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

[Alex Chekholko]

On Fri, 16 Jan 2009 09:47:51 -0800
Greg Lindahl <lindahl at pbm.com> wrote:

> On Thu, Jan 15, 2009 at 04:26:00PM -0500, Lechner, David A. wrote:
> 
> > Sometimes the newer technology uses less power and is cheaper to
> > operate....(anyone ever create a KW/MFLOP vs. Time curve?  has that
> > really gone down? )
> 
> I'd think that you would have looked that up before you posted! I
> don't have the curve, but it's falling rapidly enough that it's
> obvious that an older cluster can be replaced with a mch smaller, new
> cluster of equal performance for the cost of power. How old?  I don't
> know, but I'd guess it's less than 6 years. And anyone who rents
> machine room space probably replaces in ~ 3.
> 
> Since many clusters are funded by one-off funding sources, and often
> the cluster buyer doesn't have to directly pay for power or rent,
> you'll probably find that actual replacement dates don't make any
> sense...  not only for academic sites, but commercial ones, too.
> 

All three of those are in fact the case in my environment here.  Floor
space and power are free, and purchases are made from grant money for
capital expenditures.  The grants are not regular.

So, like many other cluster admins, we keep the hw running for as long
as possible, since the only direct cost to us is admin time.  We retire
it when users no longer even want to use it, because it's not worth
_their_ time.

Most recently, summer 2008, we retired a ~100 node cluster from ~2002.
The nodes were dual P3 1.1GHz with 2GB RAM and one 18GB disk and
100mbit Ethernet.  

The newest cluster that "replaces" that one is 64 nodes of IBM x3550,
dual Xeon 5160, 8GB RAM, 2 x 160GB SATA, gigE, purchased in summer 2007.
-- 
Alex Chekholko  chekh at pcbi.upenn.edu