The Battle of the Bandwidth Isn't What Matters


Although 100 Gigabit Ethernet is passing interoperability tests left and right - the latest tests having been organized by Japanese internet exchange provider Internet Multifeed, which expects to be deploying the technology in Internet Exchange Point environments by the end of the year - it's far too early for the overwhelming majority of IT executives and managers to do more than think about bringing the technology into the data center.


At best, I don't expect to see 100GbE make it into any but the most advanced data centers until late next year, and realistically, I suspect most organizations will be pushing out their adoption of 100GbE until 2013 or 2014.

After all, many organizations are still digesting their upgrades to 10 Gigabit Ethernet, and 40GbE devices are just beginning to turn up in data centers - for example, on June 21, the same day that the Japanese ISPs announced the results of their 100GbE testing, Force10 Networks announced that IBM had certified its 40GbE line cards as interoperable with IBM's iDataPlex servers and clusters and its new Intelligent Cluster BOM components.

But I'm not sure how much raw throughput matters anymore, since the revolution in data-center networking isn't in the constant incrementing of network speeds. Instead, the ability to expand the concept of network fabric, from the individual switch with which most of us are comfortable, into one scalable multi-path network, is the game-changer. The question is how much of this revolution will be a matter of a single vendor's proprietary implementations -Cisco's FabricPath scheme being a perfect example of locked-in technology in standards-based clothing - and how much will become an inherent part of the Ethernet of the future?

It's been my experience that one's network environment has to permit a multi-vendor approach, even if the best results are often achieved by picking a single vendor and sticking with it. The flattened networks of the future based on TRILL - the Transparent Interconnection of Lots of Links - are going to have to do more than pay lip service to interoperability; nevertheless, I just don't see that being practical for the near future.