Re: draft-nordmark-multi6-sim-00.txt

[Brian E Carpenter <brc@zurich.ibm.com>]

Iljitsch van Beijnum wrote:
> 
> On 29 okt 2003, at 18:21, Brian E Carpenter wrote:
> 
> > Er, the diffserv field isn't available for that... we use it
> > for diffserv...
> 
> Using diffserv for signalling whether rewriting is allowed doesn't get
> too much in the way of regular diffserv use. (Hm, didn't we have a very
> similar discussion just recently?)

I think it does. There are moves afoot to define recommended service
classes and recommended DSCP mappings (for example 
draft-baker-diffserv-basic-classes-**.txt) and overloading these bits within
a mere 6 bit code space seems to me to be one bridge too far. I'd much rather
look at using two different flow labels, where we have a much larger code
space and no suggestion of pre-existing semantics.

> 
> Worst case, twice the number of code points would be needed: one for
> service level x without rewriting and one for service level x with
> rewriting, one for service level y without rewriting... Best case, only
> a single DSCP would be used up: default without rewriting, in addition
> to default with rewriting and all other service levels with rewriting.
> 
> > And I wish people would think about routers that will be shipped ten
> > or fifty
> > years from now. If the solution is sub-optimal for today's hardware,
> > that
> > should not in itself be a show-stopper.
> 
> Interesting point. I sort of agree. I don't think it applies here,
> though. Using the diffserv bits is the most logical and cleanest choice
> if we can make it work, and I believe we can.

My point is that even if people think it is slow to look at an extra header
today, this is probably irrelevant in the long term. 

   Brian