Direct device-to-device transfer protocol: A new look at the benefits of a decentralized I/O model

作者: Steen Larsen , Ben Lee , Jin-Hyuk Yoon , Jae-Yeun Yun

DOI: 10.1109/NAS.2015.7255202

关键词:

摘要: Current I/O devices communicate based on the PCIe protocol, and by default, all traffic passes through CPU-memory complex. However, this approach causes bottleneck in system throughput, which increases latency power as CPU processes device specific protocols to move data between devices. This paper examines cost of centralized proposes a new method perform direct device-to-device communication. Our proof-of-concept implementation using NetFPGA shows that can be reduced more than 2×, utilization up 18%, decreased 31 W.

参考文章(4)
Steen Larsen, Parthasarathy Sarangam, Ram Huggahalli, Siddharth Kulkarni, Architectural Breakdown of End-to-End Latency in a TCP/IP Network symposium on computer architecture and high performance computing. ,vol. 37, pp. 556- 571 ,(2007) , 10.1007/S10766-009-0109-6
P. Carns, R. Ross, W. Ligon, P. Wyckoff, BMI: a network abstraction layer for parallel I/O international parallel and distributed processing symposium. pp. 213- ,(2005) , 10.1109/IPDPS.2005.128
Nawab Ali, Philip Carns, Kamil Iskra, Dries Kimpe, Samuel Lang, Robert Latham, Robert Ross, Lee Ward, Ponnuswamy Sadayappan, None, Scalable I/O forwarding framework for high-performance computing systems international conference on cluster computing. pp. 1- 10 ,(2009) , 10.1109/CLUSTR.2009.5289188
Kamil Iskra, John W. Romein, Kazutomo Yoshii, Pete Beckman, ZOID Proceedings of the 13th ACM SIGPLAN Symposium on Principles and practice of parallel programming - PPoPP '08. pp. 153- 162 ,(2008) , 10.1145/1345206.1345230