about IPv6 PPPoE

Bo Byrd bo@bbyrd.net
Wed, 1 May 2002 01:55:41 -0400


OK I see what you're saying

I have in operation a system with 10,000 active ADSL subs(1.5mbps each).
This system has a GIG-E interface.
10000 users X 1.5mbps = 15Gbps.  So this is the same as 15:1
Consider this.....my traffic on this GIGE interface has NEVER gone above
200mbps.  Never...not once. So say I had put 2 load balanced FastE
interfaces on my system instead of a single GIGE interface.. Then that
would equate to a 75:1 ratio, and my users wouldn't be able to tell the
difference.


-Bo




-----Original Message-----
From: owner-6bone@ISI.EDU [mailto:owner-6bone@ISI.EDU] On Behalf Of Joel
Baker
Sent: Tuesday, April 30, 2002 6:51 PM
To: 6bone@ISI.EDU
Subject: Re: about IPv6 PPPoE


On Tue, Apr 30, 2002 at 08:24:16AM -0400, Bo Byrd wrote:
> 20:1 seems very extreme.  From what I've seen a subscriber management 
> system can terminate around 32000 PVC's and can support 8000 active 
> PVC's, that's 4:1.  Of course for the best interests of the customers 
> you cant normally run it like that, for best performance you usually 
> see around 3:1 in datacenters with multiple SMS devices.  Surely 20:1 
> is in reference to some other set of figures.

Bandwidth subscription rates typically go at *least* 20:1 for
residential DSL; sometimes 30:1. This is not, actually, a problem for
most situations, if combined with proper monitoring of the circuit and
engineers who grok how statistical multiplexing works and when you
shouldn't be using it.

(FWIW: observed traffic showed about a 10:1 ration of capacity:traffic
on a device terminating business T1 customers, at least 2-3 of which
were running their lines full-bore 24x7; the business DSL folks ended up
being something between 15:1 and 20:1, and all of those customers pay
lots of money to never have a bottleneck inside the providers network;
that's *not* what residential customers pay for, and so they aren't
guaranteed it, and sometimes end up 30:1, or even 40:1 in one rumored
instance).

The statistics in question were taken over the span of a month and a
half, at 5 minute intervals, and processed extensively (some of it
useful, much of it for making pretty graphs to prove to the people
paying for upstream circuits that they couldn't safely try to multiplex
30:1 on the business T1s, even having hundreds of customers).
-- 
************************************************************************
***
Joel Baker                           System Administrator -
lightbearer.com
lucifer@lightbearer.com
http://users.lightbearer.com/lucifer/