ACADEMIA

PITTSBURGH, PA -- The U.S. Department of Energy has awarded $2.5 million to the Net100 Project, a multi-institutional collaboration to improve end-to-end performance on high-speed networks. The three-year grant from the Mathematical, Computational Sciences Division in the Office of Science at DOE, effective Sept. 7, supports the development of a suite of tools that enable computer operating systems to automatically tune network performance. Whether they are shipping huge amounts of data from a super-collider or remotely controlling a video conferencing camera from 2000 miles away, scientists using high-performance networks need to send data as fast as the network will allow. Today's computer operating systems come configured to transfer network data at only one speed -- usually slow -- regardless of the underlying network. For a 10 million bits per second (Mbps) corporate network or a 56K dial-up Internet connection, the default speed is fast enough. However, on a high-speed network that can potentially send 622 Mbps, such as the Department of Energy's Energy Sciences Network (ESnet), the default speed often hinders next-generation applications. The network throttle of computer operating systems is controlled through the Transmission Control Protocol (TCP). By making adjustments to TCP settings, networking experts can tweak and tune the operating system to move data faster. The goal of Net100 is to make the operating system tune itself and to do so dynamically with changes in conditions on the network. "The Department of Energy has a number of applications in computational science, distributed computing and real-time instrumentation remote control that demand optimal performance from the network," said Thomas Ndousse-Fetter, DOE program manager for networking. "Currently, scientists need networking experts to optimize individual systems and applications. Net100 will build that expertise into the operating system." "Net100 has components that will probe and sense the current state of the network, the capacity of the links or the possible paths between sites to get a real-time picture of the current capacity of the network," explains Tom Dunigan, senior research scientist at Oak Ridge National Laboratory. "Then it will feed that information to the operating system to tune TCP to the current network conditions. As the application runs, the operating systems on the end nodes will sense if there is congestion or if the link is being under-utilized, and actually re-tune the transfer as it runs." Net100 is a collaboration of the Pittsburgh Supercomputing Center (PSC), the National Center for Atmospheric Research (NCAR), Lawrence Berkeley National Laboratory (LBNL) and Oak Ridge National Laboratory (ORNL). The network sensing components are based on: NetLogger from LBNL; and Probe from ORNL and LBNL. The operating system auto-tuning facility is based on PSC and NCAR's Web100 Project, funded by the National Science Foundation. For more information visit www.net100.org/