We propose a novel parallelcomputing that allows processors to access data in predictable time without the need to access it from different locations in memory using addresses. It uses orbital data that is mapped to ...
详细信息
ISBN:
(纸本)9783031762727;9783031762734
We propose a novel parallelcomputing that allows processors to access data in predictable time without the need to access it from different locations in memory using addresses. It uses orbital data that is mapped to time and is made available to multiple processors at the same time in multiple different orbits and at a specific predictable time in each orbit. This allows processors in different orbits to share the same data, eliminating the problem of sharing data at the same time among multiple processors. It provides processors with the ability to hide the waiting time when accessing shared data by overlapping it with useful work on another data while allowing other processors to work on the shared data in another orbit. The performance of this novel method shows significant improvements in scalability compared to that of conventional parallelcomputing.
暂无评论