Author: JT Smith
doesn’t mean their workstations can’t.
Silicon Valley-based TurboLinux Inc. (www.turbolinux.com)
is taking the wraps off a revolutionary clustering solution that enables organizations
to focus the combined processing capabilities of their networked workstations and
PCs for “Big Iron” supercomputing power at a very affordable price.
EnFuzion is a high-performance clustering solution. It supports
heterogeneous platforms (Linux, UNIX, Windows NT), works with existing
applications, requires little or no coding and is easy to use. “You
don’t need a Ph.D. in parallel programming, or even a programming
degree, to use EnFuzion,” said Klaus Fabian, senior product manager for
EnFuzion.
What is Clustering?
Clustering is the manipulation of networked PCs in such a way that they
behave like a single Herculean computer. Clusters are a very popular
strategy for implementing parallel processing applications and enable
organizations to leverage the investment already made in PCs and
workstations. Clusters are also scalable. Need more computing power?
Just add more PCs to the network.
Often referred to as a “single system illusion” because the network
appears as a single system from the outside, a cluster prevents a single
point of failure. If one or more machines in the cluster fails, the
system continues to perform.
Before EnFuzion, organizations with high-performance computing needs
turned to expensive supercomputers that required constant maintenance
and expensive parallel programming. Why not use what you’ve already got?
Your CFO will love you.
EnFuzion employs a technique called Parameterized Simulation. Parametric
executions are jobs where the same task is run repeatedly but under
several different scenarios. It is typically used for modeling
complicated problems, from portfolio risk management at an investment
bank to determining gene sequences at a biotechnology company.
EnFuzion is completely platform independent, which means it can be used
with Windows NT and all UNIX-based systems including: Solaris, HP-UX,
IBM AIX, SGI Irix and Tru64. It is also fault-tolerant, so if one or
more machines in the cluster fails or is turned off, the job is
rescheduled on the remaining PCs or workstations.
And it’s secure. EnFuzion is encryption-enabled to bolster security for
root-to-node communications, authorization of trusted hosts and
executables, and user-defined decryption primitives. In addition,
EnFuzion’s application program interface (API) allows third-party
companies to seamlessly and transparently integrate EnFuzion into their
products.
Who Should Use EnFuzion?
“Any organization with a daunting computing task that takes extremely
long to complete on a single PC can get tremendous advantage from
EnFuzion,” Fabian said. “EnFuzion can speed application performance by a
factor of 100 or more.”
Companies specializing in image rendering (animation, simulations,
structure modeling, etc.) rely on EnFuzion for low-cost,
high-performance supercomputing power. EnFuzion has been used in the
areas of aerospace, bioinformatics, gene modeling, finance, engineering,
heath care, animation, mathematics/statistics, physics, telecommunications
— and more.
In the Internet realm, EnFuzion opens new opportunities for applications
that require massive computing muscle. It is ideally suited for data
mining, Web site content generation, report generation and software
development.
In the financial industry, JP Morgan and AMP Asset Management use
EnFuzion for quantitative processes involving statistical data such as
scenario analysis, used in risk assessment for portfolio and asset
management. EnFuzion enables financial services companies to minimize
their customers’ risk while maximizing returns.
What EnFuzion Users Are Saying
“Suddenly there was an enormous growth spurt in the amount of work we
were able to perform,” said Dr. Alen Varsek of AMP, Australia’s largest
insurance firm and one of that country’s leading banks. “(EnFuzion) made
it possible to multiply the amount of work done by a factor of almost 100.”
In related news, researchers at New York’s Rockefeller University said
they’ve stumbled across an amazing fact: a 16-node TurboLinux EnFuzion
cluster of Intel servers performed 20 percent faster than an SGI Origin
2000 supercomputer.
The five-person research group, which conducts large-scale comparative
protein structure modeling for the Human Genome project, turned to
EnFuzion as a cost-effective solution to add computing power to support
its growing body of work.
“We paid $650,000 for a 20-processor Origin 2000 from SGI about two
years ago. We also paid about $30,000 for a 16-node Pentium II cluster,”
said Professor Andrej Sali, head of the research group. “Given our
benchmarks, the EnFuzion cluster is about 20 percent faster than the
Origin 2000, at 20 times smaller cost.”
While EnFuzion shares the clustering space with several companies, none
of them are better suited for the enterprise than EnFuzion.
Unlike the popular clustering solution Beowulf, which requires a
dedicated cluster of Linux servers, EnFuzion allows you to utilize
existing workstations and applications. Your previous investments are
protected.
EnFuzion can be set to run at night, during lunch or anytime employees
are not using their workstations. For example, EnFuzion can be set to
run when a PC’s screen saver kicks in, thus assuring that its user is
not inconvenienced.
“Think of the efficiency — all the machines that are not used during
lunch, meetings or overnight can be used for high-performance computing
at no cost, a practice referred to as ‘cycle harvesting,'” Fabian said.
How Does EnFuzion Work?
When used with your application, EnFuzion automatically generates a
graphical user interface (GUI) for your root machine. No programming is
required! The root machine then distributes the jobs across the network
and collects the results. Awesome!
Users have the choice of using a text editor or the GUI to generate the
script that allows EnFuzion to distribute the job. Just punch in the
parameters you want to run and let EnFuzion do the rest. Once the
computation is distributed to your networked PCs, you can use the root
computer to monitor its progress via the GUI.
The root machine copies all of the input files to the client machines,
the clients then substitute the parameter and produce output files that
are ultimately sent back to the root machine
