Dynamic scheduling Monte-Carlo framework for multi-accelerator heterogeneous clusters

Abstract

Monte-Carlo (MC) simulation is an effective tool for solving complex problems such as many-body simulation, exotic option pricing and partial differential equation solving. The huge amount of computation in MC makes it a good candidate for acceleration using hardware and distributed computing platforms. We propose a novel MC simulation framework suitable for a wide range of problems. This framework enables different hardware accelerators in a multi-accelerator heterogeneous cluster to work collaboratively on a single application. It also provides scheduling interfaces to adaptively balance the workload according to the cluster status. Two financial applications, involving asset simulation and option pricing, are built using this framework to demonstrate its capability and flexibility. A cluster with 8 Virtex-5 xc5vlx330t FPGAs and 8 Tesla C1060 GPUs using the proposed framework provides 44 times speedup and 19.6 times improved energy efficiency over a cluster with 16 AMD Phenom 9650 quad-core 2.4GHz CPUs for the GARCH asset simulation application. The Efficient Allocation Line (EAL) is proposed for determining the most efficient allocation of accelerators for either performance or energy consumption.

High Performance and Low Power Monte Carlo Methods to Option Pricing Models via High Level Design and Synthesis

“…The clock frequency f clock = 125MHz implemented in this work and f clock = 200MHz in [5]. According to the resource utilization and the clock frequency, one can conclude that the FPGA power from this work is not more than that from the literature [5]. Even if the device powers of both works are assumed to be identical, the energy consumption can be considered roughly proportional to the execution time T s .…”

“…Of course, the result of any comparison between GPU and FPGA highly depends on the algorithms and the technology node used to manufacture the devices. For instance, Anson et al [5] demonstrated that an FPGA outperformed GPU in terms of both performance and energy, by implementing the MC method to Asian option using the Heston model. Several HLS techniques have been implemented as both research and commercial tools in recent years.…”

“…In (4), z 1 , z 2 are two Wiener processes, ρ is the correlation factor between them, and √ V is the volatility of the stock price. In (5), θ is the long-run mean variance, κ is the speed of mean reversion( the rate at which V reverts to θ ) and σ v is the volatility (Standard deviation) of the volatility V .…”

“…1) Asian Option, Black-Scholes Model: For this model evaluation, the same parameter values and FPGA platform as [5] have been implemented in order to meaningfully compare performance. Table V shows T s and the resource utilization.…”

“…Table V shows T s and the resource utilization. Even though the resource balancing has not been applied to this system in this study, the HLS-based implementation is still about 2.4X as fast as the one presented in [5], which modeled the algorithms in RTL.…”