Title: Hardware-based scheduling and synchronization for light-weight, fine-grained multi-threading

Issue Number: Vol. 10, No. 1
Year of Publication: March - 2020
Page Numbers: 10-22
Authors: Samer Haddad, Jeanine Cook
Journal Name: International Journal of New Computer Architectures and their Applications (IJNCAA)
- Hong Kong

Abstract:


Fine-grained, light-weight multi-threading enhances the performance and scalability of many applications due to its ability to hide memory and network latency and enhance load balance, however, it aggravates the software scheduling and synchronization overhead, which impairs the performance and scalability. In this research, we propose an efficient, dynamic hardware scheduler, which consumes much less time and bandwidth than software schedulers. The proposed scheduler replaces the software application programming interfaces (APIs) and Workers (operating-system-visible threads) with simple hardware that can be integrated on-chip with reasonable silicon area and power budgets. Simulation results show that our proposed hardware scheduling outperforms software scheduling by 22.76 times and it exhibits significant scalability improvement when compared with software scheduling.