 |
|
DAILY NEWS AND INFORMATION FOR THE GLOBAL GRID COMMUNITY /
|
Special Features:
TurboWorx CEO: BUSINESS CONSIDERATIONS TO DRIVE GRID DESIGN
Jeff Augen, the president and CEO of TurboWorx Inc, recently agreed to an
interview with GRIDtoday. Here is what he had to say:
GRIDtoday: When and how was TurboWorx formed? What provided the impetus for
its formation?
JEFF AUGEN: TurboWorx was formed in mid 2000 in New Haven, Conn. The original
intellectual property was built on top of commercial software growing out of
more than two decades of parallel computing research that took place in the
Department of Computer Science at Yale University. TurboWorx's goal has always
been to simplify the design of high performance computing solutions by
integrating many applications into workflows that can be processed in
heterogeneous distributed computing environments. Individual components of the
workflows are automatically processed in parallel and a central hub recombines
the results and manages all related logistics. Bioinformatics and drug
discovery was an early focus for TurboWorx, but our solution is applicable to
many areas where high performance computing is a key differentiator.
Automotive, aerospace, geophysical, financial services, and other technical
endeavors are all business targets for the Turbo solution.
Gt: Who are TurboWorx's competitors, and why do you feel that TurboWorx offers
an edge over them?
JA: TurboWorx's most visible competitors are niche players that offer point
solutions in specific industries. Included are Engineous (engineering
workflows and Scitegic (computational chemistry). Other vendors such as
Platform Computing offer limited workflow solutions based on older batch
queuing methodologies. Additionally, many IT departments have launched
internal workflow and parallel computing efforts. Most are based on crude
approaches that involve the creation and management of hundreds of individual
scripts.
Today, we are not aware of any other vendor that provides a generalized
solution that works across all processing environments, can function in
heterogeneous clusters, is fully extensible, and can be maintained by end
users. The TurboWorx solution is based on a unique approach to scheduling that
operates somewhat like an electronic bulletin board. Individual worker
machines retrieve tasks from the board, process the tasks, and return their
answers to the hub where results are re-sequenced and combined according to
the rules of the workflow. The system is completely fault tolerant; for
example, any failure of a worker machine results in a task being re-posted for
processing. Other systems are based on a static batch queuing approach that
cannot respond to system failures. Moreover, such systems can not respond
rapidly to dynamic changes in the computing environment because a master
server must be aware of the states of all machines in the cluster before
distributing tasks for processing.
Gt: What products and services does TurboWorx expect to debut in the coming
year?
JA: We have been expanding our fault tolerant design to include automatic
failover at the hub level. We are also beginning to focus on transactional
workflows in addition to our current efforts focused mainly on computationally
intensive computing. Finally, we hope to expand into new markets by teaming
with ISVs in focused areas such as financial services.
Gt: What are the key challenges to the commercial use of Grid computing? How
do you think business will come to grips with them?
JA: Many of the issues surrounding standards for authentication and security
remain to be addressed. Additionally, computational Grids have been slow to
emerge because of the complexities associated with dividing applications into
manageable pieces that can be processed in a geographically dispersed
environment. Predicting the performance of such applications has been a
difficult problem, and much work remains before computational Grids become a
mainstream computing environment. Alternatively, storage Grids are technically
difficult to design and expensive to build. The bandwidth required to retrieve
large amounts of data across great distances has also become a barrier to the
design of such systems.
Finally, ownership and control of resources is also an issue for the Grid
computing community. Many institutions already purchase the most powerful
computers they can afford, and the availability of free processing cycles is
often vastly overstated.
Using software such as ours, many of these difficulties can be overcome in
Enterprise computing Grids, where a single organization manages all of the
computing resources.
Gt: What will Grid computing look like in five years?
JA: There will be many variations. Dedicated computational Grids will dominate
the academic science landscape and large, geographically dispersed storage
Grids will likely dominate the medical community where the need to share
patient records is a key driving force. The design of Grids will be driven by
business and financial considerations more than by technical limitations.
However, the utility approach is unlikely to ever take hold. Utility computing
lacks the flexibility to fit the tremendous variety of computing requirements
that exist across dozens of different industries. The same forces that ended
the age of mainframe computing are likely to prevent the rise of another
utility model.
Gt: What other things should our readers understand?
JA: The trend of replacing large powerful machines with clusters of
commodity-priced Intel-based computers will continue to accelerate.
Improvements in high speed switching, network bandwidth, and software for
managing distributed problems will ultimately drive large SMP computers into a
final end-of-life situation. However, unlike other transitions, the migration
to clusters of commodity-priced machines is more a software statement than a
hardware statement and improvements on the software side are rapidly becoming
the drivers of the migration. The trend is evident in the top 500 list of
supercomputers which is now dominated by clusters. Grids are nothing more than
geographically dispersed clusters built of heterogeneous systems.
Web services, utility computing, .NET, CPU harvesting and distributed
computing are just a few of the technologies that fall under the Grid
computing umbrella. Gt04 -- a premiere enterprise Grid computing conference
targeting industrial and commercial users -- will gather experts, and outline
strategies and road maps for Grid deployment. For more information, visit
www.gt04.com .
Grid computing is here!
|