Execs from the Chicago Mercantile Exchange, J.P. Morgan Chase, and Merrill Lynch said they've experienced price-performance improvements by up to a factor of four when they replaced Unix-based systems with those based on Linux. "We're converging on Linux and Windows," said Marc Baumslag, chief technology officer of liquidity and risk technology at Merrill Lynch. "All signs are that Linux is here to stay."
Linux has been crucial to the success of J.P. Morgan's derivatives-trading business, said Scott Marcar, head of technology for global emerging markets. Linux powers about two-thirds of the CPUs that run J.P. Morgan's business of trading derivatives, such as futures and options contracts.
At the Chicago Mercantile Exchange, average daily trading volume on its Globex electronic-trading platform has grown to 2.5 million contracts, representing more than half of the exchange's total daily volume of trades. To help support that workload increase, the Chicago Merc began switching to Linux from Sun Solaris in late 2003, said Joseph Panfil, director of enterprise-technology services. By switching to Linux, the exchange has achieved orders-of-magnitude reductions in order-execution times, he said. "We're going for speed as well as cost savings."
Linux has helped Chicago Mercantile speed processing time, Panfil says.
In addition to moving to Linux, J.P. Morgan has linked servers with as many as 3,000 CPUs in a grid-computing project called Compute Backbone, which performs complex risk calculations to support the derivatives-trading business. The company charges internal users less than a dollar per CPU per hour to tap into the grid-computing system using a similar utility-computing-based approach that Sun Microsystems unveiled last fall. "The focus is on getting unit costs down as low as possible," Marcar said.
The view on grid computing at Merrill Lynch is more mixed. The company has performed multiple proof-of-concept tests of grid computing, including calculating credit risk, but the perfect match between technology and business problems has so far been elusive, Baumslag said. For one thing, Merrill hasn't yet figured out a way to split up number-crunching applications to run efficiently. "Getting a market-simulation algorithm to run in parallel calls for lots of fine-tuning of the software, as opposed to simply a brute-force grid-computing approach," he said.
Assuming the technical issues can be overcome, grid computing offers a way to support business growth without having to buy more hardware. "At Merrill, we're only utilizing half the computing power available," Baumslag said.
Merrill Lynch also has crafted a services-oriented architecture called "On Demand Risk" that provides a single, companywide portal that traders use to access market data and perform risk calculations. The approach is "catching fire across the rest of the company," Baumslag said, but so far, "We're not dictating that business units build around a services-oriented model."