[Beowulf] small file systems
James.P.Lux at jpl.nasa.gov
Tue Aug 28 12:11:13 PDT 2007
At 11:17 AM 8/28/2007, Peter St. John wrote:
>Well that's fair of course, I should have said that 5GHz is an
>unachiveable upper bound for a sufficiently distributable
>application. If I wanted to add some huge number of integers, which
>can be generated by some simple formula in ranges at each node, then
>10 500MHz machines generating ranges and then computing subtotals
>would be almost as fast as one 5GHz machine, but for any real apps
>we would not see such a gain. I'm sure I"m overoptimistic about the
>"embarassing parallelism" of my own app, but pretty much it's pretty
>embarassing I'm pretty sure :-) Got's to see.
And, of course, if you just want to get your picture taken for the
local newspaper draping your arm over a supercomputer that you built,
it's more than sufficient. "local eccentric builds supercomputer in kitchen"
(or, if you have a suitable young child available... "9 year old builds....")
My wife considers this much more acceptable than, say, making a
neutron source by building a fusor (http://www.fusor.net), smelting
ore and making your own steel
(http://www.lindsaybks.com/bks9/fmetal/index.html), high powered
rocketry, or HV experimentation (http://www.pupman.com/). All these
other highly interesting activities create fears of her coming home
from work and finding the street blocked off with guys in bunny suits
carrying things out of our garage.
James Lux, P.E.
Spacecraft Radio Frequency Subsystems Group
Flight Communications Systems Section
Jet Propulsion Laboratory, Mail Stop 161-213
4800 Oak Grove Drive
Pasadena CA 91109
More information about the Beowulf