Amidst a flurry of technology announcements this week, NYSE Euronext said yesterday that it's working with Juniper Networks to design a low latency network for the new data centers it's building in the greater New York and London metropolitan areas to supplement its existing data centers in New York and Paris. The two new data centers will support several billion daily transactions and quotes and are part of NYSE Euronext's effort to consolidate its total number of data centers from ten to four. Expected to be operational in 2010, the new facilities will use Juniper's 10 gigabit Ethernet network technology to support an internal latency of 50 microseconds roundtrip.
"To run world class markets, you need world-class technology, they're two sides of the same coin," noted Stanley Young, CEO of NYSE Technologies and Co-Global CIO of NYSE Euronext, in a recent interview. "Inevitably, with the speed and latency requirements of the modern trader, and the capacity they're looking for in terms of transaction throughput, we have to be able to accommodate their vast needs of processing power within the data centers themselves. We're seeing an almost exponential increase in their requirements for power, for space, for throughput with matching engines. Three years ago, a colo requirement was a single rack. We're now seeing firms looking at 50 to 60 racks of capacity. As they offer sponsored access and direct market access for their clients to our market, that capacity requirement grows exponentially."
By collapsing the multiple switching layers present in traditional network architectures, the new simplified NYSE Euronext network design is said to require fewer devices and interconnections, leading to improved efficiencies in space, power, cooling and management.
The data center architecture uses top of rack EX2500 Ethernet Switches to provide access connectivity to the data centers' servers with 10GbE ports that deliver wire-speed throughput. NYSE Euronext also plans to leverage routing capabilities such as MPLS network virtualization and low-latency multicast.
Between its new data centers and the nearby access centers customers connect to, NYSE is running the world's first 100 gigabit network, using technology from Ciena.
"We're looking at all technologies, I certainly think the fabric within the data center itself is a critical component of the jigsaw puzzle for low latency infrastructure," Young says. "We're looking at the whole area of shared memory access and how do we enable the algorithms of the broker/dealers and the matching engines to operate in a shared memory space, so that there's no latency caused in the passage of transactions between one chip and another. These are all areas we're exploring as we try to get to that magical zero latency."
The new data centers are sizable: the New Jersey facility will have 100,000 square feet of raised floor, the London center 70,000 square feet. NYSE itself will only use 20% of that space, the other 80% is dedicated toward hosting and colocation for customers. "If we can have as much of this low latency infrastructure for our trading firms that wish to trade on our market in the data center itself, we believe that creates a momentum, a trading environment that opens up a virtuous circle of transaction data which is self-perpetuating. So the size of the data centers, the ability to host and manage the infrastructure of a huge number of market participants becomes a huge advantage."
The data centers have been designed in pod configurations; in New Jersey there will be five pods of 20,000 square feet each and the London facility will contain seven pods of 10,000 square feet each. "We're building these so we can bring them online as required," Young says. "They're as efficient as we can make them. We've been conscious of the need to be as green as we can, but there's no getting away from the fact that these buildings are massive consumers of data, purely because of their high concentration of super computing power."