Si-Qing Zheng - US grants

High performance computing (HPC) is a major initiative at Louisiana State University (LSU). Its multidisciplinary research programs focus on grand challenges in materials science, astronomy, microsystems design, and environmental studies. To advance this effort, we request five NSF Traineeships. Consortial arrangements have been made to recruit in-state and out-of-state students, especially minorities from Historically Black Institutions. Innovative training structures have been introduced to integrate HPC into graduate education and research. These include new cross-disciplinary curricula and a dual-degree program to allow Ph.D. students in the physical sciences to get a M.S. from Computer Science and vice versa. Unique opportunities are available for graduate trainees to interact with scientists at Oak Ridge and Argonne National Laboratories, Mobil and Ethyl Corporations, and in Europe and Japan. To establish the HPC effort a LSU, two excellent parallel computing laboratories have been set up with $2. million in infrastructure grants from the State of Louisiana. These laboratories feature an 8,192-node MasPar, a 64-cell iWarp, and 8-node iPSC/860, and an 8-processor Silicon Graphics. With the latest grant of $850,000, we plan to acquire a 56-node Intel Paragon.

Switching networks serve as the core of network switches and routers, and communication
subsystems of multiprocessor and multi-computer computing systems. There has been a
growing interest in developing high-speed switches for multicasting. A wide class of
applications, such as teleconferencing, video distribution, LAN bridging and distributed
data processing, require multicasting communications. The essential network components
for such applications are multicast switches. Rearrangeable nonblocking (RNB) multicast
switching networks are used for packet switching, and strictly nonblocking (SNB) and
wide-sense nonblocking (WSNB) multicast switching networks are used for circuit switching.
Optimal RNB multicast switching networks have been constructed for a long time, and it
was shown that the lower bound for the cost SNB multicast switching network is O(N^2)
in 1980. Thus, for scalable circuit switching applications minimum-cost WSNB multicast
switching networks are the only solution. Multicast networks (a.k.a. generalized
connection networks, generalized connectors, distribution networks) were first formally
introduced in English literature by Masson and Jordan in 1972. Until now, the WSNB
multicast switching network with lowest cost remains to be Pippenger's network designed
in 1973. It has been a general belief that Pippenger's network is very difficult to
improve. One of the major goals of this proposed research is to seek WSNB multicast
switching networks better than Pippenger's network. In addition, we also consider
constructing optimal or near-optimal nonblocking multicast switching networks under
routability and optical implementation constraints. All problems considered are
constrained optimization problems with important implications. This proposed work not
only has significant theoretical importance but is also of practical value.
In the last 33 years, no progress on WSNB multicast switching networks better than
Pippenger's network has been reported. The intellectual merit of the proposed research
lies in the attempt of tackling a long-standing open theoretical problem, and its related
problems arising in new network technologies.