Visit additional Tabor Communication Publications
September 14, 2009
If you've been following the news lately, algorithmic trading, also known as high frequency trading (HFT), has been getting a lot of attention. In a nutshell, high frequency trading uses supercomputing technology to make money from short-term fluctuations in market demand. This is accomplished by aggregating tiny profits over large numbers of transactions. The key to the application model is to find small discrepancies in the buy-sell spread and execute the trade before slower computers can do so.
Not all of the attention has been flattering. By outmaneuvering long-term investors that trade via desktop computers, it is claimed that HFT'ers, unfairly squeeze profits from them. Along the same lines, high frequency trading is thought to overload the market with speculation, and in doing so, increase its volatility.
But whatever the objections, algorithmic trading is big business. No one really knows exactly how much money is being generated, but Larry Tabb of the Tabb Group estimates the take at $8 billion per year. That kind of money is bound to attract a lot of innovation. Investment banks such as Goldman Sachs and hedge fund firms like Citadel Investment Group and Renaissance Technologies generate large chunks of their revenue from this form of electronic trading and are presumed to have invested heavily in HFT infrastructure.
The network vendors are certainly benefitting from this bonanza, including newcomer Arista Networks. As an Ethernet switch vendor Arista has positioned itself as a provider of cloud computing network gear. The company is looking to ride the 10 GigE deployment wave, as the technology becomes mainstream over the next few years. But today, fully 30 percent of Arista's revenue is being generated in the financial services space, with hedge funds the largest segment. Almost all of this is focused on HFT-type deployments.
As it turns out, Arista Ethernet switches are a good fit for this application segment for a number of reasons, the primary one being the low latency nature of the company's hardware. High frequency trading needs to react as quickly as possible to market data, so minimizing latency has become an obsession. For the entire trading pipeline, best-case latencies (which encompass the market data feed, the ticker plant, the messaging platform, and the server that executes the trade) are in the 25 to 50 microsecond range. As processors become faster and the application software becomes more refined, end-to-end latency is expected to drop below 25 microseconds over the next year.
But those numbers don't include the network switch in the middle. A typical high-end Ethernet switch adds another 5 to 10 microseconds, which significantly bumps up total latency.
On Monday, at the High Performance Computing on Wall Street conference in New York, Arista Founder and Chief Development Officer Andy Bechtolsheim is scheduled to talk about some of the trends in networking that are shaping how the financial crowd is attacking the latency issue. No doubt, he will highlight his company's own 10 GbE cut-through switches, which are specifically designed to minimize latency.
In a recent conversation with HPCwire, Bechtolsheim said Arista's 7100 series switches can deliver a packet with a port-to-port latency of less than a microsecond. Compared to the aggregate latency in the rest of the data pipeline, this is almost negligible. Even the latest Nexus 5000, Cisco's unified fabric switches, deliver latencies in the range of 3 to 5 microseconds, while the Nexus 7000 and massive installed base of Catalyst products typically add over 20 microseconds to each network hop.
Bechtolsheim thinks the need for fast execution in these trading applications is making adoption of ultra low latency Ethernet switches a "no-brainer." The overarching goal is to essentially remove the switch from the equation so that HFT system architects can concentrate on decreasing latency on the application side (ticker plant, messaging platform, and trading server) by optimizing the software and upgrading the hardware with the latest CPUs, FPGAs, or whatever.
With all this talk of low latency, you would think InfiniBand would be getting a lot more attention in HFT. Mellanox, after all, advertises 200 nanosecond latencies in its current generation QDR technology. And just recently Voltaire, which employs Mellanox silicon in its InfiniBand switches, announced it had teamed with NYSE Technologies to offer an InfiniBand-based market data system that promises 10 microsecond latency, with a throughput of more than a million messages per second.
But according to Bechtolsheim, for these data feed applications, InfiniBand is not really a good fit. In general, Bechtolsheim has been a proponent of InfiniBand switching in HPC, especially in his role at Sun Microsystems. But in this area, he says, the problem is that market data feeds are arriving in GigE or 10 GigE pipes. So not only is QDR (40 Gbps, 32 Gbps net) InfiniBand throughput overkill, but the non-native protocol is problematic. "By the time you convert from InfiniBand to Ethernet, " said Bechtolsheim, "you're already 10 microseconds behind."
For this class of applications, the InfiniBand effort has focused on tuning the device drivers to increase performance, and in Voltaire's case, adding proprietary messaging software to accelerate application performance. Bechtolsheim contends the InfiniBand contingent hasn't been able to get much of an edge compared to 10 GigE switching. To date there are no field deployments of InfiniBand-based HFT -- at least none that have been publicly announced.
Besides latency and Ethernet compatibility, the other big requirement for these financial apps is reliability. "Robustness is paramount, since if the network goes down, things can get very expensive very fast," observed Bechtolsheim. Here Arista points to its EOS (Extensible Operating System) software, which contains self-healing functionality in order to keep the switch running. Bechtolsheim said it has performed essentially flawlessly and is one of the reasons they've been able to get traction in these kinds of accounts.
Arista is partnering with many of the vendors selling servers, host adapters, and message platforms appliances, into the HFT market. Message platform partners include Solace Systems, Tervela, TIBCO Software and Exegy, among others, while NIC partners include Broadcom, Chelsio, Intel, Neterion, Mellanox, and QLogic.
May 23, 2013 |
The study of climate change is one of those scientific problems where it is almost essential to model the entire Earth to attain accurate results and make worthwhile predictions. In an attempt to make climate science more accessible to smaller research facilities, NASA introduced what they call ‘Climate in a Box,’ a system they note acts as a desktop supercomputer.
May 22, 2013 |
At some point in the not-too-distant future, building powerful, miniature computing systems will be considered a hobby for high schoolers, just as robotics or even Lego-building are today. That could be made possible through recent advancements made with the Raspberry Pi computers.
May 16, 2013 |
When it comes to cloud, long distances mean unacceptably high latencies. Researchers from the University of Bonn in Germany examined those latency issues of doing CFD modeling in the cloud by utilizing a common CFD and its utilization in HPC instance types including both CPU and GPU cores of Amazon EC2.
May 15, 2013 |
Supercomputers at the Department of Energy’s National Energy Research Scientific Computing Center (NERSC) have worked on important computational problems such as collapse of the atomic state, the optimization of chemical catalysts, and now modeling popping bubbles.
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.
In this demonstration of SGI DMF ZeroWatt disk solution, Dr. Eng Lim Goh, SGI CTO, discusses a function of SGI DMF software to reduce costs and power consumption in an exascale (Big Data) storage datacenter.
The Cray CS300-AC cluster supercomputer offers energy efficient, air-cooled design based on modular, industry-standard platforms featuring the latest processor and network technologies and a wide range of datacenter cooling requirements.