The writer giving his presentation during SWoPP ASO '97
Visitor: Makoto Ando (Parallel and Distributed System Performance TRC Laboratory) Period: August 18-22, 1997 (5 days) Conferences: HPCS '97 (High-performance Computing Symposium 1997) SWoPP ASO '97 (1997 "Country of Fire" Summer Workshop on Parallel/Distributed/Cooperative Processing)
HPCS '97 and SWoPP ASO '97 were held in Aso, Kumamoto in late August under the lingering summer heat. I had been a little worried about the heat in Kumamoto, but the conference site, Greenpia Minami Aso, located at the somma of Mt. Aso, was unexpectedly comfortable.
HPCS '97
HPCS '97 (High-performance Computing Symposium 1997) was a symposium sponsored by the Special Interest Group of High Performance Computing of the Information Processing Society of Japan held on August 18. During the morning session, two tutorials were given: RISC Computers and HPC Programming, and the Recent Trend in HPF Language. Then during the afternoon session, the invited speakers addressed various fields from application-level subjects such as fluid analysis with parallel computers to architecture-level subjects such as run-time parallelization of programs. In particular, the overview of technologies for high-performance computing was of great significance to me.
SWoPP ASO '97
SWoPP is a yearly summer workshop on parallel/distributed/cooperative processing and serves as a congregational research conference, where researchers in the parallel/distributed processing field in Japan meet together for intense discussions. This summer, the tenth SWoPP was held in its birthplace, Aso, from August 19 to 22.
Six research sessions were held at the same time by the Information Processing Society of Japan and the Institute of Electronics Information and Communication Engineers with "parallel/distributed/cooperative processing" as a common keyword. There were 127 presentations in total (97 by universities, 14 by private companies, 16 by national institutes and other institutes including RWCP) with 265 participants (202 from 45 universities, 36 from 11 private companies, and 27 from 4 national institutes and other institutes). There were 11 presentations from RWCP (7 by TRC, 2 by Distributed Laboratories, and 2 by Trainees).
I made a presentation titled, "Performance Comparison of Shared- and Distributed-memory Parallel Computers Using Knapsack Problem" during the "Performance Evaluation" session under the Special Interest Group of High Performance Computing of the Information Processing Society of Japan in the afternoon on the first day. In this session, chief researcher Dr. Tanaka (Parallel and Distributed System Performance TRC Laboratory) also made a presentation titled "A Hybrid Shared Memory/Message Passing Programming on SMP Cluster". Many researchers in parallel/distributed processing area paid attention to this session, and the audience at our session site was quite large.
The research I presented was on characterization of the performance of shared-memory and distributed-memory parallel computers using the knapsack problem, a typical example of a tree-search problem. Generally, in solving tree-search problems, the size of search space and the parallelism of search processing depend heavily on input data. My presentation stressed the performance analysis of parallel computers systematically by first classifying input data into several categories and then analyzing it according to these categories. This led to active and extensive discussion during the Q&A session following my presentation on performance evaluation methods for parallel computers, solution algorithms for tree-search problems, and their parallelization. In addition, during the social session held in the evening on the second day, I had a chance to exchange opinions with researchers from various research institutes, this session was quite a useful stimulus to me.
At this SWoPP, much research was presented on technologies for configuring clusters. Specifically, this includes research on high-performance networks for connecting nodes (such as workstations, PCs, and SMPs), performance evaluation of existing communication methods on Myrinet and proposals for new communication methods, and scheduling of programs on clusters. I realized that cluster-related technologies were important subjects which researchers on parallel/distributed processing pay attention to.
In addition to cluster technologies, I spent a significant five-day period attending sessions on distributed shared memory, numerical computation, and applications.