Distributed Caching with Memcached
Installing Memcached alone is no panacea; you have to do some work to use it. Profile your application and database queries to see where you're killing the most time and then cache from there. You also have to handle updating and purging your cache, because immediate cache coherency is important for most applications. If your application's internal API is already pretty clean, and you don't haphazardly hit the database all over your code, adding Memcached support should be easy. In your getter functions, simply try Memcached first. On a miss, hit the database and then populate Memcached. In your setter functions, update both the database and Memcached. You may find race conditions and cache coherency problems to deal with, but the Memcached API provides means to deal with them.
Memcached also is useful for storing data you don't really need to put on disk. For example, LiveJournal uses it to prevent accidental duplicate submissions of requests by storing the transaction's result code in Memcached, keyed by a transaction signature. Another example of Memcached as a primary data store, as opposed to a cache, is warding off dumb and/or malicious bots, often spammers. By keeping track of the last times and actions of each IP address and session, our code automatically can detect patterns and notify us of attacks early on, taking automatic action as necessary. Storing this information in the database would've been wasteful, burdening the disks unnecessarily. Putting it in memory is fine, however, because the data is safe to lose if a Memcached node fails.
I asked the mailing list what interesting things they're using Memcached for, and here's what they said:
Many people use it like we do on LiveJournal, as a typical cache for small Web objects.
One site is using it to pass the currently playing song from their Java streaming server to their PHP Web site. They used to use a database for this, but they report hitting Memcached is much nicer.
A lot of people are caching authentication info and session keys.
One person reported speeding up mail servers by caching known good and known bad hosts and authentication details.
I continue to receive interesting e-mails and suggestions, so I'm happy that people are finding good uses for it.
If you can get away with running your threaded application on a single machine and have no use for a global cache, you probably don't need Memcached. Likewise, SysV shared memory may work for you, if you're sitting on a single machine.
A few people have suggested that MySQL 4.x's query cache might negate the need for Memcached. The MySQL query cache is emptied every time a relevant table is updated in any way. It's mostly a feature useful for read-only sites. LiveJournal is incredibly write-heavy, as are most high-traffic sites nowadays. Also, as with other databases, the MySQL caches together can't exceed the maximum address space the kernel provides, usually 3GB on a 32-bit machine, which gets to be cramped.
Another option for some people is MySQL's in-memory table handler. It wasn't attractive for my uses because it's limited to fixed-length records, not allowing BLOB or TEXT columns. The total amount of data you can store in it also is limited, so we still would've needed to run a bunch of them and fan out keys amongst them.
I'd like to thank Anatoly Vorobey for all of his hard work on the Memcached server, Lisa Phillips for putting up with early crash-prone versions and all the users on the mailing list who have sent in patches, questions and suggestions.
Resources for this article: /article/7559.
Brad Fitzpatrick has been hacking database-driven Web sites for eight years. In addition to riding his bike, Brad enjoys trying to think up alternative solutions to problems that otherwise might involve salespeople. Unless you're pitching blue pills or informing him of dead servers, Brad welcomes your mail at firstname.lastname@example.org.
Webinar: 8 Signs You’re Beyond Cron
11am CDT, April 29th
Join Linux Journal and Pat Cameron, Director of Automation Technology at HelpSystems, as they discuss the eight primary advantages of moving beyond cron job scheduling. In this webinar, you’ll learn about integrating cron with an enterprise scheduler.Join us!
- DevOps: Better Than the Sum of Its Parts
- Return of the Mac
- Drupageddon: SQL Injection, Database Abstraction and Hundreds of Thousands of Web Sites
- Play for Me, Jarvis
- Non-Linux FOSS: .NET?
- Not So Dynamic Updates
- Designing Foils with XFLR5
- Users, Permissions and Multitenant Sites
- diff -u: What's New in Kernel Development