First it was shortening the time it takes data to get between New York and Chicago, now its New York and London. The big difference here is that the NY to London route is 3 times longer and has an ocean in the middle – no small feet of engineering and no small price tag.
Such links help inform trading programs that need to consider market data coming out of two separate locations, according to Kevin McPartland, senior analyst with market research firm Tabb Group.
“Say there’s a tick up in the price of a future, and all the stocks in a certain basket will tick up a fraction of a second later,” McPartland said. “If you can see two milliseconds faster where the futures moved in Chicago and subsequently make the stock trade in New York, you can more quickly capture the spread.”
Talk of the importance of moving data between NY and Chicago has heat up considerably in the past two years. TABB Group published a piece on this “Long Distance Latency” issue back in June. Here II tackles the topic:
“Ten years ago firms would rely on telecom providers to advise them,” he says. Now the same firms are hiring away those experts to get the jump on the competition and also ensure it won’t figure out their strategy. Telecom experts are paid well for their closed-mouth discretion, sources tell us. Also fueling the need for secrecy is that although “there’s tons of connectivity,” as McPartland says, there is “very little in the fastest” modes.
This is the press release for my latest research report on data center networking. You can find a more detailed executive summary here. Now to the PR:
Despite the cost and complexity created by meeting high bandwidth, low latency and global reach requirements sought by capital markets firms trading across asset classes globally, the industry’s current data center-centric server-to-server approach has actually simplified underlying networks.
According to new research by TABB Group, less equipment, fewer hops and robust management tools are allowing networks to actually flatten as they expand. Once implemented, explains Kevin McPartland, TABB senior analyst and author of “Data Center Networking: Redefining the Total Area Network,” servers and networks can run beautifully – “if people don’t get in the way.”
This trend toward simplification has expanded to several areas of the infrastructure where TABB forecasts that globally capital markets businesses will spend $13.4 billion in 2010 on their infrastructures across equities, fixed income, FX, derivatives, commodities and other capital markets businesses, with 41% of that investment occurring in North America alone for data servers, servers, storage and networking.
“Switches are handling much of the work once left to routers,” McPartland says. “As a result, Storage Area Networks (SANs) are quickly becoming an integrated part of Local Area Networks (LANs) and lines between LANs and Wide Area Networks (WANs) are blurring. This is giving way to what TABB calls the redefined Total Area Network (TAN).” Using the Total Area Network, he explains, network equipment and protocols will be more standardized regardless of their function, and moving data between computers will be seamless despite physical location or the underlying data type. “Those who have the money and infrastructure will see serious efficiencies that will guarantee the return on investment.”
With the server-to server model, virtualization, top-of-rack switches, 10-Gigabit Ethernet (10GE), kernel bypass and other new technological approaches in place and evolving quickly, McPartland believes that although “tomorrow’s flattened network will be simpler to manage, knowing what to deploy, how to deploy it and what to think about based on business needs will be ongoing concerns.”
He gives three examples. According to TABB estimates, in 2010 the network of every major market participant in the US will consume at least 125 terabytes of market data alone with the expectation that no microsecond is wasted and no packet is lost. Second, reducing latency within an infrastructure is centered on removing hops. In a legacy co-location environment, an order message would require 10 hops to move from the client’s order-generating server to the server running the matching engine. A flattening of the network driven by new, faster, more intelligent switches requiring fewer routers can drop the hop count to six, a 40% decrease few trading firms’ CIOs can overlook. Third, assuming a network with 100 switches, which equates to 2,400 ports (physical network connections), a full upgrade to a 10GE environment for a single firm, would cost approximately $1.25 million at a cost of $500 per port, excluding network interface cards and personnel costs for installation.
“Unfortunately, there is no perfect architecture, but by understanding the relationships between latency, bandwidth, scalability and cost,” McPartland says, “the latest and greatest networking technology can underpin an infrastructure that delivers the highest possible ROI. That technology exists today; but since money never sleeps, neither can the men and women designing the networks of tomorrow.”
The 21-page report with seven exhibits is based on conversations with trading firms, network-equipment providers, telecommunications firms and high-speed trading solution providers. It provides a detailed description of how financial services firms are utilizing cutting-edge data center networking equipment and paradigms and discusses tradeoffs between low latency and high scalability and the move to 10GE and beyond.
Discover why organizations like NYSE Euronext are using some of the latest high performance networking technology to improve throughput and reduce latency throughout the trade cycle in this webcast that also features TABB Group Senior Analyst Kevin McPartland.
“From a sell side perspective, it means that those guys–major data center providers such as Equinix and Savvis– should look to expand current client footprints as the sell side looks to reduce server scatter (when firms use space in a variety of geographic locations),” McPartland explained to Securities Idustry News. “So if a major dealer is using three separate data center providers, they’ll look to use only one. Whoever wins that deal will see spending by that dealer grow significantly as they consolidate under one roof.”
So, the sell side will continue to invest in infrastructure to grow capacity as the market demand for trading U.S. equities grows, but McPartland predicts it will look for smarter ways to do this and that growth will be smaller in 2010 than in the past.
McPartland argues that because a completely virtualised and highly utilised infrastructure is still years away for sell-side equities technology, a large gap exists between what is currently possible and actually done. In that same vein, cloud computing will be a part of equity IT strategies in the future but security concerns leave it more interesting than useful.
In conclusion, he says: “Operating data centres, servers and networks with the utmost efficiency will not be a trend but the only way to do business. In 2010, the winning brokers will be those that not only have the latest and greatest technology, but can manage it most efficiently.”
Networking (both intra- and inter- data center). Growing market data message rates and shrinking latency have made networks a key focus of the sell side, said Kevin McPartland, senior analyst with the Tabb Group. “Upgrades of data center network equipment and purchases of long distance bandwidth will accelerate driven by current bandwidth requirements and future capacity planning,” explained McPartland. “And looking beyond bandwidth and transmission speed, reliability is tremendously important as downtime in today’s market is unacceptable.” The core goal: Reduce the number of hops or other factors that introduce network inefficiencies.