Low Latency & High Perf Technologies for Financial Applications

A-Team Insight: High Performance Technologies for Trading
Ultra-Fast Networks for Financial Data Centers

Competitive success in today’s financial markets is defined in fractions of a second. Studies have shown that tens of milliseconds of delay in data delivery can represent a ten percent drop in revenues, and delays of even five microseconds per trade can cost hundreds of thousands of dollars. Industry-specific requirements for uncompressed data and an end-to-end deterministic latency within tens of microseconds make attaining such performance even more difficult. These factors have combined to make raw speed a top priority for financial networks.
The challenges facing the IT professionals who manage today’s financial networks are significant, but not insurmountable. All can be addressed with a new approach to data center networks that provides ultra-high bandwidth and ultra-low latency through a simple, scalable, and cost-effective design.

Low Latency and High Bandwidth Switching

In financial markets, market data applications must deliver information at ever-increasing rates and with minimal latency. Market data arriving even a few microseconds late can have huge negative impact on a financial trading firm. 
Currently Gigabit Ethernet networks are reaching their limits in bandwidth, maintainability and latency. Multi-core processors, virtualization and converged storage and data networks require an efficient solution for adding bandwidth.
While InfiniBand offers low latency and much higher bandwidth than Gigabit Ethernet, we believe InfiniBand is not always as fast or as reliable as advertised, and is not that easy to deploy and maintain. Scaling to 10 Gigabit Ethernet is a natural evolution for Gigabit Ethernet networks, and doesn’t require expertise in deploying and maintaining separate network fabrics.

Scale-out with Less Rack Space and Better Performance
Traditional three-tier data center architectures consisting of server access switches, distribution switches and large core routers take up a lot of space. And they can require data to travel many hops just to communicate with other servers in the same rack or row. A new approach using denser racks and Top of Rack switches can shrink data center footprint, enable faster communication across fewer hops, and minimize overall application latency.

BLADE’s RackSwitch line of Gigabit and 10Gb Ethernet Top of Rack switches offer best in industry performance along with hardware features ideal for dense data center deployments and managed or unmanaged collocation facilities:

• RackSwitch G8100 and G8124 have been independently verified for 100% line-rate performance and extremely low latency.
• BLADE’s RackSwitch G8000 1/10Gb Ethernet aggregation switch shaves off microseconds of latency at Gigabit Ethernet rates while saving 50% in cost per Gbps compared to a leading competitor.
• Support for Layer 3 protocols enables faster communication with various market centers than traditional routers.
• Support for IP Protocol Independent Multicast (PIM) Sparse Mode and Dense Mode enables direct low latency connections to multicast Market Data distribution networks.
• Data center optimized airflow: All RackSwitch models support either Front-to-Rear or Rear-to-Front cooling, letting end users choose the switch that has the same airflow as their server and storage devices. Directional airflow reduces the need for extra cooling devices, reduces costs, and improves MTBF—all ideal for placement in collocation facilities. As a result you spend less money cooling your data center compared to switches that cool from side to side or side to rear.
• Low power consumption: an average of nearly $200 savings per year per 10 Gigabit Ethernet switch quickly adds up when your data center scales to hundreds of servers and switches.
• Shorter switch depth (15 inches): allows more room for cables, and makes RackSwitch easy to fit into dense, energy efficient and low cost rack-based solutions like IBM’s iDataPlex.
• Simple redundancy: Uplink Failure Detection provides sub-second failover and a loop-free topology without the complexity and slow convergence time of Spanning Tree protocol.
• Interoperable: All of BLADE's switches have been certified as interoperable in multi-vendor networks like Cisco, Brocade, Juniper and others.

• BLADE’s switch modules for IBM BladeCenter and HP BladeSystem connect blade servers to the rest of the network at 10 Gb Ethernet while reducing cabling and improving reliability, cost, and power.