In-memory databases: an industry perspective

Abstract

Main memory databases store all or part of the data in memory and subsequently reduce or eliminate disk access. This paper presents an architectural study of memory databases with respect to cost/performance trade-offs and applicability issues. Potential performance advantages depend primarily on whether the system is used as an online transaction processing or a decision support system. A major constraint to the storing of data in main memory is the vulnerability of its content to software errors. This fact may force a two level architecture even if main memories cost is considerably reduced.<>