[Beowulf] cloudy HPC?

Mark Hahn hahn at mcmaster.ca
Thu Jan 30 13:47:44 PST 2014


> Given prior art, I'm totally taking the fifth on this one ;-) hehehe!

well, I notice in that old thread, you did not explain how 
your EC2 config differs from the 66% worse-performing "internal kit".
it's a claim that falls into the same category as cold fusion.
(not impossible, but would be silly to accept at face value.)


> BTW: Chris Loken did an awesome job of explaining some of the .ca HPC
> infrastructure and made mention of some really good strategic
> investments that have been made in the past at our latest meeting down
> at the NCSA.  The slides are posted:
>
> http://www.ncsa.illinois.edu/Conferences/ARCC/agenda.html

I'm pleased he reported the messy or embarassing bits.  I prefer 
to describe Compute Canada as a dotted line around drawn around
about 20 largely independent sites.  some "sites" have a fairly
uniform user experience on multiple clusters, but most clusters 
are a single, arbitrarily diverging cluster, at a single location.

in any case, it's a huge mess now (which is not to imply we're
not getting useful work done.)

so, any substantive comments about heterogenous loads on IaaS?

thanks, Mark Hahn.



More information about the Beowulf mailing list