November 3rd, 6:00 PM, 607 Ross St.: Special meeting to discuss the future configuration of Grex

Oct 11 16:18 UTC 2001

I think we should have a special meeting to discuss what the future
hardware and software configuration of Grex should be.  I think it would be
good to get some of the people who have strong feelings about the question
into the same room to try to hash something out.  And for those of us who
don't have the technical expertise to make specific decisions, it will be a
chance to voice our opinions on what capabilities we'd like to see in the
next Grex, even if we can't make specific purchasing decisions.  Maybe we
could start by making a wish-list of features, and then move on to the
question of how to implement them.

I'll volunteer my house for a potluck, and I'm thinking sometime in early
November.  Who would like to come?

response 2 of 181:

Oct 11 16:33 UTC 2001

I would like to listen to the discussion.  Any day after November 4th should
be fine.

response 4 of 181:

Oct 11 17:39 UTC 2001

heh - you can't even get a summary of the State Of The Disunion, and you
wanna' plan a confab? ;-)

response 6 of 181:

Oct 11 23:52 UTC 2001

I would like to come and have an open schedule.  We can probably bring bread.

response 8 of 181:

Oct 12 01:08 UTC 2001

I'll be there "In Spirit"

[...tips hat to a fellow ghost...]

response 10 of 181:

Oct 12 01:55 UTC 2001

The main pressure I'm seeing for a change is the growing difficulty to
porting software to SunOS.  In the last few months I've failed to install
fully working version of (1) MySQL, (2) PostgreSQL, (3) gcc version 3.0,
(4) any version of the gnu Java compiler.  This trend toward poor support
for SunOS in open source software is not going to go away.  It is going
to get worse.

I don't think we should wait on the latest developments with OpenBSD and
large Sparc machines.  I don't think "recent developments" of any kind
are of more than theoretical interest, much less "real soon now future
developments".  I don't think anything recently developed is interesting
to Grex.  We need stability.  If we want OpenBSD then we should go with
the platfrom where OpenBSD has been most widely run for the longest time,
where it is presumably fairly well debugged:  namely Intel.

If we want to run on large Sparcs, I'd prefer Solaris.

I guess this is a matter of priorities.  My first priority is that the
system keep running reliably without continuous staff intervention.  The
current machine is the first one we've ever had for which that was true.
I'd rather stay with it than go back to the old days of constant crashes.

So I don't want anything that hasn't got years of track record handling
heavy duty server applications.  I don't have any strong preferences
otherwise.

My second priority is that we have a reasonable set of people reasonably
interested in making the plan happen.  STeve doesn't sound all enthused
about moving to a new server.  If Marcus feels the same way, then I
don't see it happening.

I expect it to take a year to get the new system on line once we get
started.  It always has before.  If so, I think we are getting around
time to make some decisions.

response 12 of 181:

Oct 12 08:14 UTC 2001

When I saw "mklinux", I was actually quite surprised with how stable it
was, especially considering it was running on (1) non-mainstream
hardware, and (2) had a fantastically strange OS architecture.  Sure,
there are support issues going with anything but the "herd", but there
are also advantages and it is definitely not a given that it will be
unstable.

response 14 of 181:

Oct 12 16:22 UTC 2001

I like the plan of slipping 'zza to the geeks after a plain-text conversation
with us eating pizza too.  Brains work differently when they have pizza
nearby. :-)

response 16 of 181:

Oct 12 18:03 UTC 2001

November schedule is unknown at this point. I'll go if the time slot selected
is available. Pizza sounds good, or not, depending on what's on it and where
from. But don't choose your toppings on account of me.

STeve, how soon until these developments get revealed?

response 18 of 181:

Oct 12 18:39 UTC 2001

I don't know anything about mkLinux, but my snap judgement is "too immature
and obscure to be of any interest to Grex".

So I went and  read mkLinux.org.  mkLinux appears to be a version of Linux
that runs on top of a mach mini-kernel mostly for power Macs.  The web
site has a broken link to it's FAQ page and it's news page hasn't been
updated since July 2000.  It looks like it was originally developed by
Apple, then passed to the open source community for further development.
Since then the web site has certainly been neglected, and development may
have petered out.  The only site less than a year old that I could find
that mentioned any version of MkLinux refered to the Pa-RISC version
and said not to use it because it was no longer actively developed
or maintained.

So basically, I don't understand why Marcus is talking about mkLinux.
It may well be surprisingly stable, but the fact that it runs at all is
somewhat surprising.  But was it ever as stable as, say, a Sun machine
running Solaris?  If we ran MkLinux could we reasonably expect to go
months between reboots?

Is it going to be easier to port MySQL to mkLinux than to SunOS?

Is there a PowerPC computer that would be a viable alternative for Grex?

So after some research, I think my snap judgement was dead on.  mkLinux was
in pretty good shape, had a fairly large user community, but it still lacked
sufficient support to make it into a serious contender.

As far as I can tell, there are only four open source operating systems
that are even plausible:  Linux, OpenBSD, netBSD and FreeBSD.  A few
of the proprietary operating systems are worthy of consideration, mainly
Solaris.  In most cases, choice of operating system dictates choice of
hardware, as it is typically obvious which hardware is best supported by
a given operating system.

Since there aren't really that many alternatives to choose between, I
don't think the choice is that hard.  In fact, I think the choice is boring.
It'd be neat to do something less mundane, but it wouldn't be practical.

Though if we meet, we should have have a network connection available, so
that we can do quick research on some of the less well-know alternatives.

response 20 of 181:

Oct 12 21:56 UTC 2001

my teacher installed mysql and it has been running smoothly ever since (sunos
5.7).  if you have any questions about porting it to grex, feel free to pass
on any errors messages to me, and maybe i could give you some feedback. 
porting software to sunos is a pain, as is porting software to sparc machines
in general.  i have been running openbsd 2.7 on a sparcstation for some time,
but namely as an aid in developing a personal project, not as a workstation.
as secure and robust as openbsd may be, third party software is less likely
to run on it than on, say, linux or freebsd.  freebsd has endured many years
of testing, more than linux, and is arguably more stable than linux.  m-net
initially had some issues with freebsd, but they have faded with the updates
and patches (i think the latest machine started out with 4.0, which had plenty
of fixes coming to it).  as for powerpc alternatives, you basicaly have
mklinux, yellowdog linux, black lab linux (the altivec yellowdog port for
g4's, costs $$$), darwin (the mac osx core, mach/bsd, close relative of
freebsd), netbsd and openbsd.  netbsd wouldn't install on my old starmax
clone, and i haven't tried openbsd/ppc.  it is of my opinion that mach kernels
are slower, but perhaps a bit more stable than linux or bsd kernels.  black
lab linux would be cool for 128-bit development, but i don't think that that's
your aim with grex (cool development station).

i only bring all of this up here because i live some couple of hundred miles
away and won't be around to bring it up face-to-face. :)

i think that either freebsd or linux on x86 would be an excellent choice,
simply because they are the tried and true, and software porting comes pretty
easy.  a decent security policy isn't that hard to implement for either, and
patching is quick and easy.  for freebsd, you can do a nightly cvsup and make
buildworld to keep the latest patches around for just minutes after security
advisories are released, and the mailing lists get more traffic than the other
aforementioned oses (freebsd and linux, that is).

/end two cents

response 22 of 181:

Oct 13 04:31 UTC 2001

A clarification on SunOS vs Solaris:  Sun's first operating system was
called "SunOS".  It was derived from Berkeley BSD sources, just like
FreeBSD, OpenBSD and netBSD.  It went up to version 4.1.4, which is
what we run here on Grex.  At that point (actually before that point),
Sun threw it away and wrote an entirely new operating system based on
Bell's System V Release 4 Unix.  The driving motivation for this was to
support multiple processors effectively.  They called the new operating
system "Solaris."

To ensure perpetual confusion, they then retroactively renamed all the
SunOS releases "Solaris".  For example, "SunOS 4.1.4" is now to be called
"Solaris 1.1.2".  Meanwhile, they kept assigning SunOS version numbers
to new releases of Solaris, so that "SunOS 5.7" is also "Solaris 7".
For a key see http://www.ocf.berkeley.edu/solaris/versions/

So most of the time when people refer to "SunOS" they mean version 4.1.4
or earlier, while "Solaris" is everything later.  Two different names
for two different operating systems.  It'd be nice if Sun did the same,
but it doesn't.

So porting MySQL to SunOS 5.7 is easy.  That's a fairly modern
operating system (released in 1998) that is directly supported by
the MySQL developers.  In fact, I wouldn't call it porting at all.
Just follow the directions and it builds.  I think you can even get
binary distributions prebuilt for it.

SunOS 4.1.4 is a whole 'nother ball of wax.  It dates from 1994 and is
really essentally identical (differing in bug fixes only) to 4.1.3 which
was release in 1992.  I was able to get MySQL built, with a little fixing,
and it even works for a while.  But after running for a day or so the
server stops responding to queries.  I'd be interested in someone getting
MySQL to run reliably on SunOS, but SunOS 5.7 doesn't count because it isn't
SunOS.

Solaris is one of the things we would seriously consider upgrading to.  But
moving from SunOS to Solaris would not be any easier than moving from SunOS
to FreeBSD.  In fact, it might well be harder, since SunOS is more closely
related to FreeBSD than to Solaris.

Speaking a person who does a large portion of the software installation and
development here, I don't care at all if we are on x86 or Intel or PowerPC.
Aside from byte-order issues, it's never made a difference to any program I've
written, and byte-order issues are easy enough to deal with.

response 24 of 181:

Oct 13 06:55 UTC 2001

For the past 8 months I have been tasked to support a large application
(3 layers) on multiple sun boxes (running solaris 6,7,&8, and three
different versions of java...don't ask).  Most of the common freeware
things one might want are already available as packages from a number of
sites.  As well, porting software to solaris is generally not much of a
pain.  Solaris/Sparc is a pretty 'hot' combination, but it does seem to
peter out under load.  It starts out real fast, quickly slows down under
load, and falls apart altogether under extreme load.  (crashes that
sun's explaination is 'well, thats just sorta the way it is' or 'so do
not do that'.)  For performance and business (cost) reasons the client
has decided to switch over to AIX/rs6000 platforms. (IBM is the OEM for
the existing SUN hardware so go figure.)  However, Solaris is very
stable considering the way this application is bitch slapping it around,
and it was performance, not stability (even of Solaris6 aka Sunos5.6)
that prompted the change.  Thus upgrading grex to Solaris (assuming all
the hardware was supported - better check carefully) would be a good
idea.
(Upgrading to AIX/rs6000 might be an option if there were specific
performance issues, but it ain't cheap so I suspect its out of the
question.)  I don't know what the license cost would be to Grex for
Solaris - some might suggest it would be free or minimal for non-profit.
Somebody should probably get an official quote from Sun - or solicit a
donation.  Sun is hurting these days so nice press even on a local basis
might be enough incentive.

If you have to lay out significant gelt for Solaris upgrade, then you
are left with the idea of putting the money towards intel hardware and
one of the freeware OSs (Solaris/Intel is somewhat freeware so ought to
be on the list).  First of all, even if you choose Solaris/Intel, the
Intel environment is going to be different, thus the difficulty in
support of SunOs you see now is merely going to be shifted not reduced. 
Secondly, while PC hardware can be inexpensive, Intel based hardware to
really do a server/multiuser system ain't.  I'm not real clear on the
details why, but my experience is that even with dual or quad CPUs and
gobs of memory a PC box under load can quickly become 'stodgy' - this
seems to be true Linux or Xbsd ( Solaris seems to do a little better).

Has the current hardware really been maxed out as far as memory/CPU?
Is there really a problem to be addressed?  (porting software may be
difficult, but is the software being ported something that your general
user of grex actually needs to perform the function of using grex?)