[j-nsp] Shaper burst size

[Saku Ytti]

On (2015-05-11 15:54 +0000), Serge Vautour wrote:

Hey,

> The setup above works fine when DeviceB is connected at 1G on both interfaces. Likewise, it works fine when Start-DeviceA is also 100M. With the exact setup above, when DeviceB does 1G to 100M conversion, it can't seem to handle the bursts of data handed out by the MX and drops traffic. Configuring the MX burst size to 0 (actual lowest value possible is 1 Byte) seems to solve the problem: no more drops by DeviceB.
> Is setting the shaper burst size to 0 a good idea? Are there any consequences to doing this?

Not only it is a good idea, it is only way to do QoS in other points than
congestion points, because you cannot know buffer sizes downstream towards
congestion.
When you configure burst-size 1, you can observe 128 in PFE, but in QX actual
burst-size is about 2.7ms times shaper-rate. As you said it solves your
issues, you should be happy, your downstream equipment can handle this burst
without dropping.

If at all possible, always perform shaping/policing at last possible moment.
-- 
  ++ytti