Visit additional Tabor Communication Publications
October 12, 2010
HPC cluster maker Appro has unveiled the HF1 server, a purpose-built box aimed at the high frequency trading business. The new server incorporates overclocked Intel Xeon "Westmere" CPUs and a self-contained liquid cooling system to deliver the best dual-socket performance this side of a tricked-out gaming machine. Although the risky design isn't geared for mainstream HPC users, for high frequency traders, it may be just the kind of gamble they are comfortable with.
The multi-billion dollar business of high frequency trading (HFT) is based on a special variant of algorithmic trading, wherein high-powered computers are hooked up to low-latency market feeds, in order to execute lightening fast trades. Typical participants include investment banks, hedge funds, and proprietary HFT firms who specialize in this type of operation. The general idea is to slice small profits from each trade by anticipating pricing ahead of other investors. To make a profit, though, many trades have to be executed on a daily basis, so a typical asset is only held for a very short time. This requires extremely fast networks and even faster computers.
For HFT, the performance of the network ultimately limits the speed of trading, but as network latency drives toward zero, the playing field has become more even. Now HFT'ers are looking to what goes on inside the server boxes to get their competitive edge. One way to speed up the software that decides which trades to make is to make the processors themselves faster. Unfortunately, CPU clock frequencies have been in a holding pattern for several years, thanks to the physical limitations of semiconductor transistors.
Today, a top bin x86 processor will flirt with 3.5 GHz. That's plenty fast if you can scale your application to take advantage of more cores and processors -- the way most of HPC works. In that case, you're able to crank up execution speed by throwing more servers at the problem. HFT applications, though, require fast single-threaded performance.
According to John Lee, vice president of Appro's Advanced Technology Solutions Group, a small number of financial services customers convinced the HPC vendor to build overclocked x86 servers designed specifically for the high frequency trading domain. Although these customers are in line to get the first hardware when it hits the streets later this year, Lee thinks all the HFT players will want to take a hard look at the new HF1. "It's a product they can't buy right now because there is nobody addressing the special needs these guys have," he says.
The server is a dual-socket box using Intel Xeon X5680 processors overclocked to frequencies up to 4.4 GHz. That's nearly a full gigahertz quicker than the top bin 3.46 GHz Xeon X5677 available today, and even outruns the fastest speeds supported by Intel's turbo boost technology. To keep the overclocked CPUs cool and collected, Appro has plumbed the server with a closed-loop cooling system (pump, coolant reservoir, pipes and radiator) to draw the extra heat away from the over-achieving chips. "We're taking the soul of a gaming machine and packaging it in a commercial rackmount server," explains Lee.
The makes for a rather unconventional HPC machine. All the internal plumbing expands what would have been a 1U server into a 3U box. And because of the overclocking and extra plumbing, a lot more power is required -- in the neighborhood of 1 kilowatt. That's fine for HFT set-ups since speed, not density or power, is the driving consideration.
A bigger concern is the short warranty period. Most Appro servers come with a standard three-year warranty, but for the HF1 that's reduced to just a single year. More importantly, the processors are only guaranteed for 30 days since the overclocking voids the Intel warranty on their chips. Lee says they are offering a one or two year warranty extension for the HF1, but at an additional premium.
Although Intel is not guaranteeing the overclocked processors, it is cheering Appro on and is certainly keeping an eye on the lucrative HFT market. It's not too big a stretch to imagine that the chipmaker would want some of the action by offering faster clocks in its mainline Xeons. Whether this would take the form of faster CPUs with fewer cores (maybe just one?), or some exotic technology that allows the frequencies to breeze past 3.5 GHz is unknown. Almost certainly, the presence of 4-plus GHz Xeons would entice Tier 1 OEMs and others to build purpose-built HFT boxes. But at least for the time being, Appro is alone in the fast lane.
Not without some risk, though. The up-front cost of an HF1 puts these servers in a class by themselves, price-wise. Given that these systems have to be hand-tuned to balance the clocking on the various motherboard components, the premium is quite steep.
Although Appro is not publicizing its pricing, Lee implied the HF1 would cost several times that of a standard $4,000 or $5,000 HPC server. Ordered in quantity, Appro would no doubt offer better deals, but even the largest deployments would be just a few hundred units, so this solution is only for the DEEP-pocketed.
The five-figure pricing and short warranty eliminates this server from consideration for the typical HPC user. A DOE lab, for example, would love to have the faster CPUs for its supercomputers, but the per unit costs are too high for a multi-thousand node machine. And in any case, by the time a lab ran a system through acceptance testing, the processors' warranty would already have expired.
Unlike the other HPC domains, in high frequency trading, server costs are a relatively small piece of the operation, especially considering the money that can be generated from these machines. A 2009 report by the TABB Group estimated that HFT generated about $21 billion in profits in 2008 (that's profit, not revenue). If a single server can slip in, say, twice as many trades in the same timeframe as its slower competition, the ROI is quite rapid. From conversations Lee has had with HFT customers, they're confident the cost of these machines can be recouped within the first few months of operation. According to him, when the subject of server cost comes up, "nobody bats an eye."
Jun 19, 2013 |
Supercomputer architectures have evolved considerably over the last 20 years, particularly in the number of processors that are linked together. One aspect of HPC architecture that hasn't changed is the MPI programming model.
Jun 18, 2013 |
The world's largest supercomputers, like Tianhe-2, are great at traditional, compute-intensive HPC workloads, such as simulating atomic decay or modeling tornados. But data-intensive applications--such as mining big data sets for connections--is a different sort of workload, and runs best on a different sort of computer.
Jun 18, 2013 |
Researchers are finding innovative uses for Gordon, the 285 teraflop supercomputer housed at the San Diego Supercomputer Center (SDSC) that has a unique Flash-based storage system. Since going online, researchers have put the incredibly fast I/O to use on a wide variety of workloads, ranging from chemistry to political science.
Jun 17, 2013 |
The advent of low-power mobile processors and cloud delivery models is changing the economics of computing. But just as an economy car is good at different things than a full size truck, an HPC workload still has certain computing demands that neither the fastest smartphone nor the most elastic cloud cluster can fulfill.
Jun 14, 2013 |
For all the progress we've made in IT over the last 50 years, there's one area of life that has steadfastly eluded the grasp of computers: understanding human language. Now, researchers at the Texas Advanced Computing Center (TACC) are utilizing a Hadoop cluster on its Longhorn supercomputer to move the state of the art of language processing a little bit further.
05/10/2013 | Cleversafe, Cray, DDN, NetApp, & Panasas | From Wall Street to Hollywood, drug discovery to homeland security, companies and organizations of all sizes and stripes are coming face to face with the challenges – and opportunities – afforded by Big Data. Before anyone can utilize these extraordinary data repositories, however, they must first harness and manage their data stores, and do so utilizing technologies that underscore affordability, security, and scalability.
04/15/2013 | Bull | “50% of HPC users say their largest jobs scale to 120 cores or less.” How about yours? Are your codes ready to take advantage of today’s and tomorrow’s ultra-parallel HPC systems? Download this White Paper by Analysts Intersect360 Research to see what Bull and Intel’s Center for Excellence in Parallel Programming can do for your codes.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?
Join our webinar to learn how IT managers can migrate to a more resilient, flexible and scalable solution that grows with the data center. Mellanox VMS is future-proof, efficient and brings significant CAPEX and OPEX savings. The VMS is available today.