Developing Applications with Networking Capabilities via End-to-End SDN

Preventing congestion in tight spots

The DANCES project addressed the problem of congested metropolitan and end site campus networks. Although the DANCES partner sites were connected in the wide-area primarily via Internet2’s 100Gbps Advanced Layer2 Services network, the end site connections were typically limited to 10Gbps. With high performance computing and data storage systems, the movement of large data sets between sites resulted in competition for network bandwidth. Unmanaged use of the network resources led to sub-optimal utilization of not only the network, but also of storage and compute resources that were left waiting for data to be delivered. Large flows could almost entirely block smaller flows and small flows could disrupt large bulk data flows.

DANCES was implemented with both software and hardware components in a prototype environment. An SDN OpenFlow capable switch was installed at the collaborating sites to enable development and testing. Benchmark throughput and functionality testing were done throughout development and deployment to verify the operation of the hardware and software enhancements. The final testing demonstrated network bandwidth control and improved performance of the collaborators’ scientific applications.

The DANCES project started at the beginning of 2014 and completed at the end of 2016.

Our impact

DANCES produced a better understanding of the interactions and technical requirements for supporting end-to-end SDN across wide area and campus cyberinfrastructure.  The software developed and lessons learned were documented and made publicly available.

The broader impact of DANCES was SDN-enabled infrastructure integrated with supercomputing infrastructure applications. Such applications enabled the request and configuration of high bandwidth connections accessible to end users and improved network performance and predictability for supporting a wide range of science applications. As DANCES demonstrated, robust support of SDN in the LAN/MAN has the potential to provide better use and control of campus edge resources, which are subject to competition among the various services offered via research and education networks. DANCES also raised the level of understanding of SDN within the XSEDE sites, campus communities, and research groups.

DANCES Partners

PSC logo

National Institute for Computational Sciences logo

Penn State logo

National Center for Supercomputing Applications logo

Texas Advanced Computing Center logo

GeorgiaTech logo

XSEDE logo

Internet2 logo