Our Computational Process Network framework makes it possible
to implement scalable, high-throughput signal processing systems
on multicore desktop computers and across multiple machines.
This figure below shows scalability results for a 3D circular
convolution beamformer that processes over 600 MB/s of data and
nearly 25 GFLOPS.
With only minor modification, the same program is executed on
two different target systems: a shared memory multi-core server
with 12 cores, and a distributed memory 8-host cluster.
The software release is
freely distributable.
More information is available in Greg Allen's PhD dissertation
entitled Computational Process Networks: A Model and Framework for High-Throughput Signal Processing.
Mail comments about this page to
bevans@ece.utexas.edu.
