1993 Fiscal Year Final Research Report Summary
Research on Massively Parallel Operating Systems with Flexibility
| Project/Area Number |
04680034
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
Informatics
|
|---|
| Research Institution | KYUSHU UNIVERSITY |
|---|
Principal Investigator |
FUKUDA Akira Kyushu University, Faculty of Engineering, Associate Professor, 工学部, 助教授 (80165282)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
MURAKAMI Kazuaki Kyushu University, Interdisciplinary Graduate School of Engineering, Associate P, 大学院・総合理工学研究科, 講師 (10200263)
SAISHO Keizo Kyushu University, Computer Center, Associate Professor, 大型計算機センター, 助教授 (50170486)
YOSHIDA Norihiko Kyushu University, Faculty of Engineering, Associate Professor, 工学部, 助教授 (00182775)
ARAKI Keijirou Nara Institute of Science and Technology, Graduate School of Information Scence,, 情報科学研究科, 教授 (40117057)
|
|---|
| Project Period (FY) |
1992 – 1993
|
| Keywords | Parallel Operating Systems / Micro-kernel / Parallelizing Compiler / Two-level Scheduling / Data Partitioning Graph |
|---|
| Research Abstract |
1. Parallel Operating Systems
We showed that promising parallel operating system architecture is microkernel-baced one, where kernel provides the minimum functions and out-of-kernel functions are provided by system servers implemented in user spaces. We developed parallel operating system prototype according to the architecture. The prototype was implemented on a computer. Through running some programs, we confirmed that proposed architecture was effective.
Environments of Developing Parallel Operating System
We showed that it is effective to model parallel operating systems as process network. We also studied debugging methods. We employed slicing method in programming languages into formal languages.
Compiler
Through simulation experiments, we showed that two-level scheduling was promising one for large-scaled multiprocessors. In addition, we investigated alternative policies in the scheduling. The simulation results said that these performance greatly depend on multiprocessor memory architectures. We also studied parallelizing compilers considering non-uniformity of memory accesses. We proposed intermediate representation called DPG(Data Partitioning Graph).
|
