Traffic Policing Issue

chenhui · Created Jun 11, 2020 03:08:24

Posted by Harry101 at 2020-06-10 07:56 Hello @chenhui Thank you for your explanation. So to get the correct peak rate at the first second ...

Hi,
As the documentation describes below:

In Color-Blind mode, the following rules apply when a packet of size B arrives at time t:
If Tp(t) – B < 0, the packet is marked red, and The Tc and Tp values remain unchanged.
If Tp(t) – B ≥ 0 but Tc(t) – B < 0, the packet is marked yellow, and Tp is decremented by B.
If Tc(t) – B ≥ 0, the packet is marked green and both Tp and Tc are decremented by B.

When the bucket P is used up, then if configure the yellow marking traffic pass, more traffic will passing through than configuring the yellow marking traffic discard. The result is a higher peak rate in the motoring result.

jason_hu · Created Jun 4, 2020 10:25:58

please wait for.

chenhui · Created Jun 4, 2020 12:39:45

Hi @Harry101,

It's normal.

According to your configuration the CBS & PBS are 2887680 bytes, which equals to 2.8M, at the very begining, the result displayed on the NMS is effected by the monitoring interval.

chenhui · Created Jun 5, 2020 06:08:34

Hi Harry,
The PIR & CIR indicates the rate which adding tokens into the bucket C and P. PIR and CIR decide the rate when the bucket is empty. Assuming that the bucket is empty, since the total number of tokens in this unit time is PIR * unit time, the total amount of traffic passing through will less than PIR * unit time.
If the bucket is full, the total traffic passing through in a short time will be PBS, which is obviously much more than PIR * unit time, which results in a much higher rate than PIR

Sergio93 · Created Jun 5, 2020 06:39:17

@Harry101

Try to reset the interface statistics for that interface, the burst occurred maybe before the configuration took effect at 11:45. After that monitor again the graph and interface statistic.
Another thing you could do is to simplify the configuration as in this example and check whether you still have any strange behavior, if not, you can adjust the configuration and introduce cbs and cbs.
https://support.huawei.com/hedex/hdx.do?docid=EDOC1000177841&id=dc_cfg_qos_0134_2_2&lang=en

Hope it's useful.

Sapte · Created Jun 5, 2020 15:35:29

Hi @Harry101

I analyzed your problem in the below

ROOT CAUSE

Since you have configured policy traffic shaping based on ACL switch will make peak traffic 15mbps comes from this address : 10.92.92.0 0.0.0.255

Between your switches there are different traffics comes from different sources.In order to see that you can capture packets on the interfaces between switches.

SOLUTION

No need the change anything probably it works.In order to be sure you can anaylze your traffic on wireshark after captured it or mirrored it

Have a nice day !

chenhui · Created Jun 6, 2020 12:40:45

Hello Harry,
Which OIDs are you using to get the interface sending rate and receiving rate? I tried simulating this, but no proper OIDs are found to get the data.

Harry101 · Created Jun 8, 2020 08:05:52

Hi @Sapte
I tested again and mirror the outbound traffic between 2 switches. I didn't any different traffic. My topology only has 2 laptops connect to 2 switches.

Hi @chenhui
I used eSight tool of Huawei. Can you try with OID:
In: 1.3.6.1.2.1.31.1.1.1.6
Out: 1.3.6.1.2.1.31.1.1.1.10

The unit of CIR/PIR is bps while CBS/PBS is bytes, so I think CBS/PBS should be converted to "bps", and that is the maximum rate when the bucket is full. How do you think about it?

chenhui · Created Jun 10, 2020 05:48:42

Posted by Harry101 at 2020-06-08 08:05Hi @Sapte I tested again and mirror the outbound traffic between 2 switches. I didn't any different ...

Hi

I tried using the Zabbix, the monitoring result is fine.
The only uncertain result is the peak rate when the traffic starts.
For example, when I set the CIR and PIR to 2Mbps, CBS and PBS to 100Mbyte. Then, wait enough time to let the bucket C and bucket P being filled up. Send packets on the iperf3 client with rate 100Mbps which equals to 12.5MByte/s. Thereotially, I think, it will take about 100/12.5=8 seconds to use up the tokens in the bucket. And anothere 2Mbps*8s=16Mbit=2MByte is less than 12.5MByte. The result output on the Zabbix is nearly the same as the infer.
What the strange is that when the PBS is less than 12.5MByte, the result shows the peak rate which can reach is hardly to calculate out, and the peak rate is related to the monitoring interval. The less the interval is, the higher the peak rate can reach.

Back to your quesion. You think that the PBS shoudl be coverted to "bps", and that is the maximum rate when the bucket is full. Actually, the monitered peak rate is based on the monotring interval, as in my test. If, the monitoring interval could be set to 0, which means the NMS could monitor every packet passing through, in such situation, the output peak rate equals to the minimum of converted PBS and input rate. But it's impossible to set the monitoring interval to 0, and the rate displayed on the NMS is an average value of passing track during the monitoring interval.

Harry101 · Created Jun 10, 2020 07:56:21

Hello @chenhui

Thank you for your explanation. So to get the correct peak rate at the first second, I should reduce the CBS?

And do you have any idea about the result of 2 commands:

car cir 15360 pir 15360 cbs 2887680 pbs 2887680 mode color-blind green pass yellow pass red discard --> output peak rate = 50 Mbps

car cir 15360 pir 15360 cbs 2887680 pbs 2887680 mode color-blind green pass yellow discard red discard --> output peak rate = 30 Mbps

Traffic Policing Issue

Recommended answer

Third Party’s Trade Secret

My Followers

Enterprise

Consumer

Corporate

Carriers

Traffic Policing Issue

Recommended answer

Third Party’s Trade Secret

My Followers