On Mon, Feb 18, 2002 at 11:04:18PM -0700, Rick Kelly wrote:

> Linux gets the attention because there are so many "distros" and the basic
> power of web/media saturation.

Let's be fair though: some of that attention is because it does things
the others don't do.

I cannot get xfs or fast 3D graphics support in a NetBSD box.  Even the
sound support is not as nice.  I also cannot get high-end server support,
can't run on IBM mainframes, etc.

It's not all hype, and I for one learned to ignore that a long time ago.
I don't even read Linux discussion groups much any more because it's just
a waste of time, nor do I care how many million dollars IBM pours into it.
They did the same for OS/2 and look where it went...

I for one want to see people like IBM make good on their promises to
boost Linux kernel strength, make SMP better, make some good hardware
for Linux... that sort of thing.

They and others talk a lot... but that's about it.  The real work of
Linux is still being done by the same people as always.

But it could be different, and not just for Linux.  I think hardware
companies would benefit a great deal, in the long run, by going back to
being hardware companies.

> The IA64 project got started back when the Alpha was clocked at 70mhz. Intel
> figured that it could go with a complex design and that the compiler writers
> would make everything work. 1993 - Alpha 150Mhz. 1994 Alpha 300Mhz. That's
> when Intel should have dropped the project. The Alpha is already faster than
> Itanic. Now. Today.

It always was faster.  I used to show some Itanic-loving associates spec
sheets on the Alpha systems out there.  It was just priceless when they
said, "Yeah, but when will this one actually _ship_ in real machines?"
That's when I'd point over to the rack with several of them in it...

Oh well, it hardly matters now.

> >I used DEC UNIX for years, and it's amazing to see servers we paid $10K
> >for just a few years ago going for $300.
> 
> Yup. I want an EV6 or EV7 machine. Used them, but never owned one.

Well, I found 800 series machines, nice little purple server towers,
for just a few hundred on ebay.  1000A/2000A systems (like 16 64-bit
PCI slots, a SCSI drive backplane, etc) for less.  They are big and
relatively slow CPUs, but they run like clockwork.

This is another reason I want SMP for NetBSD: these old servers still
work great, they just cannot handle the latest load of web crap people
are hosting, so they are dirt cheap now.

> >This won't hurt anything if they'll put support into Linux to push
> >Red Hat off the hill and push a sane distribution.  Then start working
> >on some needed kernel changes and fund increased development of the
> >VM.
> 
> But you have to see that except for HP, which has it's special Linux for
> $2000, the rest of these companies do as little value added as possible.
> They just want to sell hardware.

Well, they'll ultimately find that they cannot sell the hardware until
they get off their collective assen and do some real work.  IBM is one
of the big ones that needs to put up or shut up.  All I see from them
so far is an army of Java monkeys grunting out steaming mounds of slow,
bloated pseudocode I don't need.

> Sun is just plain nuts. Intel box margins aren't worth the effort.

Yeah, but have you looked at a "real" Sun?  Tell me that isn't a PC
with a SPARC CPU in it.  Those Blade systems are still better than a PC,
but they are bloody awful compared to the U2 systems and earlier.

DEC did the same time around 1995 or so, and they charged big bucks
for Alpha powered systems that were underpowerd otherwise (slow SCSI,
poor sound, OK but largely unsupported graphics, etc).

> >The whole UNIX world needs to push for better video support as well, and
> >not just drivers for Linux, _all_ systems running XFree servers.
> 
> Well, the worst video support is probably on Intel as it is a moving
> target,

If the manufacturers would just allow drivers to be written, the moving
target problem is not really a problem.  Besides, outside of nVidia and
ATI, everyone else is just a margin player now anyway.

Some of them are trying to change, but it's evidently very painful.

> >This is one area where the GNU zealots might actually help, if they
> >don't self-destruct or drive everyone away in the process.  I won't
> >make any bets there...
> 
> Maybe, as long as RMS doesn't explode.

Images of RMS exploding aren't that hard to imagine.  If he ever sings
again I would be willing to place bets on it.

> Well, there's ext3 and other filesystems now, but crashing a system with
> ext2 filesystems mounted async is a bad surprise.

I generally don't lose data with ext2.  Must just be lucky.  In fact,
I really don't find either ext2 or ffs to handle crashing all that well.
I lose data in both from time to time in hard crashes, so I work on
minimizing them rather than worrying about with filesystem will
recover when it happens..

That's one reason I switched to SGI's xfs on Linux.  Now I just don't
worry about it, at least nowhere near as much as I once did.  Of course,
regular backups helps.

> >Linux recovers from crashes better than my NetBSD and FreeBSD systems.
> >SGI's xfs filesystem has never let me down yet. It's an example of what
> >can be done that was previously considered impossible.
> 
> That support is new. Is it in all Linux "distros"?

The only one I use is Slackware, so I don't know, and I built it all
myself because I like to track the latest releases from SGI.  They seem
to pretty well make it better each time.  Eventually I hope they put
in streaming I/O in xfsdump.

I know some of them have it, but I couldn't tell you which.  It was only
recently that SGI decided it was ready for release.  It was frustrating
to wait for it, but given how stable it has been, it's obviously a good
thing they waited.

IBM seems to have released JFS too early, and I had trouble with that
other similar filesystem whose name I cannot remember right now.

> The news these days is that Linux 2.4.x pukes under heavy load.

Hmmm... I don't have a problem myself, but I don't have the large hardware
some people have either.  

2.4.x was an example of waiting far too long to close the door on changes.
Of course, it is a rather fantastic leap over 2.2, so after the dust
settles it will be worth it.

I'm hoping that recent FreeBSD developments make a moot point for my
workstation, but I might end up running Linux for some time.

On my workstation, I have no choice but to run Linux 2.4.x because my
video card needs it.

> One of the hats that I've worn in the past is qualifying Intel hardware for
> running UNIX. So I have been very conservative about what I have used for
> running NetBSD. Most of my NetBSD i386 machines are Pentium 1 architecture.
> Everything else is sparc accept for 1 Athlon box.

But even if you qualify hardware, that doesn't stop a kernel bug from
killing your system.

Even with NetBSD, such bugs exist, including a particularly nasty one in
1.5.2 which I have seen 4 times recently bring my system down.  It seems
it's on track to be fixed in 1.5.3.

-- 
UNIX/Perl/C/Pizza__________________________________shannon@widomaker.com