P e rformanc e C ommunic ation

Mernory-based messaging, passing mes- sages between programs through a shared memory segment, is a recognized technique for efficient com- munication that takes direct advantage of memory system performance. However, the conventional oper- ating system and hardware support for this approach is inefflcient, especially in large-scale multiprocessor systems. This paper describes interface, software and hard- ware optimizations for memory-based messaging that efficiently exploit the basic mechanisms of the memory system to provide superior communication perfor- mance. We describe the overall model of optimized memory-based messaging, its implementation in an operating system kernel and hardware support for this approach in a scalable multiprocessor architec- ture. The optimizations include address-valued signals, message-oriented memory consistency and automatic signaling on write. Performance evaluations show these extensions provide a three-to-five-fold improvement in communication performance over a comparable software-only implementation.