Dear Support Team,
I am benchmarking Wowza Streaming Server on Ubuntu 20.04.02 TLS, Intel Xeon E-2288G 8/16 CPU, 32 GB RAM remote workstation. I was expecting to get twise as better results as described in Wowza_Transcoder_Benchmarks_Transrate1080p.pdf, but I have two times worse results. I am transcoding, transrating 1080p Bunny to 720p, 360p, 240p, 160p. Transrate results 1 stream - 20% CPU, 2 - 46% CPU, 3 - 60% CPU, 4 - 85% (unusable). Transcoding results 1 - 7%, 2 - 21%, 3 - 45%, 4 - 55%, 5 - 85% (unsuable).
How would you suggest to proceed with this issue, because I have up to 4 times worse results compared to CPU (8 core, 16 thread) that is in use.
Transcode, transrate templates are used default, just enabled presets. Wowza Streaming Engine 4.8.11+5 configuration by default is optimized as described in your documentation.
Best regards,
Rimas
Which server in that PDF are you comparing against (more specifically: which column on page 2)?
Hello Karel,
I benchmarked only software based transrate and transcode functionality. I was not able to succesfully configure Intel QuickSync for Ubuntu 20.04.02 LTS, but it will be next topic.
We have Server 1 and Server 3 on page 1 in the PDF without hardware acceleration, so it will be Server 1 Default and Server 3 Default columns on page 2. Server 1 Default reaches 68% CPU utilization at 5 inputs, Server 3 Default - 7 inputs at 68% as well. It really well correlates to PassMark results: Server 1 - Intel Xeon CPU E3-1285 V3: 6919 and Server 3 - Intel i7 6700k: 8968.
It is around 1300-1400 PassMark points for every input to transrate. These numbers I used to calculate my system transrate input count. My system has Intel Xeon E-2288G with PassMark score: 17378. I would expect at least 12 inputs under 70% CPU load. My result is only 3 inputs under 60% CPU load.
Karen, how would you suggest to proceed with Wowza Streaming Server as I need all 12 inputs under 70% CPU utilization 
. It can be more, but not less.
Regards,
Rimas
Riman, I’d add a GPU card to that server; a decent NVidia Tesla should be able to handle 15+ channels. In your CPU tests did you use the legacy Mainconcept en-/decoder or did you use the new Beamr? I haven’t done any benchmarking for Beamr and it may have a different impact on CPU because it’s a different coder brand.
Hello Karel,
lets not start straight to the Nvidia Tesla cards and your answer does not relate to what I asked. Wowza provided benchmarks for systems with 4 cores, 8 threads CPU’s that in software can handle up to 7 inputs with PassMark score up to 9000. We have 8 core, 16 thread CPU with PassMark score 17000 and can handle only 3 inputs. Isn’t that ironic?
How many inputs Wowza Streaming Engine can handle with 8/16 cores in your tests? Could you provide what CPUs you used for benchmarks? Could you provide configuration XML files, that you used to get more than 3 inputs?
I will use these files and hope to provide positive feedback and topic will be closed.
Regards,
Rimas
I didn’t go straight to the Tesla cards; I just said that as a recommendation. The rest of my message went unanswered.
If you want benchmarks that are comparable to Wowza’s own publications, then you can do your own (see this article) or your can post a message in the Hire A Consultant forums and ask someone to do it for you.
Hello Karel,
thank you for the link to Capture Transcoder benchmark statistics in Wowza Streaming Engine configuration page. I already checked it. Default templates are used for transcoding and transrating , so Mainconcept coder is used by default. I hope we do want to compare apples to apples.
Karel, do you acknowledge that if we would use Intel Xeon CPU E3-1285, Wowza Streaming Engine 4.8.11+5 on Ubuntu 20.04.02 LTS we should process 5 inputs?
If we would use Intel i7 6700k Wowza Streaming Engine 4.8.11+5 on Ubuntu 20.04.02 LTS we should process 7 inputs as it is provided here. And if we would use much faster processor such as Intel® Xeon® E-2288G with Wowza Streaming Engine 4.8.11+5 on Ubuntu 20.04.02 LTS we should process more than 5 inputs?
Beamr coder from my experience reduces CPU utilization up to 10-15 % on full CPU load, but it was not mentioned in benchmarks that are provided.
Bay the way, my name is Rimas, not Riman.
Regards,
Rimas