Re: wg last call soft preemption question
Curtis Villamizar <curtis <at> faster-light.net>
2004-07-01 17:53:46 GMT
In message <1088547536.899736748cd86 <at> webmail.pi.se>
Loa Andersson writes:
> in the motiviation (section 2) for the
> draft-ietf-mpls-soft-preemption-02.txt it
> is stated that through the preemption as defined in rfc3209 traffic
> on the "to be preempted LSP" is (unnecessarily) abruptly preempted.
> The transit traffic is disregarded.
> I guess this is true, but isn't the (the most common) case here that
> the preemting LSP has a higher priority than the preemepted, and if
> you withhold the preemption, waiting for the traffic on the preempted
> LSP to be taken care of, you will waste traffic with higher priority
> during that wait?
Not at all if you are also using diffserv. The lower priority traffic
does not see an outage but will see some congestion. For most types
of Internet services this is more desirable than an outage.
Also, in another use of soft preemption, the preempted LSP may be a
backup for other traffic which may or may not be in use. If a subset
of backups are in use, you may see temporary congesstion, but all of
the traffic is in reality of equal importance in this usage. If you
hard preempt these you don't have congestion, but you have an outage
that exceeds the time threshhold that you were trying to acheive with
the backup scheme (FRR or standby from the ingress).