Chi Xue-bin

Large-Scale Parallel Simulation of High-Dimensional American Option Pricing

High-dimensional American option pricing is computationally challenging in both theory and practice. We use stochastic mesh method combined with performance enhancement policy of bias reduction to solve this practical problem in classic Black-Scholes framework. We effectively parallelize this algorithm through splitting the generated mesh by row among processors, use MPI for efficient implementation, and perform large-scale numerical experiments on heterogeneous supercomputer DeepComp7000. Numerical results of parallel simulation demonstrate that parallel simulation has good scalability in different parallel environments of DeepComp7000; large-scale parallel simulation can obtain much better speedup. The convergent performance is also empirically demonstrated. The estimated option value converges with the increase of mesh size; when using smaller mesh size, the stochastic mesh method with bias reduction can underestimate the true American option value.

The Applications and Trends of High Performance Computing in Finance

Large-scale parallel simulation and modeling have changed our world. Today, supercomputers are not just for research and scientific exploration; they have become an integral part of many industries, among which finance is one of the strongest growth factors for supercomputers, driven by ever increasing data volumes, greater data complexity and significantly more challenging data analysis. In this paper, a modest application of the developments of high-performance computing in finance is studied deeply. Attentions are not only focused on the what benefits the parallel algorithm bring to the financial research, but also on the practical applications of the High-Performance Computing in real financial markets, especially some recent advances is highlighted. On that basis, some suggestions about the challenges and development directions of HPCs in finance are proposed.

Developing high performance BLAS, LAPACK and ScaLAPACK on HITACHI SR8000

We introduce the development of high performance BLAS, LAPACK and ScaLAPACK on HITACHI SR8000. The methods and techniques are introduced and the final results are listed. According to these results, the careful tuning of a mathematical software library on one specific computing platform is very important and we can exploit a large fraction of its potential performance through utilizing its special characteristics.

Parallel algorithm design on some distributed systems

Some testing results on DAWNING-1000, Paragon and workstation cluster are described in this paper. On the home-made parallel system DAWNING-1000 with 32 computational processors, the practical performance of 1.117 Gflops and 1.58 Gflops has been measured in solving a dense linear system and doing matrix multiplication, respectively. The scalability is also investigated. The importance of designing efficient parallel algorithms for evaluating parallel systems is emphasized.

Parallel Computing for Dynamic Asset Allocation Based on the Stochastic Programming

In this paper, a multi-stage stochastic programming model is constructed, for the dynamic asset allocation with the transaction cost constraints. In the mean time in order to improve the performance, the Conditional Value-at-Risk as the risk measure, which is a very important concept in the modern risk management field, is also contained. However, with the increase of the number of scenarios, the number of constrains and decisions variable is increasing dramatically. It turns out that the memory management is a major bottleneck when solving planning problems. For this reason, this paper shows that the dedicated model generations, and the specialized solution techniques based on high performance computing, are the essential elements to tackle this large-scale financial planning. The parallel code is programmed by the C language, and the Message Passing Interface (MPI) for communication is utilized. The parallel and financial performance is performed on the DeepComp7000.

Data Processing in Space Weather Physics Models in the Meridian Project

In the Meridian Project, two space weather physics models, L1-magnetosphere-ionosphere causal chain(L1 model), and numerical magnetosphere database service, are provided as e-Science application services through the space weather computation grid environment, which provides a web-based portal for space weather studies users. We call the computation grid environment as a computing gateway. The computing gateway integrates space weather applications, space weather data, and tools. For user to use the computing gateway, the first thing that user will up against is where the input files will be located to complete the computation. Whether they are just come from user local machine, and how can I use the uploaded input data files again in the following new computing tasks? In this paper, we introduce a data space concept—the user data view space, within which the space weather data files shared between the resource entities in the space weather computation grid environment. As the size of computation data results is massive and they are raw data about magnetosphere and ionosphere, for intuitivism to show the physics meaning of the magnetosphere and ionosphere data, the raw computation result data will be post-processed by visualization software. Then how to draw graphics on the visualization node inside a parallel job? And more in numerical magnetosphere database service physics model, according to task’s designated parameters we should choose a proximate data as the initial condition from numerical magnetosphere database, and then do the computation simulation by using the L1 model.

AINV AND BILUM PRECONDITIONING TECHNIQUES

It was proposed that a robust and efficient parallelizable preconditioner for solving general sparse linear systems of equations, in which the use of sparse approximate inverse (AINV) techniques in a multi-level block ILU (BILUM) preconditioner were investigated. The resulting preconditioner retains robustness of BILUM preconditioner and has two advantages over the standard BILUM preconditoner: the ability to control sparsity and increased parallelism. Numerical experiments are used to show the effectiveness and efficiency of the new preconditioner.

Research of Acceleration MS-Alignment Identifying Post-Translational Modifications on GPU

MS-Alignment is an unrestrictive post-translational modification (PTM) search algorithm with an advantage of searching for all types of PTMs at once in a blind mode. However, it is time-consuming, and thus it could not well meet the challenge of large-scale protein database and spectra. We use Graphic Processor Unit (GPU) to accelerate MS-Alignment for reducing identification time to meet time requirement. The work mainly includes two parts. (1) The step of Database search and Candidate generation (DC) consumes most of the time in MS-Alignment. We propose an algorithm of DC on GPU based on CUDA (DCGPU). The data parallelism way is partitioning protein sequences. We adopt several methods to optimize DCGPU implementation. (2) For further acceleration, we propose an algorithm of MS-Alignment on GPU cluster based on MPI and CUDA (MC_MS-A). The comparison experiments show that the average speedup ratio could be above 26 in the model of at most one modification and above 41 in the model of at most two modifications. The experimental results show that MC_MS-A on GPU Cluster could reduce the time of identifying 31173 spectra from about 2.853 months predicted to 0.606 h. Accelerating MS-Alignment on GPU is applicable for large-scale data requiring for high-speed processing.

Parallel implementation of linear algebra problems on Dawning-1000

In this paper, some parallel algorithms are described for solving numerical linear algebra problems on Dwning-1000. They include matrix multiplication,LU factorization of a dense matrix, Cholesky factorization of a symmetric matrix, and eigendecomposition of symmetric matrix for real and complex data types. These programs are constructed based on fast BLAS library of Dawning-1000 under NX environment. Some comparison results under different parallel environments and implementing methods are also given for Cholesky factorization. The execution time, measured performance and speedup for each problem on Dawning-1000 are shown. For matrix multiplication andLU factorization, 1.86GFLOPS and 1.53GFLOPS are reached.