November 21, 2005
Sun's Bjorn Andersson Gives Insight into Big Wins
GRIDtoday editor Derrick Harris spoke with Bjorn Andersson, Sun
Microsystems' director of HPC and Grid computing, about what's going on
with the company in terms of its Grid and HPC initiatives. Andersson
discusses Sun's big customer wins for its Sun Grid and the state of the
Grid market, and also hits upon some the company's HPC on-goings, such
as its largest HPC win to-date with the Tokyo Institute of Technology.
GRIDtoday: How are things going with Sun's Grid program?
BJORN ANDERSSON: We have seen significant momentum in our program. Most recently, we
announced the biggest purchase yet of compute cycles from the Sun Grid.
VCC bought 1 million CPU hours for use in the oil and gas industry.
We're very excited to go live with such a big deal as one of the first
publicly announced customers.
Gt: Can we expect to see more Sun Grid customers announced in the weeks and months to come?
We expect this to be just the beginning of several wins with the Sun
Grid. This win may have been one of the larger wins to date however;
going forward you will see a mix of larger and smaller customers using
the Sun Grid.
Gt: Are companies ready to fully embrace utility computing? If not, when do you think it will reach its peak of ubiquity?
I would say it's a matter of how comfortable a company is with the Grid
technology and how mature they are in establishing the way of working
that Grid. Going forward the obstacles will not be so much on the
technology side, but it will be more a matter of corporate culture and
how you set up service level agreements between participants in an
internal Grid infrastructure versus setting up policies for when you
utilize a public Grid. We as an industry will get through this, but
it's a learning curve. Many companies have embraced Grid technology in
their internal infrastructure today. The volume ramp in terms of usage
is well on its way. We're now also starting to see companies use the
public Grid infrastructure of the Sun Grid. We expect the volume ramp
to follow. I believe we're not anywhere close to a peak in usage of
Grid technologies internally at companies and organizations. Utility
computing can be seen as building on that foundation, so it's even
harder to try to predict a peak for utility computing.
Gt: What about "traditional" Grid computing? Given the current state of the
market, when do you think Grid will become as ubiquitous as the
community would like?
When you say "traditional" Grid computing, I think of where Grid
computing started in terms of what problems it was used for --
traditional scientific and technical computing. We're at Supercomputing
2005 this week, so allow me to look at this from a high-performance
computing standpoint. In this area, we have seen a very pronounced trend
over the last several years of more and more computing being done on
clusters rather than on big monolithic machines. We're also seeing a
blurring of the lines between the traditional definitions of Grid and
clusters. The whole movement toward cluster computing is also enabling
Grids to be more commonplace and used where applicable, while
establishing some of the processes and tools needed along the way. What
really determines what customers use is the nature of the problem
they're trying to solve and what requirements are on latency,
bandwidth, memory size, etc. I believe we'll continue to see a diverse
environment in HPC that will integrate both clusters and Grids as it
applies to the computing requirements.
What is Sun's take on the current discussions around the need to adopt
quality standards before enterprise adoption will really take off?
has always worked for standardization of interfaces and continues to do
so also in the Grid space. To really reach a level of ubiquity that we
want, we need a solid foundation of standards. Then it's up to all
vendors to compete on the implementation of these standards.
Gt: Can we expect any big announcements at SC'05?
HPC is one of the key focus areas for Sun and we're really investing to
accelerate our growth in this market. Just last week we opened our
Solution Center for HPC in Hillsboro, Ore. It's a 10 Teraflop capable
benchmark and testing facility designed to help customers find their
HPC solutions and for us to help remove risk and uncertainties in HPC
This week, we're also announcing Sun's largest HPC win to-date -- that
the Tokyo Institute of Technology (Tokyo Tech) has purchased Sun Fire
x64 servers to build Japan's largest supercomputer. Tokyo Tech's project
will use Sun Fire x64 servers with 10,480 AMD Opteron processor cores,
Sun storage technologies and NEC's integration expertise to build the
Tokyo Tech supercomputer. The system will help provide researchers with compute power for a wide range
of scientific applications, such as analysis of the complex molecular
structure of proteins, simulated blood flow diagnosis in human brains, and
clarification of the generation mechanism of Earth and planetary magnetic
field. We have a big presence at the show and I would recommend anyone to come
by our booth (#1416) to take a look at what we're doing in
Gt: How do shows like SC help to further the cause of Grid computing? Does
it help that new Grid technologies and projects are being presented to
individuals spanning the entire HPC spectrum?
Shows like Supercomputing are great as focal points to bring key
individuals together for sharing ideas and provides a unique showcase
of what's available in products and technologies. It really helps the
whole community to be exposed to this on a regular basis.
Gt: Finally, how do you feel about the SC shows? Is there anything you're especially looking forward to at this year's event?
The Supercomputing shows are always full of exciting products,
technology demonstrations and events. It's hard to get the time to see
it all. Just to give you a couple of examples, in the Sun booth we'll
have a demo of OpenIB connected to the InfiniBand fabric at the show.
Related to InfiniBand we're also participating in Storcloud with data
management products from former StorageTek, which Sun acquired in
August of this year. We also will have a remote visualization
technology demo in the booth. The list goes on and on, but you can see
some trends. InfiniBand seems to have cleared the hurdle to be a key
low latency interconnect. Open source and free software continues to be
important, we expect to see a lot of interest in OpenSolaris and the
fact that the Sun Studio developer tools now are available with a no
cost user license. Also, if you're interested in what's around the
corner, take the opportunity to learn about the DARPA/HPCS project.
Sun, as a key participant in that project, is showcasing some
technologies, including a new programming language called Fortress.
Gt: Is there anything else you'd like to add?
ANDERSSON: From Sun we believe that we can make HPC much more attainable and
practical. As we do that, it is becoming a tool for many more usages
and starts to become a mission critical element for commercial
enterprises -- whether it's how you do crash simulation on cars, how you
do analysis to figure out how to extract more oil out of the ground or
finding that next drug that will cure a disease. Once it's being used
for the core part of the value a company delivers, it's by definition
mission critical for that company. Sun is applying engineering
disciplines from high-end data center designs and our ability to
package solutions together as a way to make it much easier and reliable
to build out an HPC infrastructure.