Traditional IP Switching/Routing Combined with Optical Technology 1. Describe your technology application, including the users of the technology and how it is used. Scientists at the California Institute of Technology (Caltech), the European Organization for Nuclear Research (CERN), the Stanford Linear Accelerator Center (SLAC) and the Los Alamos National Laboratory (LANL) set up a high performance trans-Atlantic network testbed with a 10 Gbits/s link between Sunnyvale in N. California and Chicago Illinois, and utilizing the 2.5 Gbits/s DataTAG link between Chicago and Geneva Switzerland. At each site, high-performance commercial PCs running the standard Linux operating system were connected using commercial Syskonnect 1GE and Intel 10GE Network Interface Cards (NICs). Memory-to-memory data transfers have been performed using the standard TCP stack (New Reno) and a revolutionary TCP stack (FAST TCP) developed at Caltech. The team transmitted over a TByte of data in just under an hour from Sunnyvale near SLAC in California to CERN in Geneva with a single TCP stream between 2 PCs with 10GE NICs, using jumbo frames and the standard (New-Reno) TCP. This corresponds to a sustained TCP rate of 2.38 Gbps across 10,037 kilometers for more than one hour. Utilizing 10 hosts connected via 1GE NICs, the FAST TCP stack and standard 1500Byte MTUs, the team was also able to transfer 21 TBytes in 6 hours with 10 flows at an aggregate throughput of 8.6Gbps. Today our needs are mainly driven by High Energy and Nuclear Physicists (HENP) who are using multi-hundred Gbits/s trans-Atlantic links to transfer over TBytes/day between SLAC in California and European computer centers in Lyon, Padova and near Oxford. The data rates required are doubling annually and with the turn on of the next generation of HENP experiments at the Large Hadron Collider (LHC) at CERN, Geneva the trans-Atlantic requirements are expected to have increased to multi-10Gbits/s. Other data intensive sciences with similar growing needs include global climate prediction, astrophysics, bioinformatics, fusion, and seismology. In the future we can see these speeds being valuable for the telemedicine, aerospace, oil, and media distribution industries among others. 2. How does the technology promote the development and implementation of a gigabit state-wide network by 2010? An important part of our message is to illustrate that: it is possible today, with commercial off-the-shelf components, to achieve high bandwidth; that TCP and the Internet have not run out of steam yet and continue to scale; we do not have to await difficult to deploy modifications to backbone router congestion strategies; and the new techniques will work well on a shared gigabit state-wide network. At the same time we are leading the way in demonstrating and understanding how to achieve these high throughputs. Today this takes a combination of wizards with experience in networking, operating systems etc. One result of our work will be to identify the critical components and configuration settings, to provide recipes and simplify use for later users. Another result is to feedback to vendors how individual components work, how the components work together, where the bottlenecks are, and what needs attention. 3. State why you think the technology application deserves recognition. If the network continues to improve in performance by a factor of 2 / year, as it has for the last 10-20 years, then we can expect universities, research establishments or businesses with 155 Mbits/s or 622 Mbits/s connections today will have 100 times this performance by 2010. Thus the performance that today we are pushing the envelope to achieve will be common place and even mundane in many cases in at the end of the decade. Our work is critical for setting expectations for applications and users, for planning, and for understanding how to achieve high performance and making it easy to do so. In some cases the improved network performance will lead to new models for how research and business is performed. The network, once viewed as an obstacle for virtual collaborations and distributed computing in grids, can now start to be viewed as a catalyst instead. Grid nodes distributed around the world will simply become depots for dropping off information for computation or storage, and the network will become the fundamental fabric for tomorrow's computational grids and virtual supercomputers. For example: it may no longer be necessary to ship truck or plane loads of data around the world, rather than use the network; the need to colocate data and computing may change; the business model for media distribution could be dramatically changed; geographically distributed collaborations of people working with large amounts of data will become increasingly possible.