snorthc@pokey.nswc.navy.mil (Stephen Northcutt) (08/31/90)
I have followed the RPC discussion though I don't claim to understand everything. At this point I am kind_of_sorta familiar with: ISODE, Netwise, and OSF's DCE. Three months ago, a co-worker and I experimented with the ISODE RPC system: ANS.1 compiler; app. templates. The application, though trivial, worked. So now I have a warm fuzzy for ISODE as an RPC tool. Project: This week I was assigned to investigate a networked (distributed, concurrent) approach for some scientific modeling applications as a Proof of Concept. The purpose is to compare contrast this approach against a couple "Parallel in a box" machines we have. I would prefer not to use ISODE for this, because I am alpha testing some OSI software based on ISODE and it is hard to keep my code separate from the vendor(s). Requirements: The hardware base I intend to use is SUN 3/60s (SUNOS 4.0.3), SUN 3 280 (SUNOS 4.1), DEC 3100 (ULTRIX 4.0) and a couple unloved uvaxes (ULTRIX 4.0). If the RPC I choose is available for the platform, I can utilize a CDC 990 running NOS/VE. Round 1, the systems will utlize ethernet, but we have already ordered and FDDI interface for the SUN server and I want to move the systems to FDDI as the project progresses. I will be running TCP as the transport initially, but am willing to consider XTP or TP4 later. Software. Most of the existing code is Fortran, largely based on Dr. Al Morris's math libraries. I plan to write mostly C or G++ code, to "glue" things. Now the RPC or whatever. The ideal system would be shipping by 1 OCT for the HW/SW platforms I have available. I need concurrency, I would like to be as close as possible to a POSIX compliant threads environment. In a perfect world, to do development with this system I would interface with an X Window based workbench, with at least rudimentary CASE capabilities.... whoops I was dreaming for a minute, sorry. For this experiment to return any meaningful results I will need some profiling/timing tools, preferably driven from some meta applications monitor. Possible approaches: Reinvent the wheel, we could try to develop our own client server stubs... I hear you laughing... actually before we ever heard of Netwise/OSF et al a sister group built a big hairy networked distributed simulation, it is fairly elegant at this point (3 yrs old), but does not support concurrency. Netwise SUN ONC? I attended the Netwise talk at interop 89, and have seen some of those viewgraphs mucho times. This might be a good option when they support concurrency. OSF DCE? I heard these talks a Xhibition and have follow up materials. Sounds good, IF I can get production code in a timely fashion. My gov't agency isn't going to deal with the license restrictions req'd to join OSF (we having trouble getting the right signatures for MOTIF) and I am not bullish about snapshots. Network Linda? I need to learn more about this? Any other possibilities, paradigms? Advice Requested: =================================================================== Stephen Northcutt (snorthc@relay.nswc.navy.mil) Work: (703) 663-7745 Home: (703) 371-4184 Paper Mail: Code E41, NSWC, Dahlgren VA 22448