Home // DBKDA 2014, The Sixth International Conference on Advances in Databases, Knowledge, and Data Applications // View article

Cache Management for Aggregates in Columnar In-Memory Databases

Authors:
Stephan Müller
Ralf Diestelkämper
Hasso Plattner

Keywords: Aggregates, Materialized Views, Cache Management, In-Memory Database, Column Store

Abstract:
Modern enterprise applications generate workloads of short-running transactional queries as well as long-running analytical queries. In order to improve the execution time of computationally intensive analytical queries we have introduced an aggregate cache that makes use of the typical main-delta architecture of columnar in-memory databases (IMDBs) to cope with data modifications. In this work, we contribute a cache management system for that aggregate cache, which bases the cache admission and replacement decisions on novel profit metrics. These metrics are tailored to the main-delta architecture of IMDBs. They ensure that expensive aggregates are stored in the cache while light weight query results are rejected. For the profit estimation of a cached aggregate the system also takes into account transactional data modifications triggered by the enterprise application. Along with the profit metrics we introduce an asynchronous cache management algorithm designed for the main-delta architecture as well as the transactional data modifications. We evaluate the cache management system on mixed, transactional and analytical workloads and real customer data.

Pages: 139 to 147

Copyright: Copyright (c) IARIA, 2014

Publication date: April 20, 2014

Published in: conference

ISSN: 2308-4332

ISBN: 978-1-61208-334-6

Location: Chamonix, France

Dates: from April 20, 2014 to April 24, 2014