Home > Networking

Computerworld - First Union Corp. last week said it has found a way to dramatically cut the cost of performing heavy-duty number crunching: taking advantage of peer-to-peer computing.

Peer Pressure Three top requirements for peer-to-peer computing: A strong business motive for time-sensitive turnaround An application that requires a great deal of computing power An application that has parallel construction. For example, each portfolio or subset of a portfolio can be analyzed separately from all others, and other computations don’t depend on the results. Source: DataSynapse Inc., New York

Financial services firms that need to crunch numbers have traditionally had two options: They could run the application on a high-end server, which typically costs somewhere in the vicinity of $300,000 and can take eight hours every night; or they could do the computation in real time on a dedicated server farm that can cost 10 times as much when the programming time is factored in.

But banks are finding that there's a cost-saving alternative: running the application over existing PCs by slicing up the computational work and spreading it out among the hundreds of workstations that sit idle for some parts of each day.

According to Charlotte, N.C.-based First Union, that peer-to-peer computing alternative costs roughly as much as buying one server yet produces nearly real-time computations. Joe Belciglio, First Union's managing director of trading technology, said last week that the bank has taken the peer-to-peer approach with an application for analyzing portfolio risk that integrated easily and unobtrusively with existing systems.

The work was done by New York-based start-up DataSynapse Inc., which says it has eight other Wall Street customers. But not every application is suitable for the peer-to-peer approach, said DataSynapse CEO Peter Lee.

"There are certain applications which are hopelessly written in spaghetti code," he said. "We try to stay away from those."

Lee said the applications that work are those in which computations can be done in parallel - for example, the risk associated with one investment portfolio can be calculated at a separate place or time from that of another.

Lee said DataSynapse has already developed all the middleware it requires, so converting an application to a peer-to-peer format can be accomplished in two weeks or less.

Michael Packer, managing director and head of institutional e-commerce at New York-based Merrill Lynch & Co., said the peer-to-peer concept is nothing new on Wall Street.

"We've been doing it for years," Packer said. He also noted that Merrill Lynch and many other financial firms have homegrown peer-to-peer systems or systems based on academic software to run complex risk scenarios on options derivatives.

But Packer acknowledged that commercial offerings such as that developed by DataSynapse are typically more "robust and sophisticated" than homegrown peer-to-peer systems.

Other Wall Street firms are waiting to see if commercial offerings are right for them, said Larry Tabb, an analyst at Needham, Mass.-based TowerGroup.

"It's not something they're going to roll out tomorrow," he said. "But they'll kick the tires to see if it works, if their staff are inconvenienced if these processes are going on behind the scenes, and if it works and the cost savings are as dramatic as they say."

Read more about Networking in Computerworld's Networking Topic Center.