Fault tolerance is a key requirement in process support systems (PSS), a class of distributed computing middleware encompassing applications such as workflow management systems and process centered software engineering environments. A PSS controls the flow of work between programs and users in networked environments based on a "metaprogram" (the process). The resulting applications are characterized by a high degree of distribution and a high degree of heterogeneity (properties that make fault tolerance both highly desirable and difficult to achieve). We present a solution for implementing more reliable processes by using exception handling, as it is used in programming languages, and atomicity, as it is known from the transaction concept in database management systems. We describe the mechanism incorporating both transactions and exceptions and present a validation technique allowing to assess the correctness of process specifications.

We describe the concept and implementation of an inter-process communication facility based on the exchange of events between concurrently running processes. In contrast to previous approaches based on shared data stored in a common database, our approach has the advantage of being platform independent and providing straightforward support for distribution. In addition, we also explore the problem of process atomicity and consistency when events are revoked due to the abort of processes. We present a family of recovery protocols that allow to control the effects of aborted processes.

Availability in process support systems (PSS) can be achieved by using standby mechanisms that allow a backup server to take over in case a primary server fails. These mechanisms, resembling the process pair approach used in operating systems, require the primary to send information about state changes to the backup on a regular basis. In PSS where all relevant state information is stored in a database, there are two principal strategies for synchronizing a primary-backup pair. One is to use the replication mechanisms provided by the DBMS. Another is to implement a message mechanism to exchange state information between servers above the database level. For both approaches, several variants exist that allow to trade run-time performance for failover time. This paper discusses the possible strategies and evaluates their performance based on an implementation within the OPERA process support kernel.

Exceptions are one of the most pervasive problems in process support systems. In installations expected to handle a large number of processes, having exceptions is bound to be a normal occurrence. Any programming tool intended for large, complex applications has to face this problem. However, current process support systems, despite their orientation towards complex, distributed, and heterogeneous applications, provide almost no support for exception handling. This paper shows how flexible mechanisms for failure handling are incorporated into the OPERA process support system using a combination of programming language concepts and transaction processing techniques. The resulting mechanisms allow the construction of fault-tolerant workflow processes in a transparent and flexible way while ensuring reusability of workflow components.