Wall Street IT Gets the Message: More Speed, Fewer Dollars
Do more with less and do it faster. That’s a pretty familiar order to anybody running an IT operation these days. But given the recent events in the news, no one is hearing that mandate more than people in the capital markets. At the High Performance on Wall Street conference this week, hundreds of financial services IT practitioners came looking to find ways to meet the demands of business that moves at the speed of microseconds.
The big themes of financial IT — latency, throughput and power efficiency — have intensified in the last six months, speakers and attendees at the conference generally agreed. “In the U.S. markets, what we’ve found is that tolerances have gone down by an order of magnitude,” said Peter Lankford, founding director of the Securities Technology Analysis Center (STAC). “We used to trade in hundreds of milliseconds. It’s now to the point where tens of microseconds, if not microseconds, really matter.”
So, how to speed things up and stay competitive in a volatile market? How to accelerate and improve the quality of critical applications like credit risk analysis?
“It’s a software problem,” says Ambreesh Khanna, global head of financial services at Sun Microsystems. “The biggest performance problem going forward is in the software space. Software is not taking full advantage of the processors available today. We need applications that can scale across 10,000 CPUs. Writing software that can scale is the problem today.”
Microsoft might not see it as a problem, but would agree that software is the heart of the solution. Company execs were at the conference to promote their new Windows HPC Server 2008 as a way to easily deploy scalable, high-performance applications across clusters. “We’re providing seamless scalability from desktop workstation to cluster by letting users take advantage of the power of distributed computing, and in a familiar Windows environment,” said Vince Mendillo, director of the Server & Tools Business Group. The new software includes new management and administration tools that simplify configuring and deploying cluster nodes, scheduling jobs, allocating resources and monitoring system health. Advanced failover has been built in to ensure reliability, and “there’s very low latency when sending a process from one machine to another,” he said.
Ricky Higgins, director of IT at Lloyds TSB, one of the largest banking groups in the United Kingdom, said his group was able to set up a 64-node cluster in a matter of hours. “We’ve also increased the volume of transactions and reduced processing time by about 50 percent,” he said.
Microsoft isn’t just targeting financial services with this platform. Bill Laing, corporate vice president of Microsoft’s Windows Server and Solutions Division, said during his keynote that the launch of the new HPC Server “is just another step in our vision to drive HPC mainstream.”
Another step toward that is a partnership with Cray, announced last week, to offer a diminutive supercomputer (smaller than a footlocker), the Cray CX1, that runs Windows HPC Server 2008. Prices start at $25,000.
Red Hat product manager Bryan Che said his company’s Enterprise MRG software, built on top of Red Hat Enterprise Linux, is going to help financial services companies by integrating high-speed messaging and grid technologies to provide faster response times and scalable resources. “Users can dynamically schedule virtual machines across heterogeneous environments, and if they run out of capacity, they can then go out to, say, the Amazon cloud,” he said.
Another Linux provider, Novell, is evolving its SUSE Linux Enterprise to meet the demands of real-time, scalable computing. One technique that will show up next year is “predictive virtualization,” said Moiz Kohari, Novell vice president of engineering for financial services. “The software will be able to predict when a task will run and how long it will run. The objective is to provide real-time guarantees to the application running in that process.”
But for every company at the conference promoting software solutions, there were two touting hardware. Superfast-server companies like IBM and Sun were showing their HPC systems, as were smaller companies like Super Micro Computer and Stratus. Others were focused on boosting application performance by eliminating communication or process bottlenecks. Voltaire, which makes the Grid Director line of 20 Gbps switches, said its Messaging Accelerator is designed for low-latency applications such as data-feed handling. Voltaire‘s products use InfiniBand technology to connect processors and I/O in high-performance environments. InfiniBand fabric enables users “to consume messages fast enough to accommodate market growth,” said Oleg Lukyanov, solutions architect at Voltaire. InfiniBand can move “very large amounts of data,” he said. “As you grow beyond hundreds of nodes, you reach scalability problems. But using InfiniBand, we can definitely deliver a thousand-node analytic farm.”
One topic in which conference attendees appeared interested — at least based on traffic at exhibitor stands, and a standing-room-only panel on the topic — is and using specialized hardware accelerators to speed up financial calculations.
“Accelerator hardware enables us to do things we couldn’t do otherwise,” said Henry Young, founder of TS-Associates, an IT services company specializing in financial middleware. “We can handle a 10-gigabit data stream through a hardware accelerator to speed up processing. You can’t compress a 10-gig stream like that with current CPU technology.”
As a result of companies developing devices “anyone can plug in and start using,” said Geno Valente, vice president of sales for Extremedata. Wall Street is now taking advantage of them. “Public exchanges are using accelerators,” he said. The benefits include faster processing of algorithms used in trading and analytics, for example, but also, in many cases, increased reliability with error detection and built-in failover capabilities.
Consolidation is another benefit, these hardware makers say. Exegy‘s acceleration and compression technology is incorporated in a ticker that “can handle all the North American market data feeds in one box instead of a dozen,” said chief architect Scott Parsons. (See it in action at marketdatapeaks.com, he noted.)
Of course, HPC is not new to Wall Street. “HPC is on Wall Street,” Valente said. “Most people have some kind of HPC in the back room, but they’re typically small Linux clusters.” Why these companies are adding accelerators, he said, is for more performance, but also because “an accelerator in a scalable form will distinguish a cluster. Otherwise, your box is the same as the other guy’s.”
Bottom Line: Bottom Line
There is no lack of potential ways to acquire the kind of performance, reliability and scalability that financial applications demand. Increasingly, however, these solutions will have to fit into shrinking budgets. As Microsoft’s Laing said during a briefing, “Financial services companies are going to have to be more efficient than ever before with their IT resources.”
At the last session of the conference, participants were asked what they thought the effects of “the recent market chaos” will be. Tighter budgets were definitely on everyone’s minds. “Changes like this force people to take a harder look at the reality of expenses and where the money is going,” said Dan McCormick, CEO of Atrato, a maker of high-speed storage systems. “Power costs have gone up 40 percent, so people have got to start asking ‘How can I get better performance in less space and for lower energy costs?'”
“These changes are based on fundamentals of the market,” Lankford of STAC said. “Things are just going to get tighter and tighter.”