Implementation and Evaluation of Network Interface and Message Passing Services for TianHe-1A Supercomputer

Author

Xie, Min ; Lu, Yutong ; Liu, Lu ; Cao, Hongjia ; Yang, Xuejun

Author_Institution

Sch. of Comput., Nat. Univ. of Defense Technol., Changsha, China

fYear

2011

Firstpage

78

Lastpage

86

Abstract

TianHe-1A, a large scale supercomputer features hybrid multi-core CPU and GPU computing, utilizes a proprietary high-bandwidth, low-latency interconnect fabric to achieve the optimized balance of computation and communication capabilities. In this paper, we describe the design of the network interface in TianHe-1A which supports user-level communication and offloads collective operation to deliver communication performance close to the hardware limits. Scalable and high performance message passing services are implemented on this interface and preliminary performance results are provided.

Keywords

mainframes; message passing; parallel machines; GPU computing; TianHe-1A supercomputer; communication capabilities; computation capabilities; hybrid multicore CPU; low-latency interconnect fabric; message passing services; network interface; offloads collective operation; user-level communication; Kernel; Message passing; Optical switches; Protocols; Radiation detectors; Receivers; Collective Communication; MPI; RDMA; user-level communication;

fLanguage

English

Publisher

ieee

Conference_Titel

High Performance Interconnects (HOTI), 2011 IEEE 19th Annual Symposium on

Conference_Location

Santa Clara, CA

Print_ISBN

978-1-4577-1563-1

Electronic_ISBN

978-0-7695-4537-0

Type

conf

DOI

10.1109/HOTI.2011.20

Filename

6041537