Distributed transaction-processing systems must manage such functions as concurrency, recovery, and replication. One way to improve their efficiency and reliability is to increase software modularity, which means the separate components should execute in separate address spaces to permit hardware-enforced separation. This structure offers advantages but demands efficient interprocess communication (IPC) services. In our research at Purdue University, we are investigating mechanisms and paradigms for efficient communication support in conventional architectures, such as virtual-memory, single-processor machines with no special IPC hardware support. (Some mainframes have hardware assistance where more than one address space can be accessed at the same time.) We are studying communication designs in the context of the Raid system, a robust and adaptable distributed database system for transaction processing.' Raid has been developed at Purdue on Sun workstations under the Unix operating Communication svstem in a local area network. software is critical in distributed computing systems. This research identifies efficient mechanisms and paradigms for distributed transaction processing in a replicated database environment. In Raid, each major logical component is implemented as a server, which is a process in a separate address space. Servers interact with other processes through a high-level communication subsystem. Currently, Raid has six servers for transaction management: the user interface (UI). the action driver (AD), the access manager (AM), the atomicity controller (AC), the concurrency controller (CC), and the replication controller (RC). High-level name service is provided by a separate server, the oracle.