Search

The time to solve linear systems depends to a large extent on the choice of the solution method and the properties of the coefficient matrix. Although there are several linear solution methods, in most cases it is impossible to predict apriori which linear solver would be best suited for a given linear system. Recent investigations on selecting linear solvers for a given system have explored the use of classification techniques based on the linear system parameters for solver selection. In this paper, we present a method to develop low-cost high-accuracy classifiers. We show that the cost for constructing a classifier can be significantly reduced by focusing on the computational complexity of each feature. In particular, we filter out low information linear system parameters and then order the remaining parameters to decrease the total computation cost for classification at a prescribed accuracy. Our results indicate that the speedup factor of the time to compute the feature set using our ordering can be as high as 262. The accuracy and computation time of the feature set generated using our method is comparable to a near-optimal one, thus demonstrating the effectiveness of our technique. [ABSTRACT FROM AUTHOR]

Graph coloring is used to identify independent objects in a set and has applications in a wide variety of scientific and engineering problems. Optimal coloring of graphs is an NP-complete problem. Therefore there exist many heuristics that attempt to obtain a near-optimal number of colors. In this paper we introduce a backtracking correction algorithm which dynamically rearranges the colors assigned by a top level heuristic to a more favorable permutation thereby improving the performance of the coloring algorithm. Our results obtained by applying the backtracking heuristic on graphs from molecular dynamics and DNA-electrophoresis show that the backtracking algorithm succeeds in lowering the number of colors by as much as 23%. Variations of backtracking algorithm can be as much as 66% faster than standard correction algorithms, like Culberson΄s Iterated Greedy method, while producing a comparable number of colors. [ABSTRACT FROM AUTHOR]

Component- and service-based software engineering approaches have been gaining popularity in high-performance scientific computing, facilitating the creation and management of large multidisciplinary, multideveloper applications, and providing opportunities for improved performance and numerical accuracy. These software engineering approaches enable the development of middleware infrastructure for computational quality of service (CQoS), which provides performance optimizations through dynamic algorithm selection and configuration in a mostly automated fashion. The factors that affect performance are closely tied to a component's parallel implementation, its management of parallel communication and memory, the algorithms executed, the algorithmic parameters employed, and other operational characteristics. We present the design of a component middleware CQoS architecture for automated composition and adaptation of high-performance componentor service-based applications. We describe its initial implementation and corresponding experimental results for parallel simulations involving time-dependent nonlinear partial differential equations. [ABSTRACT FROM AUTHOR]