LinkedIn - A Professional Network built with Java Technologies and Agile Practices

Slide 1: LinkedIn: A Professional Social Network Built with Java™ Technologies and Agile Practices Nick Dellamaggiore, Principal Software Engineer Eishay Smith, Senior Software Engineer

Slide 2: Learn how Java™ and agile practices are employed in building large-scale consumer internet sites like LinkedIn, the world’s largest professional network. 2008 JavaOneSM Conference | java.com.sun/javaone | 2

Slide 3: Agenda LinkedIn(troduction) Agile engineering process Architecture Building LinkedIn with Java™ Questions/Discussion 2008 JavaOneSM Conference | java.com.sun/javaone | 3

Slide 4: 2008 JavaOneSM Conference | java.com.sun/javaone | 4

Slide 5: LinkedIn Your professional identity on the web The site The numbers • Launched in 2003 • 22 million members • 3 degree network • 4+ million unique visitors/month • Strong focus on privacy • 40 million page views/day The company • 2 million searches/day • Profitable since 2006 • 250K invitations sent/day • 250 employees • 1 million answers posted • Based in Mountain View, CA • We’re hiring! 2008 JavaOneSM Conference | java.com.sun/javaone | 5

Slide 6: Engineering Process Release often • Short, 2-4 week development cycles All dev tasks are broken down into small, manageable engineering cards Strong focus on testing Minimize meetings, status (standup meetings as needed) 2008 JavaOneSM Conference | java.com.sun/javaone | 6

Slide 7: Engineering Cards 2008 JavaOneSM Conference | java.com.sun/javaone | 7

Slide 8: Testing 6500+ unit and integration tests 500 HtmlUnit tests Large-scale Hudson deployment • 20+ nodes • Continuous integration build/smoke test on SVN commit • Nightly test run (full test suite) 2008 JavaOneSM Conference | java.com.sun/javaone | 8

Slide 9: “Celebrities” Test Network 2008 JavaOneSM Conference | java.com.sun/javaone | 9

Slide 10: Hudson 2008 JavaOneSM Conference | java.com.sun/javaone | 10

Slide 11: Testing… Lessons Learned Test suite too focused on integration • Long running time • Long-running setup/teardown routine • Exacerbated by move to SOA • Tests become redundant due to overlapping integration tests • Better: use mocks (EasyMock works great) 2008 JavaOneSM Conference | java.com.sun/javaone | 11

Slide 12: Architecture 2008 JavaOneSM Conference | java.com.sun/javaone | 12

Slide 13: LinkedIn Architecture: 2003-2005 2008 JavaOneSM Conference | java.com.sun/javaone | 13

Slide 14: LinkedIn Architecture: 2006 2008 JavaOneSM Conference | java.com.sun/javaone | 14

Slide 15: LinkedIn Architecture: Today 2008 JavaOneSM Conference | java.com.sun/javaone | 15

Slide 16: LinkedIn Architecture: Today 2008 JavaOneSM Conference | java.com.sun/javaone | 16

Slide 17: Service-oriented Architecture Benefits Caveats • Each service can scale • Diminished data integrity, independently consistency • Encourages decoupling, reduces • No cross-service transactions, code+deployment dependencies referential integrity or joins • Graceful degradation of • Backward compatibility functionality • “8 Fallacies of Distributed Computing” Takeaways • Not applicable to small or simple sites, startups • Build services with interfaces, POJOs, dependency injection • Be conscious of API design, granularity 2008 JavaOneSM Conference | java.com.sun/javaone | 17

Slide 18: Services 2008 JavaOneSM Conference | java.com.sun/javaone | 18

Slide 19: The Cloud Backend server caching the entire LinkedIn Network Graph cache • Updated via Databus • Persisted to disk on shutdown Transient network cache • Computationally intensive to build • Bound to member session • Sticky load balancing to Cloud Graph operations: • findRoute(m1, m2) • visit(visitor, deg) • visit(visitor, deg, since) 2008 JavaOneSM Conference | java.com.sun/javaone | 19

Slide 20: The Cloud The numbers 22M nodes, 120 million edges 12 GB JVM heap • In-memory caches implemented in C++, accessed via JNI 40 Cloud server instances deployed in production 2008 JavaOneSM Conference | java.com.sun/javaone | 20

Slide 21: LinkedIn News Part of a new trend in LinkedIn Crawling the web for news • Distributed task management • Parsing technologies: StAX, Rome Indexing and searching • Serving the right news to the right people Duplications Duplications Duplications • Real time problem, fast solution • 500 articles batch = 125K compares 2008 JavaOneSM Conference | java.com.sun/javaone | 21

Slide 22: News Service Architecture 2008 JavaOneSM Conference | java.com.sun/javaone | 22

Slide 23: LinkedIn: 99% Pure Java™ Plumbing Services Tier Web Tier Tools Spring Remoting LinkedIn Spring Louis (in-house fwk) Hudson (CI) ActiveMQ Eh-cache Spring MVC Eclipse+Mylyn Quartz Lucene Grails JIRA/Greenhopper HttpClient Jetty DWR JUnit, HtmlUnit 2008 JavaOneSM Conference | java.com.sun/javaone | 23

Slide 24: Why we love Java™ at LinkedIn Static typing is a lifesaver • Huge codebase, 1M+ lines of code • Refactoring with confidence • Code navigation (via IDE) • 20 active branches, compiler helps with merging • Scaling the engineering organization • Currently 50+ engineers, 8 teams… and growing • Excellent talent pool of Java™ engineers to hire from Community 2008 JavaOneSM Conference | java.com.sun/javaone | 24

Slide 25: LinkedIn: A Professional Social Network Built with Java™ Technologies and Agile Practices Nick Dellamaggiore (http://www.linkedin.com/in/nick) Eishay Smith (http://www.linkedin.com/in/eishay) We’re hiring!