Buffer Cache De-duplication for Query Dispatch in Replicated Databases

Abstract

We propose a buffer cache de-duplication technique for query dispatch in replicated databases. In the field of replicated databases, there is the well-known problem called ’Buffer Cache Duplication’ problem, which means that different buffer caches share some identical data. Unfortunately, existing approaches of de-duplication have shortcomings; the only SQL statements of queries (e.g. FROM and WHERE clauses) are insufficient to estimate exactly which data the queries reference for duplication-free dispatch. Our approach uses index access patterns to construct a look-up table that allows dispatchers to determine which database it should dispatch a query. We implement a prototype and demonstrate that under a certain condition around 90% of the duplication holds down to 12% in two databases, and it cuts down the referenced data on each buffer cache to approximately 40% in eight databases. Finally, we will discuss whether the condition can be applied to actual workloads.