Buried beneath the bland verbiage announcing Microsoft's Technical
Computing Initiative on Monday is some really exciting stuff. As
Bill Hilf, Redmond's general manager of technical computing,
explained, Microsoft is bringing burst- and cluster-computing
capability to its Windows Azure platform. The upshot is that anyone
will be able to access HPC in the cloud. HPC stands for
High-Performance Computing. That's the politically correct acronym
for what we used to call supercomputing. Microsoft itself has long
offered Windows HPC Server as its operating system in support of
highly parallel and cluster-computing systems.
The new initiative doesn't focus on Windows HPC Server, per se,
which was what I'd been expecting to hear when Microsoft called to
corral me for a phone call about the announcement. Instead, it's
about enabling users to access compute cycles -- lots of them, as
in, HPC-class performance -- via its Azure cloud computing
service.
As Microsoft laid it out in an e-mail, there are three specific
areas of focus:
Cloud: Bringing technical computing power to
scientists, engineers and analysts through cloud computing to help
ensure processing resources are available whenever they are
needed—reliably, consistently and quickly. Supercomputing
work may emerge as a “killer app” for the cloud.
Easier, consistent parallel programming:
Delivering new tools that will help simplify parallel development
from the desktop to the cluster to the cloud.
Powerful new tools: Developing powerful,
easy-to-use technical computing tools that will help significantly
speed discovery. This includes working with customers and industry
partners on innovative solutions that will bring our technical
computing vision to life.
Trust me that this is indeed powerful stuff. As Hilf told me in a
brief interview: "We've been doing HPC Server and selling
infrastructure and tools into supercomputing, but there's really a
much broader opportunity. What we're trying to do is democratize
supercomputing, to take a capability that's been available to a
fraction of users to the broader scientific computing."
In some sense, what this will do is open up what can be
characterized as "supercomputing light" to a very broad group of
users. There will be two main classes of customers who take
advantage of this HPC-class access. The first will be those who
need to augment their available capacity with access to additional,
on-demand "burst" compute capacity.
The second group, according to Hilf, "is the broad base of users
further down the pyramid. People who will never have a cluster, but
may want to have the capability exposed to them in the
desktop."
OK, so when you deconstruct this stuff, you have to ask yourself
where one draws the line between true HPC and just needing a bunch
of additional capacity. If you look at it that way, it's not a
stretch to say that perhaps many of the users of this service won't
be traditional HPC customers, but rather (as Hilf admitted) users
lower down the rung who need a little extra umph.
OTOH, as Hilf put it: "We have a lot of traditional HPC customers
who are looking at the cloud as a cost savings."
Which makes perfect sense. Whether this will make such traditional
high-end users more like to postpone purchase of a new 4P server or
cluster in favor of additional cloud capacity is another issue
entirely, one which will be interesting to follow in the months to
come