The Success, Failure, and Future of Distributed Objects
Abstract: Thirty years ago Emerald proposed a compelling vision of a distributed object-based future, in which objects with identity and state could migrate seamlessly from one host to another. While Emerald improved on many of its successors, I argue that neither Emerald’s specific choices for maintaining consistency and availability in the face of failure, nor any set of fixed choices, will be sufficient for use in modern distributed applications. Emerald and other distributed languages have implicitly assumed that some fixed choice must be made, and as a result, distributed objects as a language construct have not yet gone mainstream. I believe distributed object languages have a future, but we must find a way to support a diverse set of connection and failure semantics within a language infrastructure. Some choices of semantics may also be incompatible with advanced features of languages such as Emerald; in particular, guaranteeing consistency and availability when migrating stateful objects is problematic.
Mon 20 Oct
|13:30 - 14:14|
|14:15 - 15:00|
Harold OssherIBM Research, David UngarIBM Research, Doug KimelmanIBM Research, James NobleVictoria University of WellingtonFile Attached