Sunday 2:40 p.m.–3 p.m. in Terrace
An Overview of the Facebook cache
- Audience level:
Facebook leverages in-memory data stores extensively. Even though caching is a conceptually simple service, several problems inherent to our scales make the deployment of our in-memory data store particularly interesting and challenging.
Facebook's real time in-memory data store includes two major services: Memcache, a look aside key-value store, and TAO, a write-through graph aware cache that supports structured queries. Both daemons run on thousands of dedicated servers.
In this talk, I will give an overview of the problems that we face deploying cache services that answer over a billion queries per second while maintaining sub-millisecond response time. I will describe the strategies that we use to mitigate them and since Python is a critical piece of the puzzle, I will highlight some of the areas where we use it.