[Beowulf] Utility Supercomputing...

Mark Hahn hahn at mcmaster.ca
Sun Mar 3 09:31:23 PST 2013


>> I'm guessing the answer is that your market is mainly people who are too
>> self-important (anything health- or finance-related) to do it themselves
>> (or join a coop/consortium), or else people just getting their toes wet.

James, FWIW, I regret that this was unnecessarily personal.

my thought at the time was about how, across various computing fields, there
is variance in how willing participants are to "get hands dirty" with
computing.  some are very willing to pay quite a lot for packaged,
no-hands-dirty services.  a pharma company, for instance, might self-evaluate
that it shouldn't consider building large HPC facilities because it simply
doesn't want the bother.  (this might be justified in terms of managing risk:
they might save money by cutting out Amazon's profit margin, but there's some
risk that they would not do as well as Amazon.  such a company might consider
it unwise to take on this additional risk - or that their "risk budget" is
better spent elsewhere...)

but the particular Cyclecomputing case study is particularly interesting
because the effective cost is very near, perhaps a bit better, than what a 
large-scale, competent HPC organization could do by building from scratch.
kudos to Cyclecomputing and Amazon!

this is somewhat surprising to me, but less so when I notice that the picture
is quite different if you use Amazon's on-demand pricing schedule:
for the m1.xlarge instance, the current spot price is $0.052 per hour
versus 0.480 for on-demand.

hence the question about dynamics of spot pricing.

regards, mark hahn.
(speaking, of course, for myself alone)


More information about the Beowulf mailing list