Fix seqno jumps when bwe is close to the lowest SVC layer

[vpalmisano]

It seems solving the issue described in #1069

patch.pcapng.gz

[jmillan]

I need to check it in detail but at a first glance I don't see how the issue is produced or how the fix solves it. Do you have an explanation for it?

[jmillan]

It makes sense to sync when no layer can be sent 👍

[vpalmisano]

Do you have an explanation for it?

Probably the sequence numbers keep increasing even if no layers are sent.

[ggarber]

@jmillan Isn't syncRequired marking when we are waiting for a keyframe (or something similar)? If layers are disabled is not clear for me why it should be syncRequired.

I mean, it looks clear that there is an (important) issue but I don't understand if changing syncRequired is fixing the issue or workarounding it.

Thx

[jmillan]

Yes, let's get to the root cause here.

[ibc]

We are drinking in an offsite. Sorry for the delay.

[buptlsp]

I also found a strange seqno issue. sqlno will jump per 20 packets in some scenarios. Currently I don't know what scenario. Can you please tell me if this pr of yours can solve this problem?

[buptlsp]

@ibc can you reply me about this problem? this means in some situation, the packet will lost 5% percent.

[ibc]

@buptlsp, this is a PR with a possible fix for SvcConsumer (AKA VP9 codec with SVC). Please let's focus on the topic of the PR.

[buptlsp]

the result is similar, the seq number will jumps. I don't know if it's a same issue. if this is only about svc, I will try to find the reason. thanks!

[ggarber]

Should the "this->syncRequired = true;" be done when the layers are reenabled (layer: -1 -> 0) instead of when the layers are disabled (layer: 0 -> -1)?

[ibc]

We'll jump into this after some very busy days at work.