Peformance Tuning when using GPU encode decode

Wowza Streaming Engine
, ,
1

Hi,

I have few question when I using streaming engine !

I am using AWS g4Xlarge (Now Available – EC2 Instances (G4) with NVIDIA T4 Tensor Core GPUs | AWS News Blog) for my streaming engine server .
and reference (Add graphic overlays to a live stream) this and only add simple scrolling text to my live stream.

here is my environment
Memory: 16GB
Processor Cores: 4
Wowza Streaming Engine Manager Version
4.8.14+9 (build 20210719152831)
OS:Linux (amd64)
Java Version:9.0.4
Java VM Version:9.0.4+11
Java Architecture:64

here is my transcoded config
Outgoing Stream Name:mp4:${SourceStreamName}_720p
Video Codec: H.264
Video Bitrate 1300000
Encoding Implementation: NVENC
Profile: Main
Key Frame Interval:Same as source

11 live stream at the same time need to transecode

this is source:

Network Throughput Bytes In is 1.5Mbits/s

this is transcoded stream:

Screen Shot 2021-09-06 at 12.10.12 PM

question:

  1. If I have 5 stream and source is stable ,when transcoded 3 is ok, 2 is bad network throughput bytes In
    I don’t know what I miss in configuration?.
  2. I can not use nvcuvid or cuda to decode , I got this 2 warning:
    2.1 Publisher.publish: Stream name already in use
    2.2 Video packet list is full
    after warning gpu will stop to encode and decode
2

So I decide to downgrade wowza streaming service version to 4.18.12

It’s work but seems have some performance problem

why gpu usage only 10%?
I received this logs: Video behind filter state change

Video behind filter state change
and these event happen often
destroy , unpublish

I have almost 20 x (720p 3.0Mbps) live stream at same time Which AWS instance type is recommended?