QuantHouse is an established player in the low-latency datafeed handling business, but only very recently launched a hardware accelerated solution. IntelligentTradingTechnology.com asked the vendor’s co-founder and COO Pierre Feligioni to explain the move, and provide some more technical detail.
Q: QuantBOX is your first offering to make use of hardware acceleration. With the continuing performance advances in the x86 architecture, why do this now?
A: QuantHouse has always used the latest and most efficient techniques to reduce market data delivery latency. For example, kernel bypass techniques and fine tuned Linux kernel, coupled with accelerated network cards. It is our goal to help our customers get the fastest market data available. We believe hardware-accelerated solutions are the next step to significantly reducing latency, as well as introducing processing time determinism in data processing.
QuantHouse use the latest hardware components of x86 for our servers, so we are well-informed of how the architecture is evolving. However, there are certain tasks where x86 are not as good and this is where FPGA outperform. Therefore with QuantBOX we have decided to combine the advantage of hardware capabilities with our best of breed software solution, QuantFEED.
Q: How are you positioning QuantBOX versus your existing QuantFEED software option?
A: QuantBOX must be seen to be the leading solution for ultra low latency market data delivery and proximity hosting. This is a premium solution targeted at the most demanding client.
Q: What functions have been offloaded to the FPGA in QuantBOX? And what has been the performance improvement?
A: We have off-loaded the entire network stack (IP/UDP) as well as market data processing from the software (protocol decoding, reordering and enrichment). Additionally, the FPGA offers the traditional zero copy / kernel bypass functionality to further reduce the latency and avoid unnecessary kernel context switching. After multiple benchmarks in our dedicated labs, we saw a 20X improvement for the parts that were off loaded.
The Market Data Offload Engine (MDOE) in QuantBOX can be used as a drop in replacement of a standard NIC just by changing a setting in the software.
Q: What FPGA processor/card manufacturer are you using? And what tools are/languages are you using to program the FPGAs? What programming challenges have you found, and how have you overcome them?
A: The solution is built around an Altera Stratix IV GX onboard a PLDA XpressGX4LP. The system is developed in SystemVerilog.
We started the development in VHDL, but quickly shifted to SystemVerilog to ease the integration of all the IPs from our different partners. SystemVerilog also helped us speed up the development as well as improve test coverage.
The trickiest part of the development was the integration of the PCI Express communication interface between the FPGA board and the software. It required deep hardware and software (driver + application) knowledge to be implemented efficiently as well as a good communication between the hardware, software team and the suppliers.
We also put a lot of effort in managing the external memory (QDRII & DDR3) efficiently in order to get the most performance out of the controllers.
Q: Has QuantBOX been rolled out to customers? What’s the release status? And what are plans for future versions?
A: QuantBOX is in pre-production phase at a set of trusted customer’s premises. We are confident about our progress and we plan to make QuantBOX publicly and commercially available before the end of February 2012 for production use. The next step will be to deliver a QuantBOX dedicated for Forex feeds, combined with our unique market data consolidation solution.
Q: More generally, what have been some other recent successes for QuantHouse? Where are you seeing demand, and which of your products? Where are the hot spots, and where is the growth?
A: QuantHouse has noted a significant uptake of the firm’s ultra-low latency market data solution among top tier investment banks as the sector looks to improve financial efficiency and adapt rapidly to market conditions.