If my goal is to limit bandwidth for a traffic class, why should I police the traffic to drop excess traffic when I can instead shape it to that bandwidth limit and achieve the same results?

Policing will drop excess traffic, whereas shaping will delay the excess traffic temporarily for later transmission, but at least shaping sends the traffic later whereas policing will just drop it without regard. It seems like they can both accomplish the same function of limiting bandwidth, except that shaping is a lot gentler on the traffic. So why should I police when shaping can also do the trick and is a lot gentler on the traffic? This is excluding the fact that policing can be applied on the ingress/egress and can remark packets, while shaping can only be applied on the egress.

I was thinking that shaping will use up the buffer, but does that really matter?

Thanks

Because you can only do shapping going out

Shaping effects the outgoing buffer... which you don't always have access to

Policing you can do coming in

_________________
Freedom to all the people. Brave, true and strong.
Freedom to all the people. Unless I think you're wrong

dhimes.com

sorry didn't see your "other than the incoming/outgoing" line
that's really the best reason to do it... there might be some strange reason but I can't think of any off hand

_________________
Freedom to all the people. Brave, true and strong.
Freedom to all the people. Unless I think you're wrong

dhimes.com

If you're a service provider and you want to enforce a CIR you police. It's up to your customer to shape if they don't want traffic dropped (why waste your resources on traffic that you're not being paid to pass on?). Also policing allows you to remark traffic instead of dropping it. Again, more useful for a SP than an enterprise, but still an option.

_________________
blog.brokennetwork.ca

Thanks. That's what i was thinking too. You may not want to fill up the output buffer with traffic that's exceeded its CIR because you're then using up buffer space that the other class-maps could be using.

There's also the impact on TCP to think of.

If you police, TCP gets the message: slow down.

If you shape, TCP thinks it's operating over a very long link and will continue to run faster* until you ultimately drop the data anyway. In this case, mission not accomplished, in terms of throttling the source without dropping data. Making matters worse, you've wrecked TCP's responsiveness to lost data because loss detection is delayed when the required data is languishing in buffers somewhere.

Search "bufferbloat" for lots of discussion on this topic. It's not a perfect analog, because bufferbloat usually refers to all traffic, where you'll only be impacting certain traffic classes with the shaping. Within the class, the effect is the same.

Think about this: RED (dropping data randomly even when buffer space is still available) is a feature.

If we're not talking about TCP, but rather data which must get delivered, then nevermind. Shape away.

* up to the limit of the lesser of the sender's transmit buffer and the receiver's advertised receive window. Keep in mind however, that we're comparing server RAM against network device interface buffer. The network will always lose this battle

Shaping adds delay. Policing adds no delay.

_________________
www.mellowd.co.uk/ccie/

Not exactly. High-end routers can also shape inbound



As previously mentioned, us as an SP will police inbound to ensure Mr. Customer only gets what he pays for. It's up to Mr Customer to shape to that rate when sending traffic to us, but we will police anything going over that. Traffic goes over the trust boundry and hence we need to enforce

_________________
www.mellowd.co.uk/ccie/

Nifty
Didn't know that

Do you know off hand if a 6509 can?

_________________
Freedom to all the people. Brave, true and strong.
Freedom to all the people. Unless I think you're wrong

dhimes.com

Yes, with some of the ES cards, like ES+ and ES+T

_________________
http://networking.ventrefamily.com

o rly?

_________________
"I will prepare and some day my chance will come." - Abraham Lincoln
http://danielhertzberg.wordpress.com - I blog about networks!

The 7600 with an ES+ card, and/or FlexWAN cards, does ingress shaping. I'm sure there are others, you can probably find them with the feature navigator.

_________________
http://networking.ventrefamily.com

When I first learnt about shaping & policing, the ideal "Text Book" scenario is the a CE - PE link. The CE shapes outgoing traffic & the PE polices & drops inbound traffic.

Often times however, the PE & CE are both owned by the same ISP. In this case we do add a policer to the outbound CE interface but to Re-Mark, not drop traffic, Giving each critical class that recognises it's respective marking's the bandwidth it needs to function but then applying WRED to control traffic that exceeds the quota, ie; if the policer re-marks it.

To answer the original question, Shaping add delay, policing doesn't.

Shaping queues traffic, policing marks down / drops traffic.

Shaping can/does drop traffic too.

_________________
http://networking.ventrefamily.com

Indeed. Buffers have a fixed limit

_________________
www.mellowd.co.uk/ccie/

I wasn't implying shaping cannot result in dropped traffic, more just differentiating between the main purpose of both