On 2018-12-06 23:32, Stewart Bryant wrote:
>
>
> On 06/12/2018 00:34, Brian E Carpenter wrote:
>>
>> .
>>> Of course it still needs to step through them all to do ECMP even if
>>> they are all disabled.
>> No it doesn't. That's what the flow label, in a fixed position early in the IPv6 header, is for. A line speed IPv6 router has no need to look at the layer 4 header, even if it's doing both diffserv and ECMP. Looking at transport headers is an IPv4 concept.
>>
>>
>
> A question for the operators, how widely supported in the flow label?
But there's a preliminary question: how widely is the flow label set
by sending hosts? The answer is: widely, by modern o/s releases. But not
much, by legacy o/s releases.
> (a) in native IP
> (b) IP over MPLS
>
> Certainly in MPLS and Detnet land I hear about the five or six tuple all
> the time, including the inspection of them during forwarding, but I
> don't hear much spoken about the flow label.
Why would forwarding devices bother, until a majority of traffic has
the flow label set by the source? This is a long term play. As it
becomes harder and harder to parse packet contents, the flow label
will become more and more useful for ECMP.
Brian
