2022 Fiscal Year Final Research Report
Seamless Unified Memory System over the Distributed Memories on Cluster Computing Nodes
| Project/Area Number |
18K11327
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 60090:High performance computing-related
|
|---|
| Research Institution | Seikei University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2018-04-01 – 2023-03-31
|
| Keywords | 分散共有メモリ / 遠隔メモリ / クラスタ / 仮想共有メモリ / マルチノード並列処理 / PGAS / 分散メモリ / メモリアドレス空間 |
|---|
| Outline of Final Research Achievements |
We have developed a run-time system mSMS that realizes a huge global shared memory (large-scale virtual address space) in a distributed memory type massively parallel computing system. We have constructed a programming environment that can access data in the same manner regardless of local/remote memory. In typical scientific computing, the programs in mSMS are easier to create and have higher or equal performance than conventional message-passing (MPI) programs. For large-scale N-body problems that cannot fit in the memory of a single computing node, a large-scale global tree can be constructed in a mSMS huge shared memory and address-pointer access is available in typical Barnes-Hut programs. The programs in mSMS are much easier to develop compared to MPI programs.
The mSMS achieaved comparable or better performance compared to existing MPI programs, and it dramatically improved the program development productivity on a distributed memory computing system.
|
| Free Research Field |
高性能計算
|
| Academic Significance and Societal Importance of the Research Achievements |
スーパーコンピュータでは，複数の計算ノードをネットワークで接続し，並列計算により高速処理を行なっている，しかし，計算ノードはそれぞれ独立のメモリとアドレス空間を持ち，他の計算ノードのメモリにあるデータ（遠隔データ）をアクセスするには，特別なプログラムインターフェースを必要とする．多数の計算ノードを用いた並列処理では，プログラム記述は非常に複雑でプログラム開発は大きな負担となっている． このようなシステムにおいて，プログラム開発の生産性を高め，大域共有メモリ（大規模仮想アドレス空間）を実現するmSMSを開発し，ローカル／リモートメモリの区別なくデータにアクセスできるプログラミング環境を構築した．
|
