I have a similar need as the OP's.

We have a 3845 w/ a 20M DS3 Internet that is shared among multiple sites through dark fiber on the backend.

I was hoping to divi up the 20M Internet circuit w/o having to purchase something like a Packeteer device.

How can I trick the 10/100/1000M LAN interface into thinking it only has 20Mbps of bandwidth?

I believe the bandwidth interface command is for IGP's to calculate the cost/metric.

Would that work for the purpose of QoS?

I can configure QoS policy, allocate bandwidth for each remote site, and tie it to the LAN interface.

However, QoS doesn't kick in until there's congestion on the interface.

Since this is an Internet edge router, it means the LAN interface will never see congestion.

Any ideas are appreciated.

Unclear, to me, exactly what you want to accomplish.

On a 3845, you could shape bandwidth for each site or you could shape for the total capacity (20 Mbps) and allocate minimum guarantee for each site.

e.g.

policy-map perSite

class site1

shape average 1000000

class site2

shape average 1000000

== or ==

policy-map shape20mbps

class class-default

shape average 20000000

service-policy eachSite

policy-map eachSite

class site1

bandwidth remaining percent 5

class site2

bandwidth remaining percent 5

The bandwidth statement is primarily used for IGP's to calculate the cost/metric, however at the same time, it also influences the QoS on normal routers.

If you use for instance the percent bandwidth in the policy-map, you will see the allocated bandwidth change with the bandwidth statement when displaying the policy-map interface.

Thanks for the responses to my question. It pretty much confirmed what I have read and tested.

Apologize for the lack of details.

We have a dark fiber ring that inter-connects eight locations in a metro area.

The 3845 w/ the 20M DS3 is located at one of the eight sites.

Right now there's no QoS set up on the 3845, so any one of the eight sites can potentially hog the entire 20M bandwidth.

I want to implement something that would guarantee 2.5M (20/8) for each of the eight locations.

The LAN interface on the 3845 faces down towards the eight locations, so that's where I was to implement a QoS policy.

The problem I'm facing is that in order to trigger QoS, congestion has to occur on the interface.

Being that it's a 100/1000M LAN interface, there'll never be congestion, and therefore, QoS would never kick in.

My second example, in my prior post, would address your need.

The hierarchical QoS you have above?

policy-map shape20mbps

class class-default

shape average 20000000

service-policy eachSite

policy-map eachSite

class site1

bandwidth remaining percent 5

class site2

bandwidth remaining percent 5

How can you shape 20M outbound on a 100M LAN interface?

To trigger any QoS policy, there needs to be congestion on the interface first.

Traffic would never exceeds 20M out of the 100M LAN interface because we're being "policed" by ISP on the DS3 interface.

Kevin,

Shaping method does its job without congestion on the interface. It's not Congestion management. Traffic will be passed to the shaping policy. In case of exceed,traffic will be sent to the queue. You now can apply any method in the queue. CBWFQ is what you posted above.

HTH,

Toshi

Thank you all for your help!

I didn't realize a shaping policy doesn't require congestion to occur first...thought if there's no congestion, why bother w/ shaping or policing?

Edit-

After taking a deep breath and thinking about it, I definitely got confused w/ congestion management like Toshi pointed out.

Kevin,

You will realize it soon.

Here you go: http://www.cisco.com/en/US/tech/tk543/tk545/technologies_tech_note09186a00800a3a25.shtml

Toshi