Intel’s new Ivy Bridge processor is driving new server technologies that the financial markets are keen to adopt. But in today’s cost-reduction environment, it’s not just about top performance – through accelerated CPU clock speeds – but also how to roll out and operate in a cost effective way – which brings in server density, power, ease of deployment and reliability. IntelligentTradingTechnology.com found out more from Supermicro’s global business development head Don Clegg.
Q: You just released new Hyper-Speed servers featuring the latest Ivy Bridge processors from Intel. How do they compare to the earlier Sandy Bridge-based servers. Do you expect trading firms to upgrade?
A: Supermicro believes that high-frequency, low-latency trading firms will gain significant competitive advantages by upgrading to our Ivy Bridge Hyper-Speed product line. The performance increase from the new generation architecture and firmware improvements are, we believe, unrivaled with 15% lower median latency and an incredible 94% reduction in jitter. Supermicro was already the industry leader in HFT server solutions with our previous Sandy Bridge Hyper-Speed product by providing the industry’s only hardware accelerated enterprise-class server solution, and the Ivy Bridge Hyper-Speed product line further improves our performance advantages to bring the fastest performance for mission critical trading. New features include Turbo Mode Stabilisation and an upgraded Interrupt Management System to ensure the most stable and consistent system behaviour.
Q: Your Hyper-Speed servers feature accelerated processors, which obviously give a welcome performance increase. But how comfortable are trading firms with deploying servers that are ‘non standard’? And to what lengths do you go to ensure reliability?
A: A more accurate description of the Supermicro Hyper-Speed product line is ‘far above industry standard’ in reliability. The signal integrity was fine-tuned by our design engineering team for every interconnect on the Supermicro X9DAX serverboard that serves as the engine for the Hyper-Speed system, so that generous margins were built-in to ensure highest reliability. The system boasts best-in-class thermals and thus superior stability and reliability with customised heatsinks designed to cool the CPUs nearly 30 degrees Celsius below nominal levels. This thermal headroom ensures enterprise-class reliability and allows the entire server to run at peak Turbo Mode performance indefinitely. Smart stepping features are built into the onboard firmware to accelerate the system gradually until peak performance is reached. Also, remote management is an absolute necessity in the co-location centre, so our onboard Baseboard Management Controller ensures that the system can be monitored and shut off remotely in case trading strategies need to be modified due to unforeseen events. We also burn-in every Hyper-Speed system extensively prior to shipment to guarantee smooth deployments and highest reliability and we back this up with a premium warranty and dedicated support from our HFT support team.
Q: As well as performance, there is increasing focus on cost of operation – such as power and space. How are your servers addressing these requirements?
A: Co-location rackspace is becoming a larger portion of operational expenses for trading firms, so a denser server can halve the costs associated with co-location. The 2U Supermicro Hyper-Speed server can use the fastest Ivy Bridge CPUs available in a dense form factor, while still allowing for sufficient air cooling to keep performance maximised and stabilised. The optimised airflow and Platinum-level 95% efficient power supplies minimize energy usage, allowing customers to optimise their deployments with respect to power limitations of their racks and in their budgets.
Q: What kind of pre-configuration and optimisation do you undertake? How turnkey is deployment of these servers?
A: The components integrated into the Supermicro Hyper-Speed server system are pre-screened for optimal performance, by selecting only the best performing and highest reliability parts to support this premium product line. This requires a significant validation effort and burn-in of the systems that lasts up to 10 days using Supermicro’s proprietary Hyper-Speed test suite to guarantee that these parts will be able to run at the highest performance points reliably in a mission critical environment. In addition, this server is deployed as a complete trading solution that requires no hardware or firmware tuning to achieve the lowest latency and jitter. This contrasts with building your own platform and dealing with the manufacturing variances of all the components, tuning the system, and performing iterative stress tests to make sure the deployment is operating at peak performance and reliability levels. All of this is taken care of with Supermicro’s Hyper-Speed complete server solution, providing customers with peace of mind, proven reliability, and peak performance.
Q: As well as high frequency trading, are these servers being adopted for more heavily compute tasks, such as pre-trade analytics on big data sets?
A: Absolutely, Supermicro has successfully deployed this Hyper-Speed solution into High Performance Computing and Scientific Research environments that take advantage of the elevated processing performance to speed up analytics. We have even delivered this solution to the Engineering and Media & Entertainment verticals to customers who wanted to speed up their systems to take full advantage of the extremely costly software licenses they had purchased for their workstations. This system is perfect for single node workloads that cannot be distributed, since it is the fastest dual processor solution on the market and has set numerous world records in performance benchmarks. For distributed workloads that can be delivered to a cluster of servers, we would recommend our popular MicroCloud, FatTwin and SuperBlade product families for maximum density and manageability.
Q: Given the increasing performance of your servers, and Intel’s future roadmap, are you seeing trading firms looking with less interest at technologies, such as FPGAs?
A: We believe a hybrid approach could be used. By populating FPGA cards in the expansion slots on an x86 server, any upgrade on the x86 server side can be exploited in addition to the customizations on the FPGA. The low-hanging fruit is the upgrade of the server, for example from Supermicro’s Sandy Bridge Hyper-Speed product line to the Ivy Bridge product line, because this requires very little modification of software. Customers will benefit from the latency and jitter upgrades with minimal effort so that they can focus valuable engineering resources on fine-tuning the FPGA firmware or trading software.
Q: Where do you see the next evolution and breakthroughs, in terms of performance and cost of operation of server hardware?
A: Improving speed, stability, density, and flexibility will be the targets for Supermicro’s next-generation Hyper-Speed product line. From an operations perspective, we have a strong focus and commitment to supporting our financial services customers for the long run and this is emphasised by the top-notch services, support, and software packages that are currently being offered. Management software and onsite services are additional areas where Supermicro can further benefit and add value for our customers.
Q: How is Supermicro growing and focusing itself to address the future needs of the trading community?
A: Supermicro is known for its ability to deliver first-to-market solutions utilizing the most advanced server technologies available, application optimised to exactly fit customer’s needs. This fits perfectly with the special requirements of the low-latency trading community where it is an absolute necessity to deploy the latest hardware in order to stay competitive in this leading edge technology-driven industry. Supermicro is increasingly engaged with leading technology vendors and works closely with many customers to test and deploy the newest technologies and industry-specific innovations on our platforms far ahead of the public releases, thereby allowing customers to get a head start on testing their applications and thus extending the lifespans of their trading solutions. We have strong partnerships with all of the low-latency NIC vendors and we are heavily involved with the latest low-latency flash storage technology from Diablo, VDI, SAS3, and NVDIMM solutions, and have a broad line-up of top-notch grid datacenter products available with our FatTwin, SuperBlade, and GPU SuperServer line. You can be assured that if you work with Supermicro, we will be a strong and experienced technology partner now and in the future and you will always have access to the highest performance and most reliable platforms available.