Eecient Coordination Support with Transputer Systems Eecient Coordination Support with Transputer Systems

With the increasing availability of heterogeneous networks, the coordination of distributed services, the communication between diierent applications, and the development of distributed programs is becoming more and more important. In previous work we proposed coordination extensions for existing sequential programming languages like Prolog and C, leading to Prolog&Co 1 (Prolog plus Coordination) and C&Co (C plus Coordination). The coordination extensions provide ne and coarse grained parallelism, software fault tolerance through function replication, and a high-level, reliable communication via communication objects. The semantics of the base language are not destroyed but only slightly expanded by introducing a few new language constructs. The realization of these new coordination languages employs a coordination kernel running in the background and handling all requests concerning communication with other sites. So far, we have implemented the kernel on single processor workstations in the Internet. As this kernel has to fullll a large number of functions, it may become a bottleneck if many local &Co-systems must be served that involve a lot of internal and external communication. This paper discusses how the coordination kernel can be parallelized and implemented on a transputer architecture. The idea is to have all the time-consuming functions of the kernel running on the transputer with the kernel at the host site stripped to the basic functions, i.e., waiting for requests from remote kernels, and to handle the communication with the local software systems (V P L , C&Co).