[Beowulf] Utility Supercomputing...
ajdecon at ajdecon.org
Mon Mar 4 08:41:14 PST 2013
-----BEGIN PGP SIGNED MESSAGE-----
> again, if you have a bursty load, you want to use a shared facility,
> of which, EC2 spot instances is just one example.
When talking of using a shared facility: The advantage I've
sometimes seen for "cloud" over the traditional "get time on a
shared cluster" is that you get to define a lot more of the
IT policy and configuration. You get to choose your OS, scheduler,
compute node configuration, some subset of the security policy which is
compatible with "cloud", etc. You also get your own IT folks managing
Compare to a shared cluster, which already has all this in place and is
unlikely to change it. Most folks who make an effort can probably work
with an existing system (as BOFHs are not so common in HPC, in my experience),
but there are times when it makes sense. There are also those who are just
not willing to make the effort of working with another organization's system.
Security policies are often cited.
I've seen interest in cloud HPC a few times with commercial entities which
have a limited amount of "big compute" to do, but don't already have access
to a shared HPC facility, and find the process of interacting with academic
HPC too painful for some reason. Sometimes they're also already using the
cloud for other projects, so it's comfortable.
(Also there's the Windows HPC folks, who are unlikely to find a shared
facility in existence for them. But that's a different horror story...
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (Darwin)
-----END PGP SIGNATURE-----
More information about the Beowulf