[2559] in SIPB-AFS-requests
Re: SIPB service needs
daemon@ATHENA.MIT.EDU (mhpower@MIT.EDU)
Mon Nov 25 18:32:24 1996
From: mhpower@MIT.EDU
Date: Mon, 25 Nov 1996 18:31:56 -0500
To: ghudson@MIT.EDU
Cc: star-maintainers@MIT.EDU
I'm curious why you think you have "achieved something resembling
consensus" on the item of:
> Discuss/NFS Possibly an upgrade to a Sparc 5,
> and some more disk
...
> * An upgrade to a sparc 5 for discuss and NFS (if we can do
> kerberized NFS under Solaris any time soon; otherwise this
> falls off the list entirely)
when, as far as I can tell, it hasn't even been suggested on the
rtfm-maintainers list, or shown up in the sipb zephyr logs.
My opinion is that adding a Solaris server for discuss and nfs is a
bad idea. There are a lot of costs to adding yet another SIPB server
platform, and I don't think our performance requirements for discuss
and nfs service even come close to justifying this. The costs would
include needing to maintain the various software packages used on
SIPB's server machines for three platforms rather than just Ultrix and
SunOS, and needing to follow and fix all reported Solaris security
problems. Also, I think we lack evidence on the potential reliability
of the port of Athena's kerberized NFS to Solaris (here, I'm ignoring
the issue that I don't know whether the port actually exists or not).
I don't know of particular problems with discuss or kerberized NFS
that are specific to decmips machines, and I think my preference would
be for staying with the decmips platform for these. My feeling is that
levels of interest in discuss and kerberized NFS are both dropping
(e.g., fewer new meetings, fewer changes to credentials files), so
it's probably unlikely that higher performance would be needed in the
next year to two unless that trend changes significantly.
Another possible argument is that we might want to start moving some
SIPB services to free operating systems, and moving our discuss
service to Linux might be an initially attractive option if there were
support for this concept. (For example, there are already a few
discuss servers on campus running Linux.) Obviously, there are some of
the same costs associated with free-OS platforms as with Solaris, but
it would at least give us the advantage that we could respond by
ourselves to some new classes of problems. I don't know that dealing
with syn attacks on our discuss server is a particularly useful
example here, but there might be some analogous issue in the future
where it would be valuable for SIPB people to have full access to
source code. Personally, I would pick Ultrix over even a dedicated
discuss server on a new Linux machine, but I think either Ultrix or
some reasonable free-OS platform (probably not the Hurd) would be much
better choices for us than would Solaris.
Matt
