Monday, February 25, 2008

The fallacy of bandwidth on demand

[Some excerpts from my opinion piece at Internet Evolution -- BSA]

Around the world, many National Research and Education Networks (NRENs) are focusing on various bandwidth-on-demand schemes for the future Internet architecture that will be used primarily for big science and cyber-infrastructure applications. The assumption is that in the future, big science institutions will produce such volumes of data that this traffic alone will easily exceed the capacity of today’s optical networks. If you think you have heard this story before, you’re right.

These same arguments were used to justify the need for ISDN (Integrated Services Digital Network), ATM (Asynchronous Transfer Mode), GMPLS (Generalized Multiprotocol Label Switching), and QoS (Quality of Service). These technologies were also premised on the assumption that network capacity would never be able to keep up with demand. Ergo, you needed an “intelligent” network to anticipate the applications demand for bandwidth.

Once again, we are hearing the same old litany that we need optical switched networks and optical bandwidth on demand as networks will be unable to keep up with anticipated traffic growth, especially for big science.

The fact is, no evidence exists yet that big science traffic volumes, or for that matter Internet traffic volumes, are growing anywhere near what was forecast, even just a few short years ago.

Combined with the slow ramp-up of big science traffic volumes, the business case for optical bandwidth on demand is also challenged by new optical network technology, which will dramatically increase existing optical networks by several orders of magnitude. Vendors like Alcatel-Lucent (NYSE: ALU), Nortel Networks Ltd. (NYSE/Toronto: NT), Ciena Corp. (Nasdaq: CIEN), and Infinera Corp. (Nasdaq: INFN) are developing next-generation optical networks using a new technique called coherent optical technology.

This will allow in field upgrades of existing optical networks to carry 100 Gigabits per wavelength using existing 10 Gigabit wavelength channels. The vendors expect that this can be accomplished without any changes to the core of network, such as optical repeaters, fiber, etc. Research into coherent optical technology is only at its infancy. Even more dramatic jumps in network will probably be announced in the coming years.

The argument for bandwidth on demand is further undermined with the advent of the new commercial “cloud” services offered by Google (Nasdaq: GOOG), Amazon, and others. Increasingly, big science and computation will move to these clouds for convenience, cost, and ease of use. Just like clouds allow the creation of shared work spaces and obviate the need for users to email document attachments to all and sundry, they will also minimize the need to ship big science and data files back and forth across NRENs. All computation and data will be accessible locally, regardless of where the researcher is located.