What is Hardware Processing?
Trading latency is affected by two key elements: network latency and data processing latency
Network Latency
The goal of our ultra-low latency solution is to deliver information to algorithms in the fastest possible time and with guaranteed latency. The major challenge to this goal is the TCP/IP stack, NIC (Network Interface Cards) between the information from the market and the algorithmic trading appliance.
NovaSparks has addressed this completely; by removing the need for TCP/IP and NIC cards, the required information is delivered to the algorithm without needing to access the network. This is achieved through PCIe outputs operating at 16Gb/s sending the information to a dedicated PCIe card within the algorithmic server. NovaSparks enhances this technology by providing user space device drivers that receive and process the order book efficiently with the lowest possible latency. This approach combined with our fan-out capability ensures NovaSparks clients receive the information they require, with deterministic latency and without incurring network latency during the trading cycle.
Data Processing Latency
Historically, latency has
been reduced by using software programming solutions overlaid onto the
processing speeds of the core hardware functionality. This approach
reduces the latencies significantly; however hardware processing
provides the ultimate means to reducing the processing speeds of the
core functionality itself. This is only achieved using ‘FPGA’
processing, ‘Field Programmable Gateway Array’.
NovaSparks is the only firm to offer guaranteed sub-microsecond latency because all latency reduction is achieved using FPGA programming. Other market solutions offer hardware and overlaid software combinations, however the combination is insufficient to ensure deterministic latency because the point at which software meets hardware creates additional latency in the end-to-end process.
Using a hardware-only approach, NovaSparks delivers end-to-end processing including data feed handling, order book processing, index computation & delivery direct to algorithmic trading black boxes.
Delivered through Unicast, Multicast, one or 10 Gb and PCI-Express, NovaSparks offers a proven, consistent and high performance approach, immune to market bursts and offering a fully scalable data input capacity.