Track: Architectures you've always wondered about

Abstract: Amazon's Simple Storage Service (S3) provides a key-value-oriented interface to an infinitely scalable and durable storage system.  One of the design requirements for S3 was to handle wildly varying access patterns initiated by our users.  Maintaining availability when experiencing unexpected, unusual, and often massive load is a difficult problem.  Nevertheless, a rigorously designed and implemented system will not experience widespread unavailability in these situations.   This session will focus on the approaches taken by S3 in architecting its systems to survive unexpected and overwhelming load.  The session begins with a generic description of a common service architecture and then discusses the potential failure points within the design.  It describes the necessary software and technology that all components in a service-oriented architecture must have as basic building blocks.  Then, for each layer in the system, it presents architectures which mitigate the availability impact of overload scenarios.  It concludes with a last-resort mechanism that all services must use to handle any remaining scenarios.