Why Ingress shaping?

[Tim.Yu]

As some LC on ASR9K provide capability of ingress shaping, I am wondering when this can be used and how this can be used. I agree in most cases ingress policing and egress shaping is enough. But I am wondering if scenario below needs ingress shaping:

For example, a service edge is connected with a server. On the port connecting with server, traffic coming inside is larger than outgoing. In this case, if ingress shaping will help on the burst? Is there any research on this? Also any other scenarios ingress shaping is use full and when I should use ingress shaping?

Thanks a lot.

[smilstea · Accepted solution]

If you are suspecting bursty traffic over the CIR then you can use policing or shaping or even WRED to manage the traffic. The only difference is that with shaping it allows you to burst a set amount beyond the CIR and not drop traffic like with a policer or WRED. If your pipe on egress is larger than the sum of all traffic ingressing onto it then shaping may be appropriate so that you can fully utilize the uplink, but if your ingress traffic is more than the size of your egress pipe then that will just cause drops on that interface. It really comes down to if you want to allow this server to burst or not and how much traffic you will have on the uplink. Also note that policing and WRED will cause TCP retransmissions on the server which may just end up starving the link for bandwidth and hurt end users in the end, so that is another reason to use shaping.

Sam