In the earth and biomedical sciences, individual data objects such as a 3D terrain dataset or brain image are very large, compared to what can be interactively manipulated or visualized over today's networks. Yet scientists want to interactively explore massive amounts of previously uncorrelated data - for instance, to make earthquake predictions, or to understand the structure of the human brain - to advance scientific understanding and ultimately enable practitioners to make timely and important crisis-management or health-care decisions.
Some of today's academic research and education production networks achieve rates of 10 Gigabits per second, but applications using that much bandwidth over the Internet would be shut down as Denial-of-Service (DoS) attacks. The speed of the network alone doesn't determine how fast data moves. Getting data in and out of computers, and electronically managing data flow along different fiber paths from source to destination, all affect speed. Today's personal computers have 1 Gigabit per second (Gbps) Network Interface Cards (NICs), which current high-end electronic switches and routers can handle. However, PC vendors now have prototype 10Gbps NICs and the promise of much faster CPU processors. Together with new network protocols PCs and collaborative computing applications can blast out more data than before. To handle high-performance scientific applications, experimental networks are being built that are different from production data networks, taking advantage of two major technology trends:
- Advancements in Grid middleware (a grid is a set of networked computing resources); and
- Availability of tens-of-gigabits of networking bandwidth, enabled by the ability to encode data on individual wavelengths of light (or "lambdas") on single optical fibers.
Each lambda is currently capable of transmitting 1-10 Gigabits per second (Gbps), and soon will achieve speeds of 40Gbps and greater. The resulting networks, dubbed "LambdaGrids," eliminate bandwidth as a barrier to computing, analyzing and exploring very large datasets (hundreds of Gigabytes today, Terabytes soon, and approaching Petabytes by the end of this decade).
A 10Gbps transoceanic experimental network is currently operational between the StarLight facility in Chicago, and the NetherLight facility in Amsterdam. A major obstacle to scaling up to handle scores of 10Gbps flows has been the cost of upgrading routers and electronic switches. Line cards to handle 10Gbps are currently very expensive. Scientists at UIC and UvA believe that all-optical switches can be a relatively cost-effective solution for moving massive amounts of data to keep pace with order-of-magnitude increase in bandwidth requirements. The advantage of all-optical switches lies in their ability to handle this increased data throughput without an increase in cost or complexity. Calient's DiamondWave switch can handle 10 Gbit/s per wavelength today, and does not need an upgrade to switch the 40 Gbits signals of the future.
By deploying Calient Networks' DiamondWave switches at the Starlight and Netherlight facilities (see February 10, 2003 news release), researchers at both locations will actively engage in experiments as part of the OptIPuter project. The OptIPuter itself is a "virtual machine" that sits on top of the LambdaGrid. Depending on an application's requirements, the OptIPuter schedules and configures the computational resources needed for the period of time needed. The resources (whether a cluster, data store, large-scale instrument or visualization display) and the lambdas that connect them are focused on expedient solutions to hard problems. OptIPuter scientists at UIC and the University of California, San Diego (UCSD) have also launched a major effort to develop new network protocols required to support this new vision of optical networking.
Related Links
For more information on organizations involved in the OptIPuter project, visit the following websites:
# # #
Calient Networks: Tina Tan, ttan@marketready.com, 763-548-8208
Cal-(IT)²: Doug Ramsey, dramsey@ucsd.edu, 858-822-5825
UIC EVL: Laura Wolf, laura@evl.uic.edu
