HLD Port FEC BER

[vincentpcng]

What I did

Add design proposal for Port FEC BER feature

Summary of Changes:

Enhance the collect and compute the FEC BER on each port.
Modify the port_rates.lua , portstat.py and netstat.py

PR Title	State	Context
[sonic-utilities] Add port FEC BER show changes
[sonic-swss] Add port FEC BER feature swss
[sonic-mgmt] Add the port FEC BER mgmt test

[qinchuanares]

When the port is down or flapping, or even postFEC error > 0, preFEC BER calculation does not make sense.

After link flap or link down, when link is stably up, the preFEC BER calculation uses counters reset from 0 at link up.

[vincentpcng]

hit the wrong button and accidental closed the PR, reopen it.

[vincentpcng]

When the port is down or flapping, or even postFEC error > 0, preFEC BER calculation does not make sense.

After link flap or link down, when link is stably up, the preFEC BER calculation uses counters reset from 0 at link up.

Agree, the measurement could be compromised if the link is not stable, like flapping occur, etc

[Junchao-Mellanox]

The feature only capture the realtime port FEC BER. Is it possible to capture any history port FEC BER issue so that we will be able to get some debug information when an issue only happens in a short period?

[vincentpcng]

The following code review PR is for the implementation of this HLD
(1) sonic-net/sonic-swss#3363
(2) sonic-net/sonic-utilities#3607
(3) sonic-net/sonic-mgmt#15481

[vincentpcng]

Review feedback , the following key points were mentions

(1) can we add min , max, average BER
(2) what about future serdes , can we use user data rate when compute the BER.
(3) what about other devices support
(4) the statistical average 1E-8

response:
(1) we can add the min, max, and average to the show command... this can be done as a future enhancement
(2) unless IEEE defined the enoding, otherwise we dont know (i) what the future serdes speed, (ii) what type of encoding were use in it, hence can;t be done.
One suggestion is to use the user data rate for computing. This is a possible trade off. However, the resulted BER will become not a true reflection. The community future suggest can we add more columns and use the "user data rate".
This will be left as future enhancement discussion
(3) This feature is as stated, require the SAI support. If SAI layer can provide the counter, than we should be able to support the devices.
(4) when calculate the uncorrectable BER (post FEC), the frame were dropped, we have no idea how many error is in the frame. So we can take the best, worst or statistical average for the calculation. During the document review the 1e-8 statistically average were recommended for it.

[prgeor]

@vincentpcng can you please link all the code PRs to the HLD?

[vincentpcng]

I added the code PR(s) links before the meeting minute comments...

Let me include it again here

The following code review PR is for the implementation of this HLD
(1) sonic-net/sonic-swss#3363
(2) sonic-net/sonic-utilities#3607
(3) sonic-net/sonic-mgmt#15481