[e2e] Re: crippled Internet
fred at cisco.com
Wed Apr 25 16:57:05 PDT 2001
At 03:31 PM 4/25/2001 -0700, Randy Bush wrote:
> >> please explain how we might judge whether the these are sufficient for say
> >> voip.
> > without knowing the specifics of the test scenario
OK, you didn't say that, so I didn't know that.
> and note that we have distributions of one-way delay here, not
>some standard deviation of ping times. as to how they were measured, see
><http://www.ripe.net/ripencc/mem-services/ttm/index.html>. i can not tell
>you between what host or provider pairs the particular measurements were
>taken, as that would extend the consent of the measured beyond the present
>limits. and i don't think it really matters.
It may matter. I understand and respect the privacy issues, but I wonder if
the context of the present discussion whether this says anything about
access circuits. I just don't know whether it does or not, because I don't
know if it crosses any, and if it does what their characteristics might be.
And oops, I read the graph wrong in my previous note. I read it as showing
packets with 500 ms delay and packets with tens of ms in delay. That swaps
the axis - you saw hundreds of packets with 30 ms delay and tens with 50-60
ms delay. Sorry about that.
>considering that *extremely* few experienced delay far from the mode, what
>can we say about what is reasonable for voip? can we quantify it in some
>way we can actually measure to see if we're doing well? i.e. real numbers
>please, not that someone in the press or on henning's corridor whines.
I just don't know how to respond to that.
How about we do it the other way. You produce someone, anyone, anywhere,
who says that VoIP sound quality across the North American backbone is
good. We have good experience with it within highly over-provisioned LAN
networks, and we have reasonable experience with it in networks that use
the tools we have produced for it. But I don't hear people saying that's
true over the backbone. I'm struggling to explain that fact if there is no
More information about the end2end-interest