At the Chicago Mercantile Exchange, the traders betting on future commodity price fluctuations aren't the only...
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
ones who know what it means to take a risk. Five years ago, Chicago Mercantile Exchange (CME), now part of the CME Group, placed a hefty bet: that it could trust Linux with the heart of its business, executing billions of contracts a year and processing them faster than previously.
CME won its gamble. And in so doing, it proved that an open source platform, a relative newcomer to the data center scene in general and the financial world in particular, had the power and speed to handle high-volume, real-time, mission-critical applications.Outmoded servers, transaction times prompt migration
The Exchange's 800 Solaris Sparc servers, which were simply outmoded, nudged CME in the direction of Linux. "We were starting to have issues with Sparc," said Vinod Kutty, CME's associate director and head of distributed R&D computing. Not only did they cost too much to support, but it wouldn't have been able to keep pace with the projected increase in trade volume, especially with electronic trades growing even faster than the business as a whole and pit trades on the floor becoming less predominant. "We wanted to save money and get much better performance," Kutty.said
Another factor in the conversion was latency. The technological bar is continually rising on how fast a transaction must be completed to be considered "real time," but Kutty defines it, futuristically, as a single millisecond. At the time of CME's conversion, the Sun Sparc servers were averaging 200 milliseconds per transaction, about twice what was considered real time in 2003, he said. Today 15 to 20 milliseconds is the benchmark, he said.
Although people can't physically react any faster to trades that occur in a tiny fraction of a second, it's still important to come as close as possible to real-time transactions because it affects the perceptions of traders, Kutty said.
"We have historically noticed that when we improve the performance of the trading engines, traders will trade more often," he said. "The faster they trade, the more they can trade, and the faster their orders are filled." Therefore, in 2003, CME began a gradual migration to X86s running Red Hat Enterprise Linux (RHEL) 2.1, starting with less critical applications, then progressing to order-entry programs and, finally, to the mission-critical trading applications, he said. At each stage, the system was fully tested before progressing to another application, he said. The transition was completed in 18 months at the end of 2004. The operating systems have been continually updated as well, with the servers now running a mix of RHEL 3 and 4 and certification of RHEL 5 currently in process.
The timing of the transition proved fortuitous. Two years later, CME would acquire yet another boost in transaction volume from its merger with the Chicago Board of Trade, hence the name change to CME Group.In the five years since the conversion began, the number of servers has grown from 800 to 4,000, the number of electronically traded contracts from 250 million to 1.2 billion in 2007. The dollar value of those 1.2 billion contract trades is $1,200 trillion.
In addition, the new RHEL-based servers run much faster, chopping transaction speed from 200 milliseconds to 15 to 20 milliseconds and are still heading downward, Kutty said. In two ways, CME helped to accelerate transaction speeds: first, by optimizing the TCP (Transfer Control Protocol) and second, by creating a Super HOP (Host Output Process), that routes trade orders and confirmations through a group of fast, high-powered Linux servers, he said.Not without a hitch
Although the platform switch has been successful, the transition wasn't problem-free. "We've had issues with the kernel performance," Kutty said. "Earlier versions of Linux were less stable." Another challenge was retraining Solaris users in Red Hat monitoring tools, Red Hat's certification program is "one of the better ones," he said.
An ongoing issue is the need for a broader mix of open source diagnostic tools that would provide visibility into the factors impeding real-time performance without disrupting ongoing operations, he said. In addition to more diagnostic tools, CME also would benefit from a lightweight version of the operating system that could be simulated in a lab for troubleshooting purposes, Kutty said.On the positive side, CME has reaped the benefit of problems solved within the greater open source community instead of waiting for a fix from a proprietary vendor, he said. And because OEMs have access to the source code, they can do a lot more testing, which in turn speeds issue resolution, he added.
The migration to Linux has paid off in improved performance, lower cost and greater stability, Kutty said. In addition, CME has gained better leverage with vendors, which translates into more price leverage for support services, he said.
Although CME has been an industry early adopter in moving to open source, its success has yet to create a stampede of followers. One reason, Kutty speculated, is that OEMs aren't designing servers specifically to run Linux; if they were, the hardware would be easier to use innovatively and easier to manage, he said.Even so, the experience of the Chicago Mercantile Exchange proves that Linux is ready to fill mission-critical roles in the enterprise, Kutty said.
Let us know what you think about the story; email Pam Derringer, News Writer .