Compression algorithms for log-based recovery in main-memory data management

Abstract

With the dramatic increases in performance requirement of computer hardware and decreases in its cost in recent years, the relevant research in main-memory database is becoming more and more popular and has a prosperous future. Log-based recovery, which is one of its most important research directions, is a set of problems accompanied by volatile memory. Its problem of stagnation in memory/CPU resulted from the slow I/O speed of non-volatile storage now needs to be addressed urgently. However, there is no specific platform for log-based recovery research. So the study aims to address this issue. For the specific platform issue, we design and implement a simulation platform called RecoS. RecoS aims at an implementation of recovery sub-system of the main-memory database. It uses cluster substrate to simulate more real data storage and developed interfaces for a variety of recovery strategies. We propose three log compression methods in this paper: (1) the dictionary encoding, (2) the indirectly encoding with no threshold limit and (3) the indirect encoding with a threshold limit. We also adapt ARIES and command logging on the platform, which represents physical and logical logging respectively, focusing on their recovery process and some important details. Regard the recovery platform as the core to investigate the performance of the recovery platform with different load by using different log sets.