[c-nsp] Customer Facing Interfaces: Policing vs. Shaping

David Luyer david at luyer.net
Tue Oct 19 07:37:29 EDT 2004


One point on this (haven't officially reported the below[1]):

GTS doesn't work properly with VPDN so you have to use CAR if you
want to shape L2TP PPP sessions.

The problem is present in at least 12.2(15)T8 and 12.3(6a).

Basically, if a customer on a L2TP Virtual-Interface is subject to GTS
and exceeds the permitted bandwidth, then they receive some dodgy PPP
packets with 2 bytes missing, which can have various effects from
reducing their speed dramatically to crashing some models of modems
(eg. D-Link 302g modems crash as a result of these packets).

Here's an example of a good packet:

15:16:07.255183 PPPoE  [ses 0x1428]
1100 1428 05ca 0021 4500 05c8 a119 4000
[...]

And from the same PPP session, a dodgy packet:

15:16:07.368515 PPPoE  [ses 0x1428]
1100 1428 05c8 4500 05c8 a11a 4000 3b06
[...]

Note the missing 0021 (PPP type IP) in the questionable packet, causing
PPP to think it is a dodgy PPP type.  If a customer deliberately tunes
their TCP window size and MRU way down to reduce their throughput below
the GTS bandwidth (eg, reduce to about 58k when shaped to 64k), then
they do not experience these packets when only accessing a single site.

David.

[1] we have a reasonable number of bugs which aren't officially reported,
    and tend to just hope that other ISPs in a similar environment will
    hit the same bugs and have enough spare time report and track them.
    however it seems a lot of people work on a similar assumption, 
    unfortunately.

> -----Original Message-----
> From: cisco-nsp-bounces at puck.nether.net [mailto:cisco-nsp-
> bounces at puck.nether.net] On Behalf Of Marko Milivojevic
> Sent: Tuesday, October 19, 2004 8:00 PM
> To: cisco-nsp at puck.nether.net
> Subject: [c-nsp] Customer Facing Interfaces: Policing vs. Shaping
> 
> 
>     Hello everyone, I was having brief discussion/argument with one of the
> members of this list this morning (hello Boyan :-)) regarding general QoS
> and queing. By the end of the argument we reached disagreement about
> service
> provider's customer facing interfaces. Should they be policed, shaped,
> "left
> as they are"? What are most of you guys doing?
> 
> 
> Marko.
> 
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/





More information about the cisco-nsp mailing list