Thursday, August 14, 2008

Cache Money

Scalability is a hard question, but a lot of people think that scalability is all about caching. In particular, memcached is the answer for caching. I think we can blame Facebook for this. Everybody knows that Facebook makes heavy use of memcached. Terry says that social graphs are a scalability problem for databases that is solved by memcached, so he is clearly drinking the Kool-Aid. The benefits of caching are obvious, but is memcached really the best/only way?

Earlier this year, eBay won an award from MySQL. This was for application we built that we originall Gem Cache. It is a caching tier that is built on top of MySQL. When the caching tier was designed, memcached was given a lot of consideration, but there were some very good advantages we got out of using MySQL instead.

First off, can MySQL be as fast as memcached? Absolutely. MySQL is aggressive about keeping things in memory, and if everything is in memory, it will be as fast as memcached. You can use MySQL's MEMORY engine, to accomplish this, or you can stick with MyISAM and let MySQL's caching put things in memory for you. Obviously you need to split your database, but we already knew how to do that efficiently. With that in mind, here are the advantages that MySQL offers.

1.) SQL Semantics. You are not limited to just simple "put" and "get." You can do selects and joins, aggregates, etc.
2.) Uniform Data Access. Do you use some kind of ORM? You can leverage this with a MySQL based cache.
3.) Write-through Caching. In a typical memcached setup, updates are still done to the database and this invalidates one or more objects in memcached. With a MySQL based cache, a row in the cache corresponds to a row in the "real" database. So we can write the cache and then asynchronously update the system of record.
4.) Read-through Cache. Similarly, you can always attempt to read from the cache database. If there is a cache miss, you can invisibly read from the real database and add to the cache at the same time.
5.) Replication. MySQL allows for replication of data, so it is easy to add redundancy and fail-over to your cache. Replication can also be useful when you have multiple data centers.
6.) Management. There are lots of great management tools for DBA, operations folks, etc. to use with MySQL.
7.) Cold starts. When your cache is a copy of database rows, it is easy to bootstrap it from your source, since the source and the cache are so similar.
8.) Eviction. Memcached gives you basic expiration, but otherwise you are handling eviction yourself. The caching in MySQL is a more useful LRU policy.

So there, just a few obvious advantages to using MySQL as a cache instead of memcached. Now I know that a lot of folks use MySQL as their "real" database, so it may seem weird to use it as a cache as well. But they are probably (hoepfully) using InnoDB for the "real" DB and that really is a different beast than MySQL with MyISAM or MEMORY tables. And it's not like you have to pay for extra licenses are anything... What are the advantages of memcached over MySQL? The only obvious one to me is if you want to cache things that don't fit in the database, like deep object graphs or HTML fragments, etc.

6 comments:

  1. Anonymous3:31 PM

    Interesting ideas, indeed. It's definitely food for thought. :-)

    ReplyDelete
  2. Anonymous9:19 PM

    I've related to your interesting post in a post I've made about performance.

    ReplyDelete
  3. Wow, one of the best read posts so far.

    ReplyDelete
  4. Claudia3:00 AM

    Wow, this is great! I’m about a couple weeks out from launching a new site and I start my performance caching today and look what pops up in my newsfeed… fantastic. You just saved me hours if not days!


    financial trouble

    ReplyDelete
  5. I wanted to thank you for this excellent read!! I definitely loved every little bit of it.Cheers for the info!!!! & This is the perfect blog for anyone who wants to know about this topic. You know so much its almost hard to argue with you .........
    thanks
    disney vacation home

    ReplyDelete
  6. Thanks for this great inspiration!

    ReplyDelete