Aggregates: using design patterns to create implicitly paralleldata structures in C++
Hudak, D.E.; Baughman, N.; Hodges, G.
Aerospace and Electronics Conference, 1997. NAECON 1997., Proceedings of the IEEE 1997 National
Volume 1, Issue , 14-18 Jul 1997 Page(s):239 - 246 vol.1
Digital Object Identifier 10.1109/NAECON.1997.618085
Summary:Small-scale parallel platforms have become prevalent in the
marketplace. These machines feature globally shared address spaces,
complex kernels and an emphasis on throughput-based parallelism. These
new platforms require new tools for concurrent programming. Programs
written for these new platforms must be efficient, portable and
adaptable to differing machine conditions. In addition, programming
environments for these new platforms should integrate concurrency into
the design process in order to exploit the multiple processors for
speedup when sufficient workloads arise. This functionality can be
effectively embedded at the framework/API level. Efficient design of
these features can be facilitated using object-oriented design
techniques known as design patterns. To illustrate these concepts, we
developed a framework called aggregates. Aggregates are a suggested
replacement for arrays, linked lists, or other structures for holding
large collections of C++ objects. Aggregates use run-time partitioning
to support concurrent application of a member function to all objects in
the aggregate. Experimental results on a single-processor Windows NT
system and a multiprocessor Silicon Graphics Power Challenge have
demonstrated a single program using aggregates performed comparably to
traditional arrays and linked lists in a single processor environment
while providing speedup in a multiple processor environment
View citation and abstract |
Cart
