Research Article
Testbed and Experiments for High-Performance Networking
@INPROCEEDINGS{10.1007/978-3-642-17851-1_2, author={Nageswara Rao and Susan Hicks and Stephen Poole and Paul Newman}, title={Testbed and Experiments for High-Performance Networking}, proceedings={Testbeds and Research Infrastructures. Development of Networks and Communities. 6th International ICST Conference, TridentCom 2010, Berlin, Germany, May 18-20, 2010, Revised Selected Papers}, proceedings_a={TRIDENTCOM}, year={2012}, month={10}, keywords={Network testbed dedicated channels SONET 10GigE WAN-PHY control-plane data-plane bandwidth scheduler WAN accelerators HAIPE InfiniBand}, doi={10.1007/978-3-642-17851-1_2} }- Nageswara Rao
Susan Hicks
Stephen Poole
Paul Newman
Year: 2012
Testbed and Experiments for High-Performance Networking
TRIDENTCOM
Springer
DOI: 10.1007/978-3-642-17851-1_2
Abstract
UltraScience Net is a network research testbed for supporting the development and testing of wide-area networking technologies for high-performance computing and storage systems. It provides dynamic, dedicated, high-bandwidth channels to support large data transfers, and also provides stable high-precision channels to support fine command and control operations. Its data-plane consists of 8,600 miles of cross-country dual OC192 backbone, which can be dynamically provisioned at different bandwidths. Its out-of-band control-plane is implemented using hardware Virtual Private Network (VPN) devices. In terms of the testbed infrastructure, it demonstrated the following capabilities: (i) ability to build and operate national-scale switched network testbeds, (ii) provisioning of suites of 1 and 10 Gbps connections of various lengths up to 70,000 and 8,600 miles, respectively, through automated scripts, (iii) secure control-plane for signaling and management operations, and (iv) bandwidth scheduler for in-advance connection reservation and provisioning. A number of structured and systematic experiments were conducted on this facility for the following tasks: (i) performance analysis and peering of layer 1-3 connections and their hybrid concatenations, (ii) scalability analysis of 8Gbps InfiniBand (IB) transport over wide-area connections of thousands of miles, (iii) diagnosis of TCP performance problems in using dedicated connections to supercomputers, (iv) detailed TCP performance analysis of wide-area application acceleration devices, and (v) TCP throughput improvements due to 10Gbps High Assurance Internet Protocol Encryptor (HAIPE) devices.