Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cornell!ken From: ken@gvax.cs.cornell.edu (Ken Birman) Newsgroups: comp.os.mach Subject: Re: MIG and netmsgserver vs. NIDL and location brokers Message-ID: <32531@cornell.UUCP> Date: 27 Sep 89 01:41:45 GMT References: <28300001@hplabsb.UUCP> <28300002@hplabsb.UUCP> Sender: nobody@cornell.UUCP Reply-To: ken@gvax.cs.cornell.edu (Ken Birman) Organization: Cornell Univ. CS Dept, Ithaca NY Lines: 57 In article <28300002@hplabsb.UUCP> caswell@hplabsb.UUCP (Deborah Caswell) writes: >We already have ISIS running on various machines in our lab on HP-UX. >What does ISIS have to do with Mach? Is there a version of ISIS which >is implemented using Mach ports? > We recently ported ISIS onto MACH, although not using Mach ports (what advantage would this give us? Not much, as far as we can tell). This port is available as part of ISIS V1.3, and will also be included on the standard Mt. Xinu distribution of MACH next January (along with Camelot, a system for doing transactional shared memory). Anyone using ISIS at HP is probably using an old version, since ISIS V1.3 won't really be "officially" out until next week. That version didn't run under Mach except on the NEXT machine. I take it that OSF has made a decision to go with Mach as their OS. We submitted ISIS to OSF under their recent RFT for distributed computing environments. However, OSF is a bit of a black box and there is no way to predict whether they will actually include ISIS into their offering. Whether they do or not, my group is planning to introduce a number of ISIS-based application programs next year, including a platform for dealing with sensors and actuators in realtime, a file system that extends NFS to support transparent replication and fault tolerance, and software for what we call "distributed application management" (namely, the problem of supervising/scheduling/auto-restarting a distributed but not especially robust application system). And, we will certainly have a version of ISIS running on the OSF system. Regarding the original call for a "comparison", I am obviously too biased to comment on this. But, I would like to observe that ISIS is unique in addressing the issue of "consistency" as it arises in a distributed system subject to failures and concurrency. We believe that if you don't start by tackling this fundamental problem, you simply won't end up with a robust server. On the other hand, once you DO address this problem, you get a more uniform and consistent environment, and application programs are far simplified. It seems that with enough elbow grease (mostly on my part), the ISIS system can perform as well and scale as well as any other technology. (ISIS applications generally don't see these issues.) If you want papers on ISIS, contact me or my secretary, Helene Croft (croft@cs.cornell.edu). If you want a copy of ISIS V1.3, you can copy it using anonymous FTP or via UUNET, and Helene can explain how. Its free and, with this latest version, seems remarkably robust. We also have a news group (comp.sys.isis), mostly to announce releases and bug fixes, and to answer questions about the system. Not a very active group, so far. In addition to Mach and HP UX, we run on DEC systems (Vax, 3100), the MIPS box, the Apollo 10.1 UNIX release, IBM PC/RT's under AIX, the NeXT machine, and Gould's 9800 series. Our system has interfaces to C, GCC, ANSI C, C++, Lucid and Allegro Common LISP, Fortran, and (in a limited way) C-Prolog. We are hoping to see ports to the Iris, Cray's UNIX, and the 3B series of machines soon, and a port to Bull's equipment exists but I don't have a copy of it yet. Ken Birman