P-Socket: Optimizing a communication library for a PCIe-based intra-rack interconnect
Paper in proceedings, 2016

Data centers require efficient, low-cost, exible interconnects to manage the rapidly growing internal traffic generated by an increasingly diverse set of applications. To meet these re-quirements, data center networks are increasingly employing alternatives such as RapidIO, Freedom, and PCIe, which re-quire fewer physical devices and/or have simpler protocols than more traditional interconnects. These networks offer raw high performance communication capabilities, but sim-ply using them for conventional TCP/IP-based communica-tion fails to realize the potential performance of the physical network. Here we analyze causes for this performance loss for the TCP/IP protocol over one such fabric, PCIe, and we explore a hardware/software solution that mitigates over-heads and exploits PCIe's advanced features. The result is P-Socket, an efficient library that enables legacy socket applications to run without modification. Our experiments show that P-Socket achieves an end-To-end latency of 1.2s and effiective bandwidth of up to 2.87GB/s (out of a theo-retical peak of 3.05GB/s).

PCIe

data-center servers

rack interconnects

sockets

Author

Liuhang Zhang

Institute of Computing Technology Chinese Academy of Sciences

Rui Hou

Institute of Computing Technology Chinese Academy of Sciences

Sally A McKee

Chalmers, Computer Science and Engineering (Chalmers), Computer Engineering (Chalmers)

Jianbo Dong

Institute of Computing Technology Chinese Academy of Sciences

Lixin Zhang

Institute of Computing Technology Chinese Academy of Sciences

ACM International Conference on Computing Frontiers, CF 2016; Como; Italy; 16 May 2016 through 18 May 2016

145-153

Subject Categories

Computer Engineering

DOI

10.1145/2903150.2903168

ISBN

978-145034128-8