[arin-ppml] Policy Proposal: A Modest Proposal for an Alternate IPv6 Allocation Process
George, Wes E [NTK]
Wesley.E.George at sprint.com
Fri Jun 5 16:23:47 EDT 2009
Ted Mittelstaedt wrote:
Right, but those challenges ALSO exist to do things like, for example,
run Windows 7 with acceptable speed. Heck they also exist to run Vista
at an acceptable speed.
Remember Moore's Law here.
There was a large debate in the IEEE HSSG (those responsible for the 100GE standard), and one of the notable parts was Sun Microsystems (and others), who fought hard for a 40GE standard because they believed that it would be too long before standard server technology would be capable of high enough sustained throughput to really take advantage of 100Gbps network interfaces. My guess is that they're fairly familiar with Moore's law over at Sun.
By comparison, there are multiple routers that are capable of throughput in the terabit range, today. Could you build a PC that was capable of that level of throughput? Yup, but it'd start looking (and costing) an awful lot like a router.
Running Windows quickly is not a valid comparison. Yes, at some point core routers may have to offload management of the routing table to a big bladecenter full of processors and RAM, but that still won't fix the need for them to actually know what to do with the packets as they come through, which is all done in hardware at the speeds required. Might a technology advance make that a moot point? Yes, but until that happens, I have to assume it won't (or it won't happen fast enough) in order to ensure that I can continue operating my network at a reasonable cost.
Yes, some of the ideas in this proposal might get us there. I didn't say I was totally against it, just some parts of it.
This e-mail may contain Sprint Nextel Company proprietary information intended for the sole use of the recipient(s). Any use by others is prohibited. If you are not the intended recipient, please contact the sender and delete all copies of the message.
More information about the ARIN-PPML