Adaptive Distributed Computing Middleware for Computational Finance Applications

Parallel application programming is becoming increasingly impor- tant to researchers and developers who want to leverage hardware advances and compute complex calculations quickly and effic iently in competitive domains, such as financial services. Prior re search has focused on high-performance computing and grid computing middleware to support the development of parallel applications. These technologies, however, are hard to program and manage due to the need for researchers and developers to use explicit concur- rency mechanisms and manual lifecycle management techniques. What is needed therefore, are middleware technologies and tools that are both easy to use and highly optimized for the new genera- tion of parallel computing hardware. This paper presents the structure and functionality of Zirc omp, which is adaptive distributed computing middleware targeted at parallel applications on heterogeneous computing environments, such as laptops, desktops, single-core CPUs, multi-core CPUs, clus- ters, and clouds. Zircomp allows developers to design software as if they are programming for a single computer by automatically distributing and parallelizing task executions scalably, reliably, and resource-efficiently. The result is a straightforward prog ramming model that can substantially improve the performance of complex computational finance applications via distribution and pa ralleliza- tion. This paper uses a representative case study from the fin ancial services domain to show how these types of applications can bene- fit from using Zircomp.