Processes Are Data: A Programming Model for Distributed Applications

Many modern distributed applications employ protocols based on XML messages. Typical architectures for these applications follow an approach where messages are organized in queues, state is stored in DBMS, and application code is written in imperative languages. As a result, much developer productivity and system performance is wasted on handling conversions between the various data models (XML messages, objects, relations), and reliably managing persistent state for application instances. This overhead turns application servers into data management servers instead of process servers. We show how this model can be greatly improved by changing two aspects. Firstly, by using a declarative rule language to describe the processing logic. Secondly, by providing a single, unified data model based on XML messages that covers all kinds of data encountered, including process state. We discuss the resulting design choices for compile-time and run-time systems, and show and experimentally evaluate the performance improvements made possible.