As trading firms and exchanges move beyond 10GbE connectivity towards 25GbE and 100GbE, the surrounding infrastructure needs to evolve with them. While high-density Layer 1 connectivity has long been available at 10G, providing similar levels of flexibility and port density at 25G has proved more challenging.
LDA Technologies is seeking to address that gap with LDA Maze, a new non-blocking Layer 1 connectivity fabric supporting up to 160 10/25G links, or 40 aggregated 40/100G links, with measured port-to-port latency of between 3.6 and 4.1 nanoseconds.
The new fabric is designed to provide a high-density interconnect between components including front-panel connections, FPGA boards and network interface cards (NICs), allowing trading firms to distribute signals between large numbers of devices while reducing the rack space, cabling and power required to do so.According to Vahan Sardaryan, CEO and co-founder of LDA Technologies, the development reflects a shift already under way in exchange connectivity. Trading firms have been using Layer 1 technology for well over a decade, primarily at 10G, where high-density crosspoint technology has allowed them to connect and redistribute signals across large numbers of ports. Equivalent capabilities at 25G, however, have been more limited.
“At 25G, trading firms can now have the same flexibility they had with 10G,” says Sardaryan, in conversation with TradingTech Insight. “They want to connect a lot of devices and use Layer 1 to route anything to anywhere across a lot of ports. That’s a basic use case trading firms didn’t have at 25G for a long time. Now they do.”
Moving the infrastructure with the connection
The significance of the move to 25G extends beyond the additional bandwidth available on an individual connection. Trading architectures typically need to distribute exchange market data and other network traffic across multiple servers and processing components, potentially incorporating FPGA-based timestamping, feed handling, aggregation or application switching.Layer 1 technology performs that distribution at the physical network level, electrically replicating and routing signals without introducing the processing overhead associated with higher layers of the network stack.
“It acts as a patch panel or a router, taking a signal from point A to multiple points in the chassis,” says Sardaryan. “It can take a signal from one place and put it through multiple front-panel ports. It’s basically the nerve centre of the architecture.”
Maze supports a mix of SFP, QSFP and QSFP-DD I/O cards, enabling the front-panel configuration to be adapted to a particular deployment. Firms can also operate the same fabric at 10G before moving individual connections to 25G, avoiding the need for a wholesale infrastructure redesign as higher-speed connectivity is introduced.
That flexibility could become increasingly important as the transition happens at different speeds across venues, asset classes and firms. Rather than treating 25G as a standalone network upgrade, firms need to consider how higher-speed connections interact with the rest of their low-latency infrastructure.
Trading nanoseconds for flexibility
Perhaps more significantly, Maze highlights the trade-off between absolute minimum latency and infrastructure flexibility.
For the most latency-sensitive market participants, where individual nanoseconds can influence architecture decisions, direct device-to-device connectivity or specialised ultra-low-latency fabrics may remain preferable. But Sardaryan argues that this represents only part of the market.
“There is a huge layer of firms that don’t care about an extra three nanoseconds and prefer the flexibility,” he says. “They want a large device where they can plug in all their cables, with a lot of I/O to distribute the signal to many servers. They had that at 10G; they didn’t have it at 25G.”
As low-latency infrastructure becomes more complex, that is an important distinction. Performance remains critical, but not every trading firm is pursuing the theoretical minimum latency at every point in its architecture. For many, predictable latency combined with greater density, easier configuration and the ability to redistribute signals across multiple systems may represent a more useful engineering trade-off.
The result is a somewhat broader definition of low-latency infrastructure: one that considers not only how quickly a signal can travel between two points, but how efficiently increasingly complex trading environments can be connected and scaled.
Extending beyond trading
Although capital markets are a primary use case for Maze, LDA has designed the fabric as a modular component rather than a trading-specific appliance.
“One interesting use case where we’re seeing a lot of interest is high-performance computing,” says Sardaryan. “You can connect multiple high-power FPGA boards through the fabric, which is valuable for computational scenarios where multiple FPGAs are working on a single task and need to exchange information as quickly as possible.”
That potentially extends the role of the technology into AI and other compute-intensive workloads, although the immediate relevance for trading firms lies in the convergence of high-speed networking, FPGA processing and increasingly demanding data workloads within the same infrastructure.
Maze can currently be integrated into LDA’s Neo X platform, with the company planning to make the technology available across two additional platforms due to launch in the coming weeks. Multiple Maze units can also be chained together where deployments require still greater connectivity.
As 25G connectivity becomes more widely adopted within capital markets, the challenge for trading firms will increasingly shift from obtaining faster connections to ensuring that the infrastructure behind those connections can handle them efficiently. High-density Layer 1 technology is one relatively small but important part of that transition, particularly for firms looking to increase capacity without sacrificing the flexibility they have become accustomed to at 10G.
Subscribe to our newsletter
