I agree. This can be cleaned up in the text. Brian
On 2011-02-04 14:24, Thomas Narten wrote: > Jarno Rajahalme <[email protected]> writes: > >> These are two different things: flow labeling (which labels to >> assign) and flow state establishment (how to treat specific >> flows). My own reading has been that flow labeling mechanisms may be >> stateful or stateless (for example, pseudo-random sequences were >> mentioned explicitly), and that signaling mechanisms may be needed >> to provide flow-specific treatment. > > This is an important observation. > > I think we can allow for stateless use of the Flow Label while > preserving the original definition for other types of usages. > > The original restrictions about not reusing Flow Label values had to > do with stateful setup protocols. There was an assumption of sorts > that there was agreement between the source of the traffic and the > routers on what a particular Flow Label value meant and how the router > was to assign resources. If an originating node reboots, and reuses a > Flow Label value too soon, the routers may not have timed out their > state and undesirable things might happen. > > We should preserve those semantics, not that I'm optimistic we will > ever need to, but because we can easily enough. > > For stateless usages of the flow label, it really doesn't matter if a > particular Flow Label value is reused too soon. > > It's in the stateful case that it matters. So just require that when > the Flow Label is set to a value as part of a stateful flow usage, in > *that* case, make sure we don't reuse Flow Label values too soon. That > is, for the more complex case (stateful involving setup) it's still > fine to required adherance to the original semantics. For the > stateless use, we can relax the requirements. > > The only place where this gets potentially messy is if you have an > originating node setting the Flow Label using stateful means, but is > also originating packets with zero Flow Label values, and then some > intermediate device changes those values from 0 to something else -- > potentially colliding with the stateful value. But this is a bit of an > edge case. We could require any node that uses stateful Flow Setup to > set all Flow Labels properly, or at a minimum not to send out Flow > Label values that are zero. > > Thomas > -------------------------------------------------------------------- IETF IPv6 working group mailing list [email protected] Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
