[Beowulf] RE: Capitalization Rates - How often should you replace a cluster? (resent - 1st sending wasn't posted ).

Robert G. Brown rgb at phy.duke.edu
Wed Jan 21 12:10:00 PST 2009


On Wed, 21 Jan 2009, Lux, James P wrote:

> Without going into a big research project.. Here in Southern California,
> rents are about $1/sq ft/month for cheap industrial space. Electricity is
> about $0.20/kWh (maybe for a big bulk consumer, it gets down to the
> $0.10/kWh range)
>
> Figure a node draws 400w, and you can stack 20 nodes in a rack, which
> occupies 10 square feet (need to have room to stand in front of the rack,
> etc.)
>
> So each node is costing $144/20= $7.20/yr in rent and 0.4*8000*0.2 = $640/yr
> in electricity.
>
> The rent is tiny compared to utilities.  $2K/node seems plausible (3*
> utility cost), but that doesn't include any of the infrastructure costs
> (racks, cables, installation, copies of Windows Server edition, etc.).  If
> the real power draw is more like 200W, and you don't run them 24/7, then
> you're getting down to the $100/yr/node costs..

But often you can't get "cheap" industrial space.  You need excellent
climate control and server-room class wiring.  Otherwise you can add the
cost of outfitting it with a couple of Lieberts, a harmonic-balancing
transformer, and a rewiring job (amortized however).  Call it a few
hundred thousand over N years?

In server-class space, this is all provided but I doubt you'll get away
as cheaply as all that -- after all each node costs THEM an amortized
chunk of that same set of Lieberts and transformers.  Also, may server
rooms offer "management" services -- somebody onsite to e.g. reboot your
nodes or babysite the power and AC while you sleep.  This too adds cost.

So I think your estimate above is a lowball -- not implausibly low for
space that is "free" but you pay for power, but I'm guessing low by a
500-100% for a space you can just put your nodes in and not worry about
them (adequate space, security -- you've got tens to hundreds of
thousands of dollars of hardware in there and may have valuable data as
well, adequate AC and power, and maybe a systems admin qualified
babysitter.

    rgb

Robert G. Brown                            Phone(cell): 1-919-280-8443
Duke University Physics Dept, Box 90305
Durham, N.C. 27708-0305
Web: http://www.phy.duke.edu/~rgb
Book of Lilith Website: http://www.phy.duke.edu/~rgb/Lilith/Lilith.php
Lulu Bookstore: http://stores.lulu.com/store.php?fAcctID=877977



More information about the Beowulf mailing list