What would you do with fat fiber?

[gdt]

Doc Searls annoys me, I don't know why and it's probably be unfair. His blog was quoted on Linux Weekly News with a question I know a little about: What would you do with fat fiber?

Once you have 'enough' bandwidth, say 1Gbps, another set of factors comes into play to restrain network performance.

Simple latency is one. Unneeded round-trip times in applications are disliked by network engineers just as applications which make unneeded system calls are disliked by the kernel folk. TCP itself performs poorly once the bandwidth-delay product gets near cross-country gigabit.

So one thing that people can do with cheap large bandwidth is to help each other reduce latency: content distribution networks, P2P, and other overlay networks.

Having such cheap large bandwidth also makes sharing the results of computation more economic. Say you're down-rating a HDTV broadcast to SDTV. You can share the results of that intensive computational load to other viewers of the broadcast that might desire a SDTV feed.

Of course, it's a short step from there to full distributed computation, as with the Grid. Although that brings significant trust issues in some applications -- your neighbour may give malicious results.

Cheap large bandwidth also moves the line between what is "inside" the computer and what is "outside". Is large mass storage needed "inside" the computer anymore? At the least there's a lot to be said for using a RAID mirror with the one disk located across the network.

At 10Gbps the network finally rivals sending a station wagon full of tapes for throughput. So there's no reason not to do all archiving across the network.

Large bandwidths also allow distributed sensor networks. The Square Kilometre Array is a massive and early example, but more mundane applications are obvious. If applied to surveillance cameras and microphones these networks would provide a serious intrusion into privacy.

Large bandwidths also allow existing applications to be larger. As the Access Grid and video instant messaging shows, teleconferencing can grow from one-on-one to many-to-many. Quality can also increase, this can be marginal or dramatic. For example, the improvement in experience moving teleconferencing from SDTV to HDTV is remarkable -- someone can hold up a printed document and you can read the text they are pointing to.

Finally, large bandwidths allow us to reassess the economics of connection-oriented protocols. The connectionless packet protocols of the Internet are very vulnerable to denial of service attacks. For applications where these attacks are undesirable we may be able to use GMPLS controlling optical switches to give dedicated bandwidth on demand between end-points.