How to increase video transcoding speed?

PeerTube
1

In my opinion, the transcoding speed is too low, for the provided resources.

My PeerTube instance (version: 5.2.1) has 8GB RAM, 8CPUs with barely any viewers for now.

I’m running VOD Transcoding jobs via Remote Runners (to increase performance).
I’ve provided 10 (concurrent) runners on a 64GB RAM, 32CPU machine.

There’s a gigabit connection between the 2 instances.

The transcoding is still too slow!

My settings:

image
image804×681 31.2 KB

Example, (sanitised) logs of a job from my instance:

-----------------------------------------------------------------------------------------
-----------------------------------------------------------------------------------------
> 93a91565-feb5-40ec-8d06-c592e7e977b9	vod-web-video-transcoding	Completed	100			12/28/23, 12:48:51.407 PM

Parent job: -
Processed on 12/28/23, 2:55:25.614 PM
Finished on 12/28/23, 3:08:37.359 PM
Payload:
{
  "input": {
    "videoFileUrl": "https://my-domain/api/v1/runners/jobs/93a91565-feb5-40ec-8d06-c592e7e977b9/files/videos/<video-uuid>/max-quality"
  },
  "output": {
    "fps": 30,
    "resolution": 1080
  }
}
Private payload:
{
  "videoUUID": "<video-uuid>",
  "isNewVideo": true
}
-----------------------------------------------------------------------------------------
> e44378c3-6773-4ef8-ab37-ac1953a9bba1	vod-hls-transcoding	Completed	100			12/28/23, 12:48:51.468 PM

Parent job: 93a91565-feb5-40ec-8d06-c592e7e977b9
Processed on 12/28/23, 3:14:00.069 PM
Finished on 12/28/23, 3:22:18.576 PM
Payload:
{
  "input": {
    "videoFileUrl": "https://my-domain/api/v1/runners/jobs/e44378c3-6773-4ef8-ab37-ac1953a9bba1/files/videos/<video-uuid>/max-quality"
  },
  "output": {
    "fps": 30,
    "resolution": 720
  }
}
Private payload:
{
  "videoUUID": "<video-uuid>",
  "isNewVideo": true,
  "deleteWebVideoFiles": false
}
-----------------------------------------------------------------------------------------
> c7e26893-8b5e-48db-8f9e-a0e514748a60	vod-web-video-transcoding	Completed	100			12/28/23, 12:48:51.458 PM

Parent job: 93a91565-feb5-40ec-8d06-c592e7e977b9
Processed on 12/28/23, 3:13:52.905 PM
Finished on 12/28/23, 3:22:03.179 PM
Payload:
{
  "input": {
    "videoFileUrl": "https://my-domain/api/v1/runners/jobs/c7e26893-8b5e-48db-8f9e-a0e514748a60/files/videos/<video-uuid>/max-quality"
  },
  "output": {
    "fps": 30,
    "resolution": 720
  }
}
Private payload:
{
  "videoUUID": "<video-uuid>",
  "isNewVideo": true
}
-----------------------------------------------------------------------------------------
> 75f84c27-aad1-4ce6-b0d6-0bcebfc00886	vod-hls-transcoding	Completed	100			12/28/23, 12:48:51.447 PM

Parent job: 93a91565-feb5-40ec-8d06-c592e7e977b9
Processed on 12/28/23, 3:13:31.589 PM
Finished on 12/28/23, 3:19:36.317 PM
Payload:
{
  "input": {
    "videoFileUrl": "https://my-domain/api/v1/runners/jobs/75f84c27-aad1-4ce6-b0d6-0bcebfc00886/files/videos/<video-uuid>/max-quality"
  },
  "output": {
    "fps": 30,
    "resolution": 480
  }
}
Private payload:
{
  "videoUUID": "<video-uuid>",
  "isNewVideo": true,
  "deleteWebVideoFiles": false
}
-----------------------------------------------------------------------------------------
> 0c6cd1d7-727e-43a9-b991-913b276491fc	vod-web-video-transcoding	Completed	100			12/28/23, 12:48:51.436 PM

Parent job: 93a91565-feb5-40ec-8d06-c592e7e977b9
Processed on 12/28/23, 3:12:03.211 PM
Finished on 12/28/23, 3:18:08.778 PM
Payload:
{
  "input": {
    "videoFileUrl": "https://my-domain/api/v1/runners/jobs/0c6cd1d7-727e-43a9-b991-913b276491fc/files/videos/<video-uuid>/max-quality"
  },
  "output": {
    "fps": 30,
    "resolution": 480
  }
}
Private payload:
{
  "videoUUID": "<video-uuid>",
  "isNewVideo": true
}
-----------------------------------------------------------------------------------------
> b986a9b4-6f6b-4362-8244-1ba1713ae591	vod-hls-transcoding	Completed	100			12/28/23, 12:48:51.423 PM

Parent job: 93a91565-feb5-40ec-8d06-c592e7e977b9
Processed on 12/28/23, 3:08:37.856 PM
Finished on 12/28/23, 3:20:25.380 PM
Payload:
{
  "input": {
    "videoFileUrl": "https://my-domain/api/v1/runners/jobs/b986a9b4-6f6b-4362-8244-1ba1713ae591/files/videos/<video-uuid>/max-quality"
  },
  "output": {
    "fps": 30,
    "resolution": 1080
  }
}
Private payload:
{
  "videoUUID": "<video-uuid>",
  "isNewVideo": true,
  "deleteWebVideoFiles": false
}
-----------------------------------------------------------------------------------------
-----------------------------------------------------------------------------------------

This job took more than 30 mins.!

Is this expected?

Is there any way that i can speed this up significantly?

1 « J'aime »
2

Any feedback on this, anyone?

3

Hi,

Check the runner config to increase ffmpeg threads

4

I’m already running 10 runners/threads in parallel!

Are there any benchmarks for performance (time taken) vs resources, which i can use as a reference, to compare with my scenario?

5

I’m already running 10 runners/threads in parallel!

I’m not sure to understand. You don’t need to run multiple runners in parallel. Juste update ~config/peertube-runner-nodejs/common/config.toml of 1 runner and update concurrency to 10. If you want faster transcoding you can also set a different value for threads under [ffmpeg] section.

6

I seem to have expressed this improperly.

My current config is-

[jobs]
concurrency = 5

[ffmpeg]
threads = 2
nice = 20
7

Then increase threads value to have faster transcoding jobs

8

Will try that.

Also, where do i find reference information, to understand the relation of concurrency & threads values with performance?

9

@Chocobozzz

With this config:
2 servers (8 CPUs), each with -

[jobs]
concurrency = 4

[ffmpeg]
threads = 8
nice = 20

it took a total of 10 videos-

  • 5 mins. to get imported
  • 4 hours to get transcoded (HLS only: 1080p, 720p, 480p)

This seems to me really slow!

Kindly correct me if my perception is inappropriate.

If not, what could be the issue here? And how to remedy it?

10

Try with concurrency = 1 as ffmpeg should already use all you CPUs with thread = 8. Then ensure on your server using top that ffmpeg consumes all available CPU.

11

How long are your videos? (« 4 hours » does not give any information, if we don’t know how long videos are).

12

What cpu are you using? What video codec are you using? What quality are you transcoding to?