A Compile Time Model for Composing Parallel Programs

Many distributed memory machines support connection-based communication instead of or in addition to connection-less message passing. Connection-based communication can be more efficient than message passing because the resources are reserved once for the connection and multiple messages can be sent over the connection. While long-lived connections enable more efficient use of the communication system in some situations, managing connection resources adds another level of complexity to programming such machines. iWarp is an example of a distributed memory machine that supports long-lived connections. To aid the iWarp programmer and program generator tools, we developed a tool chain that enables the programmer to define connections and compose parallel programs. The communication tool chain has been in use for four years in various forms. In that time, we have found many benefits and a few pitfalls in our model. This paper describes the design of the programming model and tools and discusses our experiences with this implementation. Supported in part by the Defense Advanced Research Projects Agency, Information Science and Technology Office. under the title "Research on Parallel Computing," ARPA Order No. 7330. Work furnished in connection with this research is provided under prime contract MDA972-90-C-0035 issued by DARPA/CMO to Carnegie Mellon University. The views and conclusions contained in this docum~ent are those of the authors and should not he interpreted as representing the official policies, either expressed or implied, of the U.S. Government.