Currently computility shows an explosive growth trend. Heterogeneous software and hardware architecture, super-computility distribution and performance difference bring significant challenges to super-computing grid i...
详细信息
ISBN:
(纸本)9798350379860;9798350379877
Currently computility shows an explosive growth trend. Heterogeneous software and hardware architecture, super-computility distribution and performance difference bring significant challenges to super-computing grid integration. We propose a super-computing grid integration system to access and schedule heterogeneous super-computing, and provide high-quality service. Queue based super-computing virtualization and multi-level scheduling system are adopted respectively to address the problem of heterogeneity of super-computing centers and scheduling system. The system has unique scheduling policies, is theoretically compatible with majority of super-computing scheduling systems like Slurm and also has practicability and extensibility compared with traditional super-computing center. The system supports historical performance evaluation by agent to support applications deployment between architectures and it can optimize scheduling. Example case analysis proves the feasibility of our architecture. And we also find that the computational performance of heterogeneous computing resources needs to be comprehensively analyzed in terms of instruction sets, IO(Input/output), network and specific optimizations.
暂无评论