rab@brillig.umd.edu (Bob Bruce) (01/31/88)
[ This is in response to a note I sent out asking for the status of some ] [ on-going research projects. Status resports are always welcome. I would ] [ like to get comp.os.research so it is regarded as a medium for publication ] [ of interesting results prior to publication in a conference or an archival ] [ journal. This would shorten turn-around time and increase interaction. DL ] We have built several machines in the lab. The first, called `ZMOB' was built in 1979, and consisted of 128 Z80 microprocessors arranged in a slotted ring architecture, with a 48 bit wide communication path between them. When a processors sends a message over the communication belt it can broadcast it to all other processors, address it to a particular processor, or send it using a pattern register so that only a subset of other processors receive it. All of this is done in hardware. The next machine built, called `MCMOB', is similar, but consists of 16 mc68010's with MMU, and DMA support. It uses the same communication belt, but runs it at a higher speed. The latest machine (not quite operational yet) consists of 4 mc68020's that have a megabyte of shared memory between every two processors. There are a number of research projects currently being done on these machines. The most significant aspect of these machines is the very high speed, and low cost of communication. We are attempting to make the inter-processor communication as invisible as possible to the user. When we are done our operating system should do automatic load balancing and resource allocation to take advantage of all the distributed computing power and memory available. In addition to OS research, there are projects involving data flow algorithms, parallel inference, parallel programming languages, numerical simulation, and combinatorial optimization. Most of the software that has been developed is very machine dependent (these are some pretty weird machines), so I don't know if it would be very useful to other researchers at this time. I think that we should eventually clean up the code to make the hardware interface a little more general, but that is something that we never seem to be able to find the time to do. ARPA: rab@mimsy.umd.edu UUCP: uunet!mimsy!rab