Getting Priorities Straight: Improving Linux Support for Database I/O

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

Christoffer Hall, Philippe Bonnet

The Linux 2.6 kernel supports asynchronous I/O as a result of propositions from the database industry. This is a positive evolution but is it a panacea? In the context of the Badger project, a collaboration between MySQL AB and University of Copenhagen, we evaluate how MySQL/InnoDB can best take advantage of Linux asynchronous I/O and how Linux can help MySQL/InnoDB best take advantage of the underlying I/O bandwidth. This is a crucial problem for
the increasing number of MySQL servers deployed for very large database applications. In this paper, we rst show that the conservative I/O submission policy used by InnoDB (as well as Oracle 9.2) leads to an under-utilization of the available I/O bandwidth. We then show that introducing prioritized asynchronous I/O in Linux will allow MySQL/InnoDB and the other Linux databases to fully utilize the available I/O bandwith using a more aggressive I/O submission policy.
Original languageEnglish
Title of host publicationProceedings of the 31st International Conference on Very Large Data Bases
Publication date2005
Pages1116-1127
Publication statusPublished - 2005
EventVLDB 2005 - Trondheim, Norway
Duration: 29 Nov 2010 → …
Conference number: 31

Conference

ConferenceVLDB 2005
Nummer31
LandNorway
ByTrondheim
Periode29/11/2010 → …

ID: 3184948