[c-nsp] Full meshed corporate network, QoS - best practices

Sergey Voropaev serge.devorop at gmail.com
Mon Sep 15 10:39:44 EDT 2008


Hi members,


I would like to know best practices in the following network design.
We are some company with distributed office location all over the
world. We lease IP/MPLS service on our service provider, but in some
offices we buy 10Mb/s and in another 1Mb/s. The aim is to configure
our CE-routers such that all shaping and traffic prioritisation will
performing on its and no data was lost at the SP core. As I understand
SP use policer to inbound and outbound traffic.

It is not acceptable use common shaping to the outbound interface. For
example OFFICE_1 buy 10Mb/s and OFFICE_2 only 1Mb/s. If we configure
shaping on the outbound interface on OFFICE_1's router there will be
congestion on the OFFICE_2's PE-router and traffic will be dropped by
SP.

We can configure shaping for every single class. For example we can
create class "OFFICE_1<--->OFFICE_2" and shape it to 1Mb/s. And then
shape all this classes by parent shaper applied to outbound interface.
But this is also unacceptable because of if traffic to OFFICE_2 will
be forwarded from TWO sites like as OFFICE_1 there will congestion on
OFFICE_2's PE-router.


Is there QOS solution for such design?

Thanks in advance.


More information about the cisco-nsp mailing list