Lorenzo Mezzalira Politecnico di Milano Milano (Italy)
The term "distributed covers a very broad ran-
presented papers.
ge of meanings, both with respect to the elements which are distributed and with respect of the domains along which the distribution takes place.
The first paper presented is "A language coprocessor for the interpretation of threaded code".
In fact we can think of distributing devices in space, duties among devices, operations along the time, etc. It is interesting to mention different and possibly overlapping motivations for distributing something in some way: a) to partition complexity into more manageable parts - the divide and conquer principle; b) to allocate resources where they belong, in order to spare in terms of connections and transportation; c) to use more specialized elements for specific tasks to improve the performance/price ratio; d) to decrease fault-tolerance by having tasks performed by more than o n e e l e m e n t and/or in more than one place; e) to manage concurrency and parallelism of actions, required for increase the throughput or reduce the response time of high performance systems. Of course distributing elements may solve some problems but certainly creates new ones. In fact as a consequence of distribution, such entities emerge as interconnections and the related interface problems, which have to be carefully mastered in order to achieve the intended performances. Actually interface management and optimum distribution, far from being solved problems are more and more worth the attention of many research and design efforts. The title "distributed architectures" of this session applies very differently to the three
In present days powerful CPU's are often realized putting together co-processors sharing the computer resources but each one very specialized and very efficient in its own task: so why not extend the same idea and use a co-processor taking care of all the housekeeping involved in the interpretation of indirect threaded code. Also interesting is the idea of implementing the coprocessor as a microprogrammed device, able to assume different personalities according to the microprograms downlaoded into its control store. The second paper presents an actual implementation: "Parallel control of waste-water treatment plant using a real-time multi-tasking operating system". The industrial application is based on VME bus and PCDOS operating system kernel: it is interesting to know how the activity has been partitioned into processes and which have been the design criteria about hardware and software tools and resources. The third paper presents a system distributed in the more usual meaning of the word: it is composed by various processors linked together with loose cooplings. The problem presented by the paper "Architecture for a serial data link for TUMULT" is centered on changing the present parallel connections into serial ones: the design objectives and altervative solutions are presented and discussed with emphasis on reliability and reconfigurability of the system.