Date: Fri, 19 Jun 1998 09:38:09 -0600
From: Sean Reifschneider <jafo@tummy.com>
To: lug@boulder.nist.gov, nclug@moss.verinet.com
Subject: [lug] Usenix Report for Thursday June 18, 1998

Usenix Report for Thursday June 18, 1998
Sean Reifschneider, tummy.com, ltd.
===================================

The opening session today was a historical perspective on Unix.  It included
an analysis by Dennis Ritchie on what things they did right and wrong in
Unix.  For the most part they did things right over all, with some of the
details such as pipes and the file-system implementation that they would
have liked to have done differently.  Plan 9 and Inferno apparently handle
many of these details "better".

Richard Miller discussed how they made the first port of Unix to an
InterData machine.  They started by writing some "glue" libraries
so that some of the Unix commands could run on the InterData OS32,
then porting the C compiler, and kernel so it ran under OS32.  Then,
they just whipped OS32 out from under Unix like the magician whips the
table-cloth out from under the glasses.

About this same time, the folks at Bell Labs were also working on a port to
this same machine.  Stephen Johnson talked about their approach, which was
to make Unix more portable in general.  This included a large number of
changes to the C language including type-casting, header files, structures as
we know them today, etc...  The InterData apparently had some quirks
which InterData wasn't willing to fix.  Shortly thereafter the VAX was
released, and the rest is history...

Finally, Juris Reinfelds wrapped the historical session up.  The original
Unix kernel was 10K lines long, and was not portable.  NetBSD's kernel
is close to 2 orders of magnitude larger (over 600K lines of code).
This is because it's so portable and using more modern methods, right?
Well, the Plan 9 kernel is 40K lines of code...

Simics was presented which is a full sun4m emulator, which allows one to
profile a full system.  For example, they ran Linux on the simulator,
and Netscape on top of that.  Reloading a web page generated over 2
million Sparc instructions, and even though this simulated machine is
running 1% the speed of a real sun4m, Netscape spent 25% of it's time
in an idle loop.  They were also able to port Linux to a simulated sun4m
that had 16 processors, and see how things scaled.

With network computers being (apparently) so exciting these days, the
Lava cache presentation seemed timely.  The more network computers you
have, the more powerful a server you need, right?  That may not be the
case if you can push a lot of this work off to dedicated servers that
only service cache hits.  Their system is able to process around 100K
transactions per second, saturating a system at around 670Mbps (with 8
tulip network cards).  What struck me was that they did full predictions
and when they came up with 10% better performance than their anticipated
optimum they were worried.

Darrell Anderson presented "Cheating the IO Bottleneck".  Hard drives
are mechanicly limited in performance by the typical 8 milli-seconds
latency and limited bandwidth.  Networking has gotten faster at a faster
rate than discs have.  For example Myrinet is close to PCI bus speed at
around 100MBps.  

Linus was expecting more of a BoF (Birds of a Feather) session rather
than him giving a talk ("I hate doing that").  The room was packed, I'd
guess well over a thousand people there.  Ted Tso spoke for a while about
what's in Linux's far future, Journaling and B+Trees for ext2 for example.
Linus then took over and talked a bit about what the near future would
be with the 2.2 release.  Then the long list of standard questions were
asked ("What's new for SMP?").  A notable quote about Java: Linus said
"We already have a binary standard, it's called Windows."  This was
the answer to the standard question "What do you think about WINE?".
The quote of the evening was "I do Linux because it's the most fun
thing I can do...  With a computer."

We thought this over and remember at the last Usenix there was quite a Java
presence.  Gosling gave the keynote and Arnold was here.  Bell Labs was giving
out demos of their Inferno environment (which to the untrained eye is similar
to Java).  This year however when Linus said that Java was irrelavent, it
looked more like observation than opinion.  There wasn't even a Java BoF
session.

Jordan Hubbard gave a great presentation on FreeBSD.  He gave a brief
introduction to FreeBSD and their philosophy, then spent a good hour
answering questions.  Ever since I first loaded FreeBSD to port XVScan
a couple of years ago, I've thought that Linux could learn a lot from
FreeBSD.  I think that's more the case than ever.  While Linus has a
great attitude (he doesn't care how, he just wants Linux to be useful),
a lot of people in the Linux community are taking things too personally.
Jordan also just wants people to find FreeBSD useful.  The interesting
quote was "if a company came to me and wanted to port to FreeBSD, I'd
propose they port to Linux.  We can run their binaries, and they get
a much larger supported base."

I wanted to go to the LDAP/IMAP BoF (Birds of a Feather) session, but we
only stayed about 5 minutes because one of the things they were talking
about got Evelyn and I started in a conversation.  We took it outside and
decided on one more project we'd like to work on.  That's the down side
of these conferences -- I now have 6 projects I have no time for but
would like to work on.

Despite the fact that I set up a Python BoF for the same time that Stallman
and Maddog had their BoFs, there were around 20 people in attendance.  I
was a bit worried 4 minutes before when there was only one person there.
There was a good mix of experienced folks and "just looking" people.
Not surprisingly, many of the folks were dissatisfied with Perl.  I can't
help but wonder if the Perl BoF gets a lot of people who are dissatisfied
with Python.

Sean and Evelyn
------------------------------------------------------------------------
Boulder Linux Users Group:  http://stripe.colorado.edu/~dickinsg/lug.htm
------------------------------------------------------------------------