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