|
Linear based algorithms are used by researchers the world over for
solving complex systems of equations, looking for best fit solutions
to problems, identifying characteristic features of data and manipulating
data to a more user-friendly form. Many of these algorithms employ
linear algebraic operations such as singular value decomposition
(SVD), eigenvalue and eigenvector de composition (eig), lower-upper
matrix decomposition (LU), matrix inversion, multiplication and
transposition, to name a few.
Many linear algebra applications increasingly involve either larger
datasets or simply many more datasets. In each case, the computational
load increases dramatically, limiting throughput when processing
large datasets or many matrices on a serial computational platform.
When handling many small matrices, task parallel computation on
a cluster of processors would be ideal—but how can one easily distribute
the work? Similarly challenging is the implementation of data parallel
computations, which often require C or Fortran, and MPI coding to
implement the inter-processor communication demanded by the linear
algebra algorithms.
With Star-P, researchers can quickly take advantage of both data
parallel and task parallel line ar algebra to solve their problems
in a fraction of the time. Star-P lets us ers develop and optimize
their applications in a familiar environment, such as MATLAB®®
from The MathWorks, while seamlessly computing interactively on
a multiprocessor server. With the native MATLAB® linear algebra compatibility,
Star-P readily enables simple conversion of existing serial MATLAB®
code for parallel computation shaving precious execut ion and development
time.
Data Parallel Computations
- Minimal modification of MATLAB® code
to take advantage of powerful parallel computing
- No need to program in C/C++ with MPI
to take advantage of parallel computation
- Scalable, high-speed parallel processing
using linear algebra techniques readily achieved
|
Star-P Support
Blog
Email
Page
Summary