[conspire] Another low-power server

Rick Moen rick at linuxmafia.com
Wed Oct 7 17:24:47 PDT 2009


Quoting Luke S Crawford (lsc at prgmr.com):

> wow. sweet.  I was pretty excited about the atom boards, too...
> but I was so dissapointed when I got them and they put this decently low-power
> cpu on a board with a chipset that draws 50w (meaning the 2gib max ram
> atoms draw about 20% less power than a 8GiB max core2duo that supports
> unbuffered ecc.  useless.)

Haha.  Man, having worked in the Linux hardware industry, I'm still
continually amazed at suckage where you don't expect it.  Thanks very
much for dropping the other shoe about Atom-based boards.  That's pretty 
pathetic, indeed.

If I see new Atom-based server boards, I'll make a point of remembering
your experience, and test actual power draw.  Fortunately, Casa Moen
now has a Kill-A-Watt tester (inline watt monitor for AC), kindly
acquired by my mother-in-law Cheryl, after Mark Weisler brought up the
idea.

Back in VA Linux Systems days, one of the final models the firm
introduced was the movel 1124, one of the very first 1U rackmount 
machines to use dual Athlons (using AMD's 760MP chipset on a Tyan
Thunder K7 motherboard).  It was a very fast but somewhat bizarre
system (required registered ECC DDR266 SDRAM).  Those power-sucking
Athlons ran so hot that just designing the airflow through the box was a
major engineering problem, and VA Linux had several key patents on the
baffle design.  Just as it was about to hit market, though, Larry
Augustin pulled the firm completely out of the hardware business (and
turned it into VA Software Corp.).  The model 1124, patents and all, 
got sold cheaply to APPRO.[1]

Anyhow, I remember that one customer wanted rackmount clusters of those
puppies, and it turned out that a full-sized floor-mounted rack of 45(?)
of those units drew more power than could be delivered by any of their
electrical circuits.

And the thing is, hardly anyone seemed to care.  The physical inability
to power a full-sized rack of them was widely seen as an unfortunate
inconvenience, rather than a sign that burning that much power might be
A Bad Thing that might be worth avoiding.

I haven't yet used the Kill-A-Watt on my VA Linux model 2230, but
that'll be an interesting baseline, as it was sort of a bog-standard
PIII box (Intel L440GX+ "Lancewood" motherboard, cruddy proprietary PSU, 
up to four 100 or 133 MHz unbuffered or registered SDRAM DIMMs).  Mine 
has three circa-2001 SCSI drives.  So, in the power-consumption
department, it wins in the CPU area, and (relative to newer SAS and SATA
drives) loses on the HDs.


I'm certainly interested in these ARM--based options, especially the
ones able to support a reasonable amount of RAM, and able to mount a
mirrored pair of hi-speed or "SuperSpeed" USB drives, so you're not 
limited to Flash.  However, doing without floating point (thus, painful
SSL/ssh) seems a high price to pay.  

Yes, I know that most people who say "SOHO server" really mean just
"gateway appliance", but I like to have a real server.  (Whatever
machine's in front of me gets treated as just a Web browser and xterm
appliance, mostly.)


> These would also be good for things like kerberos servers.  

With ARM CPUs and thus no FPUs?  Really?

[1] There's a good article about it at AnandTech.





More information about the conspire mailing list