The basic idea is to push the pages that are discarded by the InnoDB buffer pool LRU mechanism (buf0lru) to memcached and before attempting to read a page from disk (buf0rea), an attemp is made to read the page from memcached first. With a proper tuned libmemcached, accessing memcached is much faster than accessing the local disk. The patch to InnoDB is suprizingly small and simple, less than 100 lines are affected. Furthermore, the patch makes MySQL not dependent on memcached.
Typical applications are: 32 bits OS with more than 4 GB of RAM (not uncommon), Improving performance of MySQL using the memory of the passive server in a Heartbeat/DRBD pair, large OLTP servers by helping to reduce the disk load by removing much of the read operations, Datawarehousing/adhoc queries using a farm of memcached servers to scale the memory to TB size. In fact, memcached could help InnoDB in nearly all applications that are disk bound.
Benchmarking results will be presented for the various use cases along with DBT2 results with and without memcached.
Matthew Yonkovit is a 12 year database veteran, who has worked on large performance tuning adventures on most major database platforms for several fortune 500 clients. He currently is a Solution Architect for Sun/MySQL and lives in St. Johns Michigan.
MySQL Consultant for North America, specialized in Heartbeat/DRBD, NDB Cluster and Performance tuning. Former system administrator and DBA for a company specialized in large scale opensource based infrastructure for schoolboards in Canada.
For information on exhibition and sponsorship opportunities at the conference, contact Sharon Cordesse at firstname.lastname@example.org
Download the MySQL Sponsor/Exhibitor Prospectus
For media-related inquiries, contact Maureen Jennings at email@example.com
To stay abreast of conference news and to receive email notification when registration opens, please sign up for the MySQL Conference newsletter.
View a complete list of MySQL contacts.