 |
|
DAILY NEWS AND INFORMATION
FOR THE GLOBAL GRID COMMUNITY / AUGUST 25, 2003; VOL. 2 NO. 34
|
Applications:
GRID PLAYER POWERLLEL WINS
FINANCIAL ISV AS OEM PARTNER by Rachel Chalmers for the451.com
Axiom Software Laboratories, a vendor of risk management and enterprise
data
software, has chosen software from Powerllel to integrate with its RiskMonitor
application. Axiom sells to big banks, asset managers and energy companies.
Its customers will now be able to use Powerllel's technology to run
RiskMonitor as a parallel, distributed application – in other words, across
computing grids.
Uniquely among grid players, Powerllel offers a software abstraction layer
at
the algorithm level. In doing so, it complements products from other grid
companies – particularly those from Platform Computing, which has taken the
unprecedented step of making an equity investment in the firm. Powerllel has
its roots in Wall Street and is particularly strong in the financial services
sector through ISV deals like this one with Axiom. Its ambitions are not
confined to banking, however. Powerllel hopes to take its parallel computing
software to government, life sciences and manufacturing customers as well.
Impact Assessment
The message In its 12-year career, Platform Computing has made just one
equity
investment: in algorithm-layer abstraction player Powerllel. That gamble may
be about to pay off.
Competitive landscape Although it's compared with DataSynapse's
LiveCluster,
Platform's LSF and Sun's Grid Engine, Powerllel's software actually
complements all three.
The451 Assessment
With an A-round of venture financing in the works, the
company only needs to score a few more ISVs in banking and life sciences to
make itself a highly attractive acquisition target for Platform or Sun.
Context Abraham Gulko, formerly of HSBC and Open Link Financial, and David
Mellor, a veteran of Parallel Computer Systems, founded Powerllel in 2000. As
an application developer supporting trading desks and bank back offices, Gulko
identified an opportunity to speed applications without diluting the power of
the algorithms on which they are based.
At present, Powerllel executives say, users of compute-intensive
applications
face a trade-off between speed and accuracy. Speeding complex financial
algorithms isn't as easy as throwing in another expensive Sun box, because the
application still needs to figure out which functions run on which processor,
thus incurring a performance overhead. Internal dependencies make it hard to
split apart computations, especially when one piece of the algorithm takes
longer to run than another.
Right now, the best application managers can hope for is to deploy an
intelligent software load balancer and an army of computer science PhDs to
break down algorithms for parallel processing on a case-by-case basis. Gulko
and Mellor propose replacing this arrangement with a layer of software that
will decompose applications in a standard way.
To put it another way, other grid companies tend to concentrate on pure
infrastructure and plumbing. By contrast, Powerllel focuses on decomposition
services and application-level fault tolerance to get the most out of cost-
effective distributed computing infrastructure. The company employs 15 people,
five of whom are engineers.
Technology Powerllel's software runs on Windows NT/2000, Unix or Linux and
handles applications written in C, C++ and Visual Basic. There are three
components to the software suite. First is a range of adapters that manage the
specific requirements of parallel computing for different applications. There
are adapters for Monte Carlo, end-of-day processing, risk management,
portfolio reporting and model calibration applications, as well as for pricing
and hedging single securities and portfolios using multifactor models, and for
recursive structured algorithms. Still to come: more adapters, especially for
legacy applications (in which category CEO Eliot Listman rather surprisingly
includes software written in J2EE, Perl and Python).
The second component of the suite is Lobster, the 'load-balancing sub-task
executer.' Lobster controls the amount of parallel computation dynamically,
depending on how complex the algorithm is and how many resources are
available. It parses and maps the workload, handles fault tolerance and
balances the load.
With Lobster, developers should not have to worry about how to implement
parallel processing tasks (failover, rollback or load balancing, for example)
within the application. That's especially attractive for banks with a big
Visual Basic talent pool, a group of developers who may not be as
sophisticated as the average programmer in C or C++. What's more, given ideal
circumstances, it should only take a few hours to get an application up and
running under Lobster.
Finally, DNET ('distributed network') fleshes out the Powerllel suite with
a
virtual system that can run on various types of cluster middleware, including
DataSynapse, Entropia, MPI, Platform LSF, PVM and Sun's Grid Engine.
Partners On the standards side, Powerllel has participated in early OGSA
(Open
Grid Services Architecture) meetings and intends to contribute to an
application-onto-grid subcommittee. While the company sees IBM moving briskly
forward with OGSA, CEO Listman says it's too early for anyone to bet on the
outcome of the grid standards game.
As far as partners are concerned, Platform Computing is Powerllel's number
one
ally, both as an equity investor and as a co-marketer. Powerllel is also a
member of Egenera's Assured Solution Alliances program, which means
Powerllel's software has been validated for the BladeFrame environment after
testing in Egenera's labs.
Axiom is the first vertical application developer to integrate Powerllel's
software with its own offerings. It's unlikely to be the last. Look for more
partnership announcements in the immediate future, especially with ISVs and
especially in the key financial vertical.
Sales Like many other grid companies, Powerllel markets its grid software
tactically – by downplaying the increasingly suspect 'grid' buzzword and
emphasizing the potential to improve application performance. The company
estimates that it earns $150,000 per trading unit. Deals are moving toward the
$500,000-700,000 range, Listman said. The average sales cycle takes six months
from call to close. Sales made through the crucial Platform partnership do
take longer, but may prove to be much larger. Right now more than half the
company's sales – about 60% – are direct, but Listman hopes to see
partnerships account for 50% of sales over time.
Competition Powerllel's software has been compared with DataSynapse's
LiveCluster, Platform LSF and Sun Grid Engine, but as we've seen, it actually
runs on top of all three. Rather than distributing applications or aggregating
unused cycles, it breaks the application itself apart to more effectively feed
it to a distributed resource pool.
Like Platform Computing before it, Powerllel offers a useful niche
capability
not to be found elsewhere. Unlike Platform, however, the company is poised to
announce an A-round of venture investment. Those investors will doubtless seek
a return on their investment, making Powerllel a likely acquisition
target.
Platform, already a part owner, is the obvious buyer, although the company
could also merge usefully with fellow Wall Street specialist DataSynapse or
serial acquirer Sun. A merger with a financial-sector ISV like Axiom can't be
ruled out, although it wouldn't necessarily be in Powerllel's best interests,
as Lobster does seem to have useful applications beyond the trading desk and
bank back office.
|