Gigabit Test beds

The Internet backbones operate at megabit speeds, so for people who want to push the technological envelope, the next step is gigabit networking. With each increase in network bandwidth, new applications become possible, and gigabit networks are no exception.

Gigabit networks provide better bandwidth than megabit networks, but not always much better delay. For example, sending a 1-KB packet from New York to San Francisco at 1 Mbps takes 1 msec to pump the bits out and 20 msec for the trans-continental delay, out of a total of 21 msec. A 1-Gbps network can reduce this to 20.001 msec. While the bits go out faster, the trans-continental delay remains the same. Thus for wide area applications in which low delay is critical, going to higher speeds may not help much. Fortunately, for some applications, bandwidth is what counts, and these are the applications for which gigabit networks will make a big difference.

One application is telemedicine. Many people think that a way to reduce medical costs is to re-introduce family doctors and family clinics on a large scale, so everyone has convenient access to first line medical care. When a serious medical problem occurs, the family doctor can order lab tests and medical imaging, such as X-rays, CAT scans, and MRI scans. The test results and images can then be sent electronically to a specialist who then makes the diagnosis.

Doctors are generally unwilling to make diagnoses from computer images unless the quality of the transmitted image is as good as the original image. This requirement means images will probably need 4K x 4K pixels, with 8 bits per pixel (black and white images) or 24 bits per pixel (color images). Since many tests require up to 100 images (e.g., different cross-sections of an organ being inspected), a single series for one patient can generate 40 gigabits. Moving images (e.g., a beating heart) generate even more data. Compression can help some but even the most efficient compression algorithms reduce image quality. Furthermore, all the images must be stored for years but may need to be retrieved at a moment's notice in the event of a medical emergency. Hospitals do not want to become computer centers, so off-site storage combined with high-bandwidth electronic retrieval is essential.

Another gigabit application is the virtual meeting. Each meeting room contains a spherical camera and one or more people. The bit streams from each of the cameras are combined electronically to give the illusion that everyone is in the same room. Each person sees this image using virtual reality goggles. In this way meetings can happen without travel, but again, the data rates required are stupendous.

Starting in 1989, ARPA and NSF jointly agreed to finance a number of university-industry gigabit test beds, later as part of the NREN project. In some of these, the data rate in each direction was 622 Mbps, so only by counting the data going in both directions do you get a gigabit. This kind of gigabit is sometimes called a "government gigabit".

The first five projects on gigabit testbeds are mentioned here. They have done their job and been shut down, but deserve some credit as pioneers, in the same way the ARPANET does.

1. Aurora was a testbed linking four sites in the Northeast: M.I.T., University of Pennsylvania, IBM's T.J. Watson Lab, and Bellcore (Morristown, N.J.) at 622 Mbps using fiber optics provided by MCI, Bell Atlantic, and NYNEX. Aurora was largely designed to help debug Bellcore's Sunshine switch and IBM's (proprietary) plaNET switch using parallel networks. Research issues included switching technology, gigabit protocols, routing, network control, distributed virtual memory, and collaboration using video conferencing.

2. Blanca was originally a research project called XUNET involving AT&T Bell Labs, Berkeley, and the University of Wisconsin. In 1990 it added some new sites (LBL, Cray Research, and the University of Illinois) and acquired NSF/ARPA funding. Some of it ran at 622 Mbps, but other parts ran at lower speeds. Blanca was the only nationwide testbed; the rest were regional. The research interest here was in protocols, especially network control protocols, host interfaces, and gigabit applications such as medical imaging, meteorological modeling, and radio astronomy.

3. CASA was aimed at doing research on super-computer applications, especially those in which part of the problem ran best on one kind of supercomputer (e.g., a Cray vector supercomputer) and part ran best on a different kind of supercomputer (e.g.. a parallel supercomputer). The applications investigated included geology (analyzing Landsat images), climate modeling, and understanding chemical reactions. It operated in California and New Mexico and connected Los Alamos, Cal Tech, JPL, and the San Diego Supercomputer Center.

4. Nectar differed from the three test beds given above in that it was an experimental gigabit MAN running from CMU to the Pittsburgh Supercomputer Center. The designers were interested in applications involving chemical process flow-sheeting and operations research, as well as the tools for debugging them.

5. VISTAnet was a small gigabit testbed operated in Research Triangle Park, North Carolina, and connecting the University of North Carolina, North Carolina State University, and MCNC. The interest here was in a prototype for a public switched gigabit network with switches having hundreds of gigabit lines, meaning that the switches had to be capable of processing terabits/sec. The scientific research focused on using 3D images to plan radiation therapy for cancer patients, with the oncologist being able to vary the beam parameters and instantaneously see the radiation dosages being delivered to the tumor and surrounding tissue.