储租户的数据以获取规模经济效益,但基于传统数据库搭建的Chunk Folding共享存储架构,其缓存管理机制缺乏良好的多租户特性,导致数据库性能恶化,租户的SLA得不到保障.为此,提出了基于Chunk Folding的自适应多租户缓存管理机制,该机制以租户的SLA需求作为驱动,依据租户当前访问模式,动态生成缓存单元集并计算缓存单元集的I/O效益,通过贪婪算法来选择缓存单元集,使得租户SLA得到满足的同时最小化缓存资源的消耗.通过实验分析证明了该缓存管理机制的有效性.