New Xeon Phi-based system delivers a 30-fold speed up in calculation time
“Sharing joy” with its clients in the form of outstanding financial returns, is one of the stated goals of Mizuho Securities Co. The institution is a member of the Mizuho Financial Group, one of Japan’s three major megabanks.
Hiroshi Motoyama, President and CEO of Mizuho Securities, in his announcement last year of the company’s Medium-Term Business Plan, indicated that increasing the company’s overall capabilities is a key part of realizing this objective. And that includes enhancing its industry-leading financial technology – including HPC.
In particular, the company’s Credit & Derivative Trading Department, a part of the Fixed Income Group, was interested in streamlining their operations to take advantage of global financial opportunities. Low interest rates have ramped up the demand for structured bonds, a pre-packaged investment based on a combination of bonds and derivatives.
Takashi Akahane, Director of the Credit & Derivative Trading Department, explains, “It is necessary to solve complicated stochastic differential equations to evaluate a derivative.”
He says that the huge amount of calculations required by Derivative Valuation Systems to obtain the indicative pricing of the derivative were taking too much time to meet their customer’s needs. Also, the calculation of risk values – something that traders must have to manage their portfolio – was also prolonged by the increase in transactions, taking as much as a full day to complete.
Akahane and his team of quants investigated several options to boost their processing power, including NVIDIA Tesla technology and the powerful Fujitsu K Computer. They determined that the NVIDIA solution, because of its complexity, would require too much development time. The K computer was not only expensive, but it also featured a new architecture that was unfamiliar to the Mizuho Department.
The quants had already developed an evaluation module of the derivatives on an 8-core Intel Xeon platform. The computational demands of the fast-growing structured bond marketplace led the Mizuho team to investigate the use of high performance Intel Xeon Phi coprocessors. The team, with help from Fixstars, their development partner, created the necessary algorithms to take full advantage of the Xeon Phi’s highly parallel processing capabilities.
“Xeon Phi looked easier than the other platforms,” says Akahane. “Actually, our quants developed a prototype module in six months – that’s very quick.”
The Department began using its new system in a production environment in March, 2014. Each calculation node consists of a Linux server with four Xeon Phi coprocessors based on the Intel Many integrated Core (MIC) Architecture. The systems grid computing architecture is based on IBM Platform Symphony.
The application running on the system evaluates interest and foreign exchange (FX) derivatives. The system is currently calculating “plain vanilla” derivatives – the standard, most basic version of a financial instrument. Structured bonds – bonds embedded with exotic derivatives – are slated to be up and running this summer.
“Moving to the Intel MIC Architecture cut our server costs drastically,” Akahane comments. “We estimate that the cost to other financial institutions with this big a system is more than 10 times the cost of our system.”
He adds that compared to running the algorithms on the previous Xeon-based platform, the new Xeon Phi-based system has provided a 30-fold speed up in calculation time.
But Akahane is far from finished – he has new goals that will bring more joy to Mizuho’s clients.
For example, “Our goal is to build a system that finishes all calculations in real time plus one second,” he says. The high-speed Xeon Phi coprocessor will help him make that ambition a reality.