[0024] daemon@ATHENA.MIT.EDU (Robert A Basch)  Committee to Examine Athena SGI Effort  10/05/99 18:30 (70 lines)
Subject: IRIX development effort
Message-Id: <199910052014.QAA48450@aupair.mit.edu>
To: cease@MIT.EDU
Date: Tue, 05 Oct 1999 16:14:57 -0400
From: Robert A Basch <rbasch@MIT.EDU>

These are the areas which required a significant development effort
for us to be able to support IRIX 6.5, along with an indication of the
likelihood that the effort would need to be repeated for a future
release:

1) Figuring out how to make diskless boot work in the install.  This
is not well documented, and I needed to figure out how it had changed
in 6.5.  This could be an issue again in future major (6.6 or later)
releases.

2) Porting the ram disk driver to 6.5.  We used the 6.2 sample device
driver originally, but that no longer worked under 6.5, and was no
longer provided.  This will likely be an issue again in future
major releases.  There were other, miscellaneous issues in getting a
read-only root to work.

3) Handling overlay packages in our install software.  They will
presumably continue to use overlays in 6.6.

4) Dealing with the inst virtual memory issue; trying to get SGI to
understand and help us was mostly a waste of time.  Figuring out an
install scheme using multiple sets of distributions, installed in many
passes, instead of installing from all distributions in one pass, was
a major effort, as was determining an optimally-sized swap file for
our miniroot.  While that system should work for future releases, it
will still be a significant effort to determine how to split up the
distributions and passes, and adjust the miniroot swap file size.

5) Debugging problems installing packages to a non-/ target, and/or
in symlink mode.  This will probably always have to be dealt with,
even to some extent with patch releases.

6) Install performance issues.  Current hacks to avoid time-consuming
procedures will hopefully be useful in future major releases,
additional ones could well be needed.

7) Design and implement an OS integrity checker, based on inst
database information.  This should work fine for future releases,
though probably will require some maintenance, and could be enhanced.

8) Port Athena source tree to 6.5, using N32 format binaries with the
new compiler.  We should not have to repeat this effort.

9) Getting nanny/xdm/xlogin to work together.  This will most likely
require some attention for each new major release, a problem inherent
in using the vendor xdm.

10) Dealing with nsd problems, particularly in Athena login.  Another
SGI support nightmare.  For the next Athena release, we will consider
implementing an nsd library to support Hesiod, so that the native
passwd look-ups work, and avoiding caching issues arising from adding
entries to passwd, etc.

11) Other problems where we spent a lot of time dealing with SGI's
lack of understanding, and (sometimes) seeming lack of desire to help
us.  Examples: inst core dumping during update; man tty modes bug;
network problems; setmonitor; system hangs.  Unfortunately, it seems
that this may be an ongoing headache.

12) Miscellaneous: extra cost in supporting multiple hardware
architectures, as each type of hardware requires its own kernel; need
to build root and swap partitions into the kernel.  This is not likely
to change.  Also, there was some work in generating diskless and
miniroot images supporting both Indy and O2, but that effort should
not have to be repeated.
--[0024]--