Twitter is increasingly relying on services written in either Java or Scala and running on the JVM. While JVM gives us a world-class runtime in terms of operational stability, performance, and manageability, it is still not trivial to achieve the desired performance characteristics of the programs being run on it, especially when you're dealing with services that need lots of memory, or have very low latency requirements, or both. In this talk, we'll present examples of performance problems we encountered while operating JVM-based services and the ways we solved them.

Twitter is built on Open Source software: we run on Linux and MySQL; on Open Source Ruby, Java, and Scala runtimes; we use tons of Open Source libraries both in our back and front ends. We also contribute back to community. In this talk, we'll briefly catalogue the various Twitter software that is released under an Open Source license. We have a rich asynchronous RPC stack for multiple protocols, a highly efficient message queueing system, a database sharding library, a graph database, and a library of commonly used Java objects. All of these can be adopted by anyone to build their own scalable services.