On Sep 8, 2010, at 11:44 AM, Christopher Morrow wrote: > On Tue, Sep 7, 2010 at 9:18 PM, Brian E Carpenter >> If this is correct, it is futile to assert that the flow label >> MUST be delivered unchanged to the destination, because we >> cannot rely on this in the real world. >>
Anything that cannot be proven to have the same original value at the end of the ride has this characteristic. >> Are we ready to accept this analysis? I, personally, am very ready to accept this analysis. It might be key to making the flow label useful. > what's the threat if it changes in flight? multiple times even? Speaking for myself, I see no threat. On the other hand, the question to me is what might be the benefit. What has been proposed is some variation on a load-sharing hash, one that might, for example, let the ingress node in a network direct specified traffic streams to one of several egresses - essentially what is done using MPLS but without the fuss and bother. One could imagine this being something akin to an ECMP route, but without any real discussion of "routing cost". Imagine that we could assign egress numbers to routers - akin to an IP address, but fits in 20 bits. An ingress node could look at some tuple (source/destination address, DSCP, whatever else) and decide that the correct egress router is <something>, and put the <something>'s number in the flow label to direct traffic towards the egress. One could imagine several policies - traffic toward a given egress is up to some limit rate, and arriving traffic that exceeds that rate goes to a different egress for example, or 1/3 of sessions this way and 2/3 that, or whatever served the ISP. -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
