[rescue] "i do my best thinkin' on the bus..."

[Skeezics Boondoggle]

wow.  good thing i had Too Many Beers yesterday and read rescue in digest
mode... moving slow this morning... otherwise i love a good flamefest, now
and then.  reading the list with the safety on helps. :-)

so, two things emerge:  first, i'm glad i ride the bus/train to work every
day.  :-)  seems some of you are getting a jump on some kind of 'mad
max'-style future, apparently.  ("geeks with guns" is a good band name, at
least. :-) there is something eerily appropriate about people who surround
themselves with stinky old computers driving heavily modified vehicles
laying smoke screens, talking about spikes and gun racks, etc.  hell, you
could load the back of your pickups with vt102's and launch them like
depth charges - those things are bloody indestructible (cue apocryphal(?)
url about guy who ricocheted a .50 cal machine gun bullet off the glass of
one and nearly shot himself...)  anyway, y'all seem like an interesting 
group, and i'm probably nutty enough to fit in. :-)


second, re: the usual blather about x86:  nobody is going to "win" this 
one.  this is the 00's, and the world is all about entrenched ideas, 
and computer-related religious issues are just as silly as everything 
else, but then, i always stand up for the underdog, so here goes. :-)

i take my stand firmly on the side of the anti-intel team.  why?  simple
things, to start with:  real serial consoles.  openboot proms instead of
useless bios firmware.  graceful degradation under load.  consistently
high throughput, predictable latency, reasonable failure modes.  all the
things that box stuffers have to cut corners on to maintain commodity
price points and eke out their slim margins.

but most importantly, i support the anti-wintel point of view for the
simple reason that monopolies are bad for everyone - *no one company* has
all the best ideas about how to design chips or how to write software, but
the wintel world seems to believe that - and that vision of the future
makes me ill just to think about it...  without someone to copy or buy
ideas from, what happens to innovation?  if next/apple/be/others hadn't
been around, what would m$ be pushing down everyone's throats today?  
tiled windows and dos boxes?

you _have_ to give intel credit for doing amazing (perverse? :-) things
with process technology.  doesn't it seem really strange that everyone
else has stalled out around 1.05-1.3ghz, while intel is pushing twice
those clock rates?  i suppose that says _something_ about architecture,
but i think if hp/decpaq/sgi hadn't sacrificed hppa, alpha and mips on the
altar of commoditization most of the things that make the itanic
interesting would not have seen the light of day...

sure, you can dismiss my arguments against wintel as "emotional" or
"illogical", but i am for the notion of elegance, and monopolies are not
elegant.  brute force, incessant marketing, vicious and overly-aggressive
business tactics - those things *are* repulsive.

but to back it up with almost two decades of programming/sysadmin
experience of my own, i have plenty of "real world" examples:  *6 years
ago* at an isp where a spiffy high-end (at the time) 486 box couldn't
handle 400 users and 16 dialup lines, whereas three years later a handful
of sparc5's and 20's were handling 25,000+ users over 1200+ phone lines in
10 cities and four t1's worth of traffic.  we had 128mb sparc5's running
_sunos4_ handling 120+ interactive logins, 2-way ss20's handling 450,000
emails a day, dns for over 2000 domains (and uucp feeds for 350 of 'em!),
and we were one of the first isp's in the country to sustain over 1M web
hits/day on our main site, not including 120+ virtual servers... in
fairness, the main filestore was a netapp F330, built 'round a pentium-90,
and we were highly skeptical at first...

altogether, we're probably talking about a combined compute power that now
could be matched, on paper, by a couple of top-end p4's.  and i absolutely
defy you to spend less money to get the same amount of real world work
done.  those stinky old sparcs took a beating, 24x7, year in, year out.  
laugh all you want about specmarks or how a modern pc can "beat the crap"
out of a typical sun in some single-user application - and i'm sure it's
true.  at least until you spill out of cache and have to do, oh, y'know,
memory fetches to the same skinny pc133-based memory bus or burn cycles
servicing ide drives...

just a few months ago on the netapp list people were asking if three 2-way
or quad 700mhz p3 xeons backed by a netapp F760 cluster would be "enough
hardware" to manage 3,000 users running mail through an exchange server.  
i didn't know whether to laugh or cry...

closer to home:  256mb 600mhz p3 running the same solaris 7 install as the
128mb ultra 1/200e on my desk.  i bootstrapped the solaris environment
here on both platforms in three days - building solaris rpms for over 70
software packages.  the ultra 1 was running:  my windowing environment,
dns, nis and syslog for the internal network, all the jumpstart-related
stuff (rarp/bootparams/tftpd/nfs), the license server for the sun
compilers, plus the sparc compiles.  the p3 was running:  compiles.  the
u1 with half the physical memory and some poopy old 5400rpm scsi drives
consistently outperformed the p3 under load.  sure, in a simple
side-by-side comparison with nothing else active, the p3 would "kick the
crap" out of the old, 1st generation ultra.  but under serious load, the
old ultra kept up just fine, remaining responsive to my interactive
sessions while compiling in the background.  the p3 running just a plain
old console (no windowing) suffered noticeable pauses and lag.

i do run a handful of solaris x86 boxes - dell 2450s that i had to bang on
to get to jumpstart properly, and which use those idiotic "square hole"  
rack mounting systems, and a compaq dl380 which is satan's spawn - that
fucker took almost a DAY of my time to get properly installed, and i've
never, ever had a sparc install take more than an hour or two - from
shipping crate to fully-configured and in production.  sure, commoditizing
the hardware makes for lower up-front costs - at least until recently, 
when sun's schizophrenic product marketing people finally saw fit to drop 
prices out of the stratosphere for certain low- to mid-range models - but 
the thing x86 vendors _still_ struggle with is "fit and finish" - stuff 
that apple has down pat, and sun (or any of the high-end unix vendors) has 
had many years to perfect.  elegance matters.  design matters.  you get 
what you pay for.

or, as one old boss said:  "you can put a V8 in a vega, but it's still a 
vega."

all that said, i have softened my view over the last year or two toward
the x86 stuff, because it has been improving over time.  the unix world
cannot afford to ignore it.  it seems *insane* that for the cost of a
single 400mhz/8mb ultra-ii cpu module - parts that by now you'd think
would cost a few hundred $'s - still sell even at discount for thousands
of bucks.  the price pressure on the unix shops is intense, and it makes 
me sad to think that if sparc and power/ppc eventually go the way of hppa, 
alpha and mips, that we'll be left with zero choices - and at that point, 
i'll probably go do construction work and just give up on the stupid 
computer industry altogether.

in the meantime, the only intel-based box in my basement is a solid old
netapp F330 - a machine that proved itself worthy through two years of
providing an average (sustained!) rate of over 700 nfs ops/sec - more than
enough for my little network at home. :-)  but would i rather move up to 
an alpha-based filer someday?  you betcha.  the chrome bezels on the F630 
are dead sexeh!

ramble, ramble...

anyway, i thought the point of rescue was to talk about Neat Old Junk, so
who cares about the spiffy new stuff out there?  we can play with new toys
at work - on rescue i'm interested in hearing about the old gear.  all
things exist in their historical context, something the younger lads seem
to forget...

-- skeez