On 8/22/07, David Newman <dnewman@networktest.com> wrote:
quoted text > -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> >
> >> Also, I noticed today that Google marks all their stuff with a DSCP of
> >> 0x38 (high throughput, low delay). Nice trick, but also an excellent
> >> argument for re-marking capability in all routers.
> >
> > nice trick? rather useless. I'd be extremely surprised if it makes any
> > difference at all.
> > i mean, who is really
> > 1) looking at DSCP/TOS at all,
> > - and -
> > 2) using them for different forward9ng priorities
> > - and -
> > 3) has congestion/fwd capa shortage so that it actually makes a
> > difference,
> > - and -
>
> For various reasons I can't name names, but I can tell you that there
> are some VERY large service provider and enterprise networks using DSCP
> classification and prioritization.
>
> ISPs tend to run at much higher utilization levels than enterprises and
> congestion is a reality on at least some of their pipes. So is the
> layer-8 urge to charge a premium to one set of customers over another.
> And even in the absence of congestion, there's still a desire to service
> delay- and jitter-sensitive voice and video ahead of other traffic.
>
> > 4) trusts externally set TOS/DSCP
>
> No one should trust external TOS or DSCP markings. Again, what Google is
> doing is an excellent argument for re-marking capability in all routers.
Yeah, really. Maybe we are misunderstanding, but wouldn't remarking
capability be exactly the ability to say "I don't trust these
externally set TOS/DSCP bits"? Henning, could you explain to the
luddites?
quoted text > And here we come full circle. Given the OpenBSD now IS a router --
> whether it's a little two-interface pf box for home use or some big
> studly hardware running OpenBGPD and OpenOSPFD box for ISPs, I would say
> the addition of support for DSCP re-marking would be a very desirable
> feature.
