Well, the need to reduce silicon real estate is very noble indeed. The price and density of silicon goes down by a factor of 2 every 2 years so, and the flow-label legacy would probably last beyond that. So the pertinent question to answer would be how do we assign the bits in a way that they would still be useful 10 years from now.
Subrata -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Michael Thomas Sent: Wednesday, December 26, 2001 4:29 PM To: Perry E. Metzger Cc: Michael Thomas; [EMAIL PROTECTED] Subject: Re: Flow Label Perry E. Metzger writes: > Michael Thomas <[EMAIL PROTECTED]> writes: > > Perry E. Metzger writes: > > > I'm looking for statements from several router vendors that look much > > > like this: > > > > I don't speak for Cisco. I speak for myself. > > > > Sorry. > > Well, speaking for yourself, can you describe your simulations of > hardware performance with and without the flow label? I'm not especially interested in this game because some hardware geek somewhere is bound to throw enough transistors at the problem and claim that it doable. Big deal. It's an empty claim because it doesn't say what was given up in the process. I have witnessed firsthand hardware engineers on high end platforms react somewhere between disbelief and outright hostility at the prospect of chasing down header chains at line rate. CAM's -- as I've mention three times now -- are especially sensitive to bone headed standards potato blunders of this kind; I forget, but the transistor count is O(n^2) or maybe worse with the number of bits needed to form the key. IP6 is already at a big disadvantage given the need to key off of ~256 bits vs. 64 bits just for addresses. And lest you think that I'm just concerned about CAM based forwarding planes, think again; they all need to view enough of the header to classify, and that always impacts silicon as it gets bigger. Thus you get these choices: pay more than you should have, get worse performance, or delete other features you might have liked to go faster on a finite transistor budget. Oh, and you might ask your hardware vendors who claim to do this whether they can deal with flow classification of mobile IP with route optimization at line rate. And since MIPv6 route optimization is about as stable as jello, will they have the resilience to change their flow classier if it changes too? Mike -------------------------------------------------------------------- IETF IPng Working Group Mailing List IPng Home Page: http://playground.sun.com/ipng FTP archive: ftp://playground.sun.com/pub/ipng Direct all administrative requests to [EMAIL PROTECTED] -------------------------------------------------------------------- -------------------------------------------------------------------- IETF IPng Working Group Mailing List IPng Home Page: http://playground.sun.com/ipng FTP archive: ftp://playground.sun.com/pub/ipng Direct all administrative requests to [EMAIL PROTECTED] --------------------------------------------------------------------
