Lechner, David A. dlechner at mitre.org
Wed Jan 21 06:00:03 PST 2009

Just as a word of Thanks  !
Although we had many of the same intuitive views on the life expectancy of servers, but there is little real data on it....
(and in the military environment the amount of dirt, humidity, and temperature variation is worse than the average R&D lab...!)
But the discussion is highlighting several excellent issues and views that we thought might be discussed.
Greg's model of operating cost per FLOP vs. Capital Acquisition Cost per FLOP in particular was helpful - 

When I mentioned licenses, some of our systems are large and complex, with many real time modules as well as 
publish/subscribe middleware, Data  Base servers, and Human/Computer Interface products.
Normally those products seem to have a 3 year tempo of upgrades, but we run out of Server and OS "leniency" after 
6, 7, 8 or 9 years of aging, so we get maybe one major refresh loaded then have to go static and forgo the upgrades 
until the next hardware upgrade (which is what we're trying to decide as a resultant variable, not make it a given and an input!).

In military or intelligence programs there is not much of an option to co-locate!  ;-]
and our personnel are all our own, so admin costs tend to be high.  

Dave Lechner
MITRE Corp. 

More information about the Beowulf mailing list