Journal of Computations & Modelling

High-performance Cloud Computing for Symbolic Computation Domain

• [ Download ]
• Times downloaded: 10107

• Abstract
  
  

  We investigate the potential benefits of high performance cloud computing (HPCC) for novel application domain namely symbolic computation, and specifically for computational algebra as provided by systems like Maple, Mathematica or GAP. HPCCs potentially offer the computational power of a large number of hosts, flexible configuration, and ease of access to a specialized, high-performance configuration. Computational algebra deals with the symbolic manipulation of mathematical problems, often and many algebraic computations are time consuming and therefore promising candidates for parallelism. However, the nature of these computations is fundamentally different to classic high-performance scientific computation: many are highly dynamic, use complex recursive data structures, exhibit high degrees of irregularity, generate large intermediate data structures and primarily use arbitrary precision scalars rather than floating point values. We present an initial study on how to use existing HPCC frameworks for computational algebra. We port a C+MPI parallel implementation of a representative computational algebra application, the parallel determinisation of a non-deterministic finite state automaton, to an HPCC and evaluate the performance on several cloud infrastructures. The key issies for the HPCC implementation are the efficient management of massive intermediate data structures, up to 1.1TB, and fast access to the file system in order to store such big data structures.