Check out the WaffleGrid patch (InnoDB,Memcached)

[Arjen Lentz]

Memcached as a L2 Cache for Innodb - The Waffle Grid Project

See http://www.bigdbahead.com/?p=73

[Mark Atwood (fallenpegasus)]

This patch needs two more things to be production ready.

The memcached server specification needs to be a configvar instead of compiled in.

The memcache key needs to have a unique-per-run id encoded into it, so that when mysqld is restarted without restarting the memcached servers, it doesnt try to pick up stale cached data, and get corrupted.

It would be good to insert the block id's being cached into a bloom filter. That way it can check the bloom filter before checking the memcached, and thus the cost of cache misses will go way down.

A nice-to-have would be to add cache hit and cache miss counters, and display them as statistics.

[Arjen Lentz (arjen-lentz)]

http://en.wikipedia.org/wiki/Bloom_filter

[MadMatt (yonkovim)]

Also Form the stats perspective I think we should do something like cahce retrieval timings;) but that's a ways away. Yves just sent me a not that he has the configvar and stats done.

[Yves Trudeau (y-trudeau)]

I like the idea of a per run ID used as a prefix to the key, this would remove many issues I have with flushing the cache at startup. What about a 4 bytes random number?

[Arjen Lentz (arjen-lentz)]

Hi Yves

On 01/12/2008, at 2:37 AM, Yves Trudeau wrote:
> I like the idea of a per run ID used as a prefix to the key, this
> would
> remove many issues I have with flushing the cache at startup. What
> about a 4 bytes random number?

Random is not unique.
We don't want to play roll-the-dice with customer data.

Cheers,
Arjen.
--
Arjen Lentz, Director @ Open Query (http://openquery.com.au)
Training and Expertise for MySQL and related tools

OurDelta: free enhanced builds for MySQL @ http://ourdelta.org