Architecture and Execution Model for a Survivable Work

We present a novel architecture and execution model for an infrastructure supporting fault-tolerant, long-running distributed applications spanning multiple administrative domains. Components for both transaction processing and persistent state are replicated across multiple servers, ensuring that applications continue to function correctly despite arbitrary (Byzantine) failure of a bounded number of servers. We give a formal model of application execution, based on atomic execution steps, linearizability and a separation between data objects and transactions that act on them. The architecture is designed for robust interoperability across domains, in an open and shared Internet computing infrastructure. A notable feature supporting cross-domain applications is that they may declare invariant constraints between data objects and furthermore declare dependencies on constraints maintained by other applications, leading to flexible, incidental atomicity between applications. The architecture is highly evolvable, maintaining system availability and integrity during upgrades to both application components and the system software itself. Type of Report: Other Department of Computer Science & Engineering Washington University in St. Louis Campus Box 1045 St. Louis, MO 63130 ph: (314) 935-6160 Architecture and Execution Model for a Survivable Workflow Transaction