The design, implementation, and evaluation of a symmetric banded linear solver for distributed-memory parallel computers

Abstract

This article describes the design, implementation, and evaluation of a parallel algorithm for the Cholesky factorization of symmetric banded matrices. The algorithm is part of IBM's Parallel Engineering and Scientific Subroutine Library version 1.2 and is compatible with ScaLAPACK's banded solver. Analysis, as well as experiments on an IBM SP2 distributedmemory parallel computer, shows that the algorithm efficiently factors banded matrices with wide bandwidth. For example, a 31-node SP2 factors a large matrix more than 16 times faster than a single node would factor it using the best sequential algorithm, and more than 20 times faster than a single node would using LAPACK's DPBTRF. The algorithm uses novel ideas in the area of distributed dense-matrix computations that include the use of a dynamic schedule for a blocked systolic-like algorithm and the separation of the input and output data layouts from the layout the algorithm uses internally. The algorithm also uses known techniques such as blocking to improve its communication-to-computation ratio and its data-cache behavior.

Trading Replication for Communication in Parallel Distributed-Memory Dense Solvers

“…That is, the algorithms are asymptotically work-efficient only with smaller numbers of processors, P = O(nm/r 2 ) for the triangular solver and P = O(n 2 /(r 2 log n)). This behavior is common to linear-algebra algorithms with long critical paths, such as triangular solvers by substitution and triangular and orthogonal factorizations (see, e.g., [9]). …”

Cache-optimal algorithms for option pricing

“…For computations, where data is reused many times, this technique reduces memory traffic to slower memories in the hierarchy [Hennessy and Patterson 2007]. The cache blocking technique has been extensively applied to linear algebra applications [Dongarra et al 1990;Anderson et al 1999;Kågström et al 1998;Gupta et al 1998;Goto and van de Geijn 2008;Agarwal et al 1994a]. Since accessing data from a slower memory is expensive, an algorithm that rarely goes to slower memory performs better.…”

Hypergrammar-Based Parallel Multi-Frontal Solver for Grids With Point Singularities

“…This allows for a reduction of computational cost related to the construction of the elimination tree. Different implementations of the multi-frontal solver algorithm also exist that target specific architectures (see, e.g., [13,14,15]). There is also a linearly-computational cost direct solver based on the use of H-matrices [27] with compressed non-diagonal blocks.…”