Terabit-LAN project lead by NTT Network Innovation Laboratories attempts to provide Terabit capacity for future high-end applications, clusters, or PCs. Obviously, 10GbE/GbE interfaces in LAN and DWDM technology in WAN transport have well matured. Also, parallelism had penetrated in CPUs in PCs and lambdas in DWDM transmission technologies. So it seems to be an inevitable consequence that large-scale and high-resolution visualizing applications, clusters, or high-end PCs will incorporate multiple lambdas of 10GbE/GbE into their I/O interfaces in combination with DWDM or CWDM technology.

When we would provide 1Tbit/s capacity for edge equipment based on 10GbE channels, we have to handle 100 channels or lambdas at the same time in switching nodes, and also have to transport these channels with sufficient reliability. Accordingly, it might be an approach where we would use large optical MEMS switches for that purpose, and switch these lambdas individually. However, these lambdas will be expected to be used as a group of lambda in a limited number of optical paths in high-end applications. So these grouped lambdas should be switched and transported as a single path to simplify the switching complexity and to reduce cost.

Terabit-LAN/WAN project focus on this parallelism in switching and transport of multiple lambdas.

To realize parallel switching of these multiple lambda, Terabit-LAN/WAN adopts lambda group switching and establish lambda group path (LGP). High-end application users would require multiple lambdas as a whole for their clusters, so Terabit-LAN/WAN switch would provision and establish a LGP upon request of these applications. The number of lambdas in each LGP will be determined and provisioned by such applications. And Terabit-LAN/WAN switches these LGPs as a single end-to-end optical path.

From the viewpoint of parallel transport of 10GbE/GbE lambdas over Terabit-LAN/WAN, relative latency deviations in each LGP could be a serious problem to achieve high quality of service. This relative latency arises from several reasons, such as optical path diversity or group velocity dispersion in optical fibers. To meet such issues, we have proposed OVC (Optical Virtual Concatenation) function. OVC can be realized by OTN (Optical Transport Network) function defined in ITU-T Recommendation G.709. OVC can completely compensate the above-mentioned relative latency deviations to achieve a virtual terabit bulk transport. Therefore, visualization applications, clusters, and high-end PCs do not suffer any impairment that could result from the relative latency deviation in a single LGP.

OVC/Terabit-LAN demo
in iGrid2005 San Diego

We will demonstrate OVC/Terabit-LAN function in the upcoming iGrid2005 in San Diego. In the demo, we would establish a single LGP (Lambda Group Path) between Chicago and San Diego, which consists of two GbE channels. These two GbEs carries two one-halves of double-wide animation streaming on SAGE.

In normal operation, these two GbE channels in a LGP have an identical optical path, thus no relative latency deviation between the two. But, a fault in a fiber link would evoke a restoration procedure, and a protection path of quite different route could be established in a worst case. In such a case, relative latency deviation could cause noticeable degradation in the double-wide video streaming.

An option to overcome the issue would be to implement some synchronization function in the application. However, as we scale up the applications, such as a large number of tiles or higher resolution, cluster load to manage such synchronization could be a fundamental limit of scalability.

OVC/Terabit-LAN provides relative latency-free transport for these applications and expels away the limitation of scalability.