By working closely with Intel, Algorithmics has dramatically improved the performance of its risk management solutions. These performance improvements enable a radical change in risk management: A reactive post-trade perspective is replaced with a proactive pre-trade capability, while at the same time dramatically reducing CPU usage and operating costs in the data centre due to the enhanced power thermal features of the latest processors. Full risk simulations for large derivative portfolios of global institutions are reduced from hours to minutes, and pre-deal, what-if risk profiles for plain vanilla and exotic derivatives at the transaction, portfolio, and counterparty level are completed in milli- or sub-seconds.

In the past, says Neil Bartlett, chief technology officer at Toronto-based Algorithmics, trading desks had to use short-cut add-on methodologies to measure counterparty exposure intra-day because the Monte Carlo simulation approaches used in the middle office were too slow.

“Until quite recently it hadn’t been possible to measure risk at the point of taking on the risk. In the past year we have seen the front office want to measure risk in the same way as the middle office. Now with modern hardware we are doing just that.”

Over the years Algorithmics has kept abreast of Intel’s development plans, so it was able to design its latest release of software to take full advantage of Intel’s advances in processor architecture such as the quad-core, 45nm Intel® Xeon® processor L5400 series.

“They tell us their road map and we make sure we are on board,” explains Bartlett. As Intel released its plans for multi-core processors a few years ago, Algorithmics’ developers began designing multi-threaded software that could take advantage of their power.

“We fully understood Intel’s multi-core plans a few years ago, so we are ready for scores of cores on a CPU. The Xeon processor L5400 series is more than 200 times faster than the previous chip, even though the clock speeds are very similar. Much of that improvement in speed comes from the size of the caches on the Xeon processor L5400 series, which is important to us because we are moving so much data. For us, Intel’s development hit a sweet spot in timing because we were planning our development at the same time the multi-core trend was emerging.”

While each financial institution that uses Algorithmics can select its own hardware platform, Bartlett says that the Intel Xeon processor L5400 series running on the Sun Solaris* operating system makes a powerful combination.

“The Xeon processor L5400 series has a very large cache onboard the CPU and it also has very, very good pathways in and out of the CPU. You need that for risk management because you are moving a lot of data around. What Intel has done with low-level instruction sets means you can get really good access to the CPUs; the fabrications systems are small die sets. It is just so fast….”

To get the real value of multi-core, software has to see each core as a single CPU, adds Bartlett. Software which has been designed to run on single-core machines and processing in serial steps has to be completely rewritten for multi-core to maximize the advantages.

Financial institutions are often buying new hardware to take advantage of the speed the Xeon processor L5400 series offers in risk management, he added.

“The quad-core CPUs are very popular at the moment. Financial institutions are buying dual quad-core formats and sticking them together in racks.”

One of the best-known software systems for risk management, Algorithmics makes heavy demands on hardware infrastructure, says Bob Boettcher, senior director, risk solutions at Algorithmics. Counterparty credit risk is very compute intensive because it requires simulation of every over-the- counter transaction that a financial institution holds, and it may be many hundreds of thousands of transactions across thousands of scenarios forward in time, across 100 or 150 time steps.

“You want to know your exposure not just today but over the life of all of these transactions.” All the dimensions of risk management computing are increasing while budgets are remaining flat.

“One reason we work with Intel is that derivative trading portfolios continue to grow,” he adds. “The number of trades is increasing and regulators are pushing banks and banks are pushing themselves for ever more granular risk measurement The standard number of scenarios a few years ago was 1,000. Now regulators want 5,000 scenarios, so the problem just got five times bigger. And regulators also want more time steps to capture roll-off risk. Where they had asked for 30 to 40 time steps, now it is more like 100 to 150.”

To provide the performance that clients need, Algorithmics has a major initiative underway to speed up its calculation engines. Part of the effort involves targeting some of the new capabilities in Intel® processors. Algorithmics is working in the Intel Faster FS Lab outside London to test its software with a full array of the latest fast processor hardware.

“We are running large performance benchmarks in the Intel labs, and they are providing the latest and greatest hardware, starting with the latest Intel chips. But speed means not just fast chips but also fast disk I/O as this is important during the risk aggregation phase. So in addition to providing its processors, Intel is giving access to system engineering so new are assured of taking advantage of features in the silicon. The Intel labs are supported by their partners, so we get access to both pre- and post-production hardware platforms and OS such as Linux* and Solaris*, plus enabling middleware if necessary.”

Financial institutions will take all the capacity that Algorithmics and Intel have to offer.

“In the past, financial institutions would run these big credit exposure calculations once a day. Now they want to run calculations for selected counterparties and industry groups multiple times a day. If you are a credit manager you might want to measure exposures intraday to figure out what transactions you could do to reduce that credit exposure. If you want to do it efficiently with the response time that the business requires, you need very, very high performance. Fast is never fast enough.

The next frontier is using Monte Carlo* simulation-based exposure calculations to assess the contribution of a proposed deal prior to inception. This fits with Algorithmics’ theme of “Measure risk, then do business.” One client is a large American investment bank which is using Algorithmics’ intraday engines to calculate credit exposure for e-commerce. ”Say a hedge fund wants to make a trade. The bank posts prices and the hedge fund has DMA trading terms or goes to the bank’s Web site and wants to buy or sell. The bank doesn’t have a human in the loop on its side, and it needs to be able to do an enabling credit check in 10 milliseconds or the hedge fund will do the trade with someone else or the bank is exposed to execution risk.”

Nigel Woodward, global director of FSI at Intel, concludes, “We have identified risk management as one of the crucial functions for high-performance compute. Our roadmap towards many cores and sockets on a processor opens massive performance opportunities for software houses to upgrade. Risk in FSI will never go away, and these developments get us closer to (proactive) management as opposed to (limiting) control, making capital work harder and safer in a credit crunch world.”

*Other names and brands may be claimed as the property of others.

(No Ratings Yet)  Loading ...

Filed under: Issue 6 - Autumn 08