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

Efficient Aggregate Cache Revalidation in an In-Memory Column Store

Authors:
Stephan Müller
Lars Butzmann
Hasso Plattner

Keywords: Aggregation, Materialized View Maintenance, In-Memory Data Bases, Bit Vector

Abstract:
Modern enterprise applications do not separate between online transactional processing and online analytical processing anymore. To ensure fast response times for expensive analytical queries, we implemented an aggregate cache in an columnar in-memory database. The separation into read-optimized main and write-optimized delta storage is exploited to cache only the aggregates based on the main storage and aggregate all records in the delta storage on-the-fly. This works with insert-only workloads, but not with deletes and updates that invalidate records in the main storage and consequently invalidate a cached aggregate. In this paper, we introduce an approach to revalidate a cached aggregate using efficient bit vector operations. A revalidation is superior to an invalidation since the old cached aggregate is reused. A further contribution is an evaluation of the influence factors that determine whether to invalidate or revalidate a cached aggregate. Our implementation shows that an aggregate cache revalidation outperforms an invalidation when less than 50% of the relevant records are invalidated.

Pages: 66 to 73

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