Video render performance - secondary question

[Soebirk, Thorsten]

Hi Deim,

I am not streaming from a browser but from a second gstreamer application over the internet. Generally I don't see any broken images, but if network bandwidth is severely constrained it does occasionally happen. In these rare cases webrtcbin and/or the underlying streaming plugins never seem to recover and the only solution is to reinitialize the connection. This hasn't been a significant issue in my experience, so I haven't looked into how to automatically detect and recover from these situations.

/Thorsten

From: gstreamer-devel [mailto:gstreamer-devel-bounces at lists.freedesktop.org] On Behalf Of deim31
Sent: 30 November 2020 17:18
To: gstreamer-devel at lists.freedesktop.org
Subject: Re: Video render performance - secondary question

EXTERNAL MAIL:


Hi Thorsten,

I don't have answers but let me ask something.

I'd like to know if you notice some broken image during streaming?
It seems you stream from browser to GStreamer? Right? So do I.
I tried to stream screen share or window share and image get broken after some time. Soon or later - always.

Thanks,
deim
Dne 25. 11. 20 v 9:43 Soebirk, Thorsten napsal(a):
I'm working on an application that renders video streamed via WebRTC, and I am a little puzzled by the amount of CPU and GPU that this uses. The pipeline is basically as follows:

webrtcbin ! queue ! glimagesink

While streaming an HD video source, my application (on Windows) is using appr. 10% CPU and 25% GPU. If I view the exact same WebRTC stream in the Chrome browser, it uses appr 4% CPU and 0% GPU.

I know the absolute number given here mean nothing, but I am struck by the fact that Chrome can render this using half the CPU and without using the GPU at all.

I am fairly certain that the resources are consumed in glimagesink and not the other elements of the pipeline, because if I replace glimagesink with fakesink, then my application's resource consumption essentially goes to zero. Also, if I insert a caps filter before the sink to reduce the resolution, this also reduces total resource usage, despite being an additional processing step.

I've also tried replacing glimagesink with d3dvideosink. Both use roughly the same resources.


Given that Chrome is likely highly optimized for this particular use case where Gstreamer is a much more general framework, I would not necessarily expect performance parity in this case, but does anyone have any suggestions for possibly reducing resource usage while maintaining video quality and resolution in this scenario? I'm also struggling to understand why glimagesink and d3dvideosink would be using the GPU at all (much less so heavily) when Chrome seems to do entirely without.

/ Thorsten



_______________________________________________

gstreamer-devel mailing list

gstreamer-devel at lists.freedesktop.org<mailto:gstreamer-devel at lists.freedesktop.org>

https://lists.freedesktop.org/mailman/listinfo/gstreamer-devel


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/gstreamer-devel/attachments/20201202/54f1bfab/attachment.htm>