RE: DS20L

["Peter Petrakis" <voodoo alphadriven org>]

> I have set these up to run customer benchmarks twice now.
> Once a long time ago, and once just two weeks ago.
> Both times, when room temperature rose to about 25C after a few hours I had
> stalled machines.

So you where running hardware that is known to dissipate significant heat in
a machine room that had inadequate cooling? That's 77 F you're talking about.
That is uncomfortably warm for a machine room or any room for that matter, You
should never be past 70F, 65F ideally. At API, Our labs had 'two' full air
conditioners and even if one failed we wouldn't get the lab hotter than 74F
tops. It was a cool 63-67F when they where both operational. Cold enough that if
you worked in there for
a significant period of time you either wore a long sleeved shirt (me) or
a vest.

You think that cooling is overkill? At it's peak that room housed 64 running
UP2000s
(EV67 too), 20+ Intel 2/4/8 way servers, 12+ SPARC servers, 16-20 UP2000 (RMA
testing), and a half dozen UP2K/DP264 servers. At that time, The room wasn't
hotter than 72-73F. That 64 node cluster was only temporary though even when it
was up 'we' didn't see machines failing left and right due to over-temp. Your
customers will buy multiple 15K  servers  but wont spend the 6-8K for another AC
to cool them?


> On the last go I ended up putting a room fan in front to blow over them,
> and it "fixed" the problem.
> I think that in a 20C air-conditioned room they will be fine, but at 25C
> they get "iffy".

Like they're spec'd for :-).

> >There are several people on this list that worked very hard on the CS20
> >project and I for one can say that I don't appreciate people making false
> >statements about it. Was it perfect? No, of course not, no product is
> >perfect, they all have their faults, but in the case of the CS20, over
> >heating was not one of them.

Agreed.


> I also believe that they could have been a big success, if they were
> marketed properly.

Not touching that, someone from marketing may wish to comment :-).

> But when they were released they were a bear to configure, were offered
> with no warranty,

If I recall correctly, The only ones that came with no warranty where the
prototypes that VARS requested 'advanced' access to. All production CS20's had 1
year warranties and also if I recall correctly, you could 'buy' a warranty if
you had one but the price was very high.

> were offered at a 15% price premium over the cost of
> fully configuring a UP2000+ based machine

I don't pay attention to pricing though even 'if' that where the case. If you
could
triple the existing density of a cluster that had 3U CH320's (Those chassis cost
1K alone). You don't think that premium  is worth the gain in computer power
never mind the reduced maintenance?

> and could not run TRU64.

Absolutely incorrect. Just because it had an onboard symbios chipset doesn't
mean
it automatically runs Tru64. 1) We didn't pay for the licensed BIOS 2) Tru64
didn't
even support that sym 1010 chipset and wasn't for some time to come. 3) We where
a LINUX company :-). UP2000/+ onboard SCSI didn't support Tru64 either but BOTH
CS20 and 2K had IDE which was supported, What are you getting at?

> We consistently saw customer disapproval, from people who were still buying
> UP2000+ based machines.

Then those customers where looking at the CS20 for the wrong reasons. Unless
you are concerned with GIGIFLOP/SQ FT then the CS20 is 'probably' not for you.


Finally. If there are so many things wrong with the CS20 why is it that I
cannot recall a single user 'ever' posting to this or any other list that
they where having a problem with their hardware or software? Considering
anyone who is an authority on CS20 is at least on this list.

Peter
--
www.linuxalpha.org	www.alphadriven.org
Peter Petrakis         Warrior/Engineer        voodoo@alphadriven.org
"<BLAM!!> Who the hell are you!? Name's Ash <click clock> Housewares..."