Special Features:
SGI EMBRACES THE GRID
By John
Hurley, Editor-at-Large
How profound is SGI's commitment to grid technology?
GRIDtoday's Editor-at-Large John Hurley and Mike Bernhardt, Contributing
Editor interviewed Walter Stewart, Director, Business Development-Canada
Global Co-ordinator, SGI Grid Strategy; Addison Snell, High-End Server
Marketing Manager Advanced Systems Modeling; and John Watson, PR Programs
Manager, Sciences & Energy, to obtain a meaningful perspective.
GRIDtoday: What is SGI's position on grid computing?
STEWART: SGI is focused on the grid, as it touches creative and technical
users. We believe that SGI can play a vital role in academia, as well as, in
business enterprises such as manufacturing (design-side) and gas/oil
(exploration-side) companies.
These industries are interested in using the Grid to share information and
to enable collaborative decision-making, visually, in multiple locations.
Current grid projects include: UK cosmology grid led by Prof. Stephen Hawking,
Reality Grid also in the UK, RAVE - a visualization grid for Welsh
universities, among others in North America, Australia, Japan, and Europe.
BERNHARDT: Why SGI and not someone else?
STEWART: Long ago, SGI met the challenges of bandwidth within machines. Now
we work together with the grid community to bring the advantages of SGI's
internal system bandwidth to the grid. SGI always has focused on the movement
of enormous data sets.
SNELL: The real distinction is in the way that SGI has treated metadata,
including the ability to use a Visual Area Network (VAN) to send pictures of
data, instead of entire data sets. SGI's CXFS filesystem, for instance, allows
communication via the picture of data, not the data itself, which unlocks the
ability to move large data sets over long distances without moving the data
itself, with high performance characteristics still intact.
BERNHARDT: How long has CXFS been around? SNELL: XFS, for which CXFS is
based, has been around for almost 10 years. CXFS is proprietary to SGI, but
XFS has been put into the public domain and has been adopted as an industry
standard.
GRIDtoday: Do you have ideas of any specific applications that might be
enhanced by these advances?
STEWART: The film industry is an immediate example because of the need to
quickly move data between different areas. Historically, moving data over
networks had stressed the networks to the point that it was actually easier to
instead move the machines among departments.
The automotive and airline industries also come to mind as multiple groups
around the world are asked to look at the same data at the same time
especially in the areas of design. Aircraft and automotive industries are not
designed in the same location, but rely upon the quick availability of current
and relevant information.
SNELL: In addition, we see that traditional problems associated with
versioning may be dramatically reduced.
GRIDtoday: How do you perceive being able to deal with the different
architectures, OSs, etc. associated with grids?
STEWART: We actually believe that this kind of problem will be solved by
organizations like the Global Grid Forum (GGF). We are interested in bringing
a unique capacity to the grid and firmly believe that those who are "Trying to
own it, will fail."
We are ensuring that nothing that we develop is functionally incompatible
with the open source effort and engage as much as we can with the working
groups of GGF.
SNELL: Some existing technologies will help solve issues like security of
the grid. We believe that a lot of the security issues can be solved through
traditional file permissions management at an OS system administration level
because you have different classes of users that are allowed to read, write
and copy data, etc. If you can get access without having the real data, then
better control can be assured.
GRIDtoday: Other Potential users?
SNELL: When we get to the point that we can release CXFS across WANs, it
would
then be possible to have multiple hospitals across the continent sharing the
same data. This establishes the potential to have a variety of different
"grids" associated with different departments, for example: radiology grids
where radiology depts. of key hospitals are collaborating to further some
dimension of research that is separate from cardiovascular grids.
We believe the largest potential revolution is in the workflow in medical
research and in the treatment of patient data by hospitals. It augments rather
than replaces existing initiatives, where companies have staked out very
sizable businesses in the hospital data management chain.
GRIDtoday: How has SGI positioned itself relative to the grid and enabling
Grid technologies?
STEWART: People issue absurd statements such as "We are the Grid". We
believe
that the Grid should be built, not bought. Discussions should be focused on
"who" needs to access "what" technologies.
Our position in grid computing is the same as in computing, we are
interested
in the capability segment--providing a tool so that people interested in high
productivity grid computing can get their job done. The ability to do
visualization over a grid and to have single file systems used between grid
partners is unique to SGI.
GRIDtoday: What do you see as the greatest potential stumbling blocks for
grid
computing?
STEWART: The greatest hurdle has been to get people to realize what is
already
possible and get them thinking beyond it. If all grid computing accomplishes
is that which can already be done cheaper, then it will fail. Grid computing
has the capacity to provide access to resources to solve problems of a
completely higher order.
People need to dream about those problems and look to build the
infrastructure
that can solve them. grids can enable the widespread asking of whole new
questions.
STEWART: In addition, a significant number of leading- edge academics push
us to provide more. However, a major problem with the evolving of this
technology is the funding of grid activities in the academic community. If one
adopts the position that research computing is a service that an academic
institution accesses when needed, then the problem is how does one get funding
to buy this service.
An investigator can get funding to buy a small server, but the researcher
may be better served with a budget to access additional cycles. Unserviceable
distinctions between operating and capital dollars makes it very difficult to
pay for compute on demand. This problem is true not only for the academic
communities but also for startup companies.
BERNHARDT: If GRIDtoday were able to focus over the next couple of months
on the one theme or trend that would really add to the challenge of educating
the public about Grid computing, what would it be?
STEWART: People must stop thinking of the grid as a particular
implementation and start thinking of it in the spirit that it was intended,
i.e., distributed access to computing resources which does not necessarily
mean distributed hardware. Do you want to distribute the computer itself, or
do you want to distribute bits? And what is the most efficient way to do
that?
WATSON: Visualization on the grid is where SGI believes technology is next
headed and SGI expects to be leading the way.
STEWART: Computing will be totally interactive and collaborative, which
shortens the workflow on what you are already doing. Beyond that it gives
access to new levels of capabilities that are beyond what currently exists,
resulting in possible processing and capability breakthroughs.
GRIDtoday and BERNHARDT: We would like to thank you very much for the
opportunity to discuss with you the new breakthroughs that you have noted. We
look further to additional news of further accomplishments in the very near
future.
|
