High Performance Network Programming on the JVM GeeCON, May 2013 Erik Onnen Wednesday, May 15, 13 About Me Wednesday, May 15, 13 About Me Vice President, Architecture at Urban Airship • Most of my career biased towards performance and scale • Java, C++, Python in service oriented architectures • Wednesday, May 15, 13 In this Talk Terminology and Key Theorems • Foundations for this talk (WTF is an “Urban Airship”?) • Networked Systems on the JVM • Choosing a framework • Critical learnings • Q&A • Wednesday, May 15, 13 Lexicon What makes something “High Performance”? Wednesday, May 15, 13 Lexicon What makes something “High Performance”? Low Latency - I initiate an action with a service, how long • does that take Throughput - how many of those operations can I drive • through my architecture at one time? Scalability - how far can we push one service, how does it • fail Productivity - how quickly can I create a new operation? A • new service? Sustainability - when a service breaks, what’s the time to • RCA Wednesday, May 15, 13 Key Theorems Wednesday, May 15, 13 Key Theorems Programming language can have a material impact on • runtime performance - it matters at scale Wednesday, May 15, 13 Key Theorems Programming language can have a material impact on • runtime performance - it matters at scale Writing code is often the easy part of a developer’s job • Wednesday, May 15, 13 Key Theorems Programming language can have a material impact on • runtime performance - it matters at scale Writing code is often the easy part of a developer’s job • Virtualized servers are often the victim of egregious crimes • against networking and system throughput (e.g. ec2) Wednesday, May 15, 13