Slow video transcoding ffmpeg h264 0.6 fps

[www-data]

Hi since weeks I'm trying to reach best and fast method to transcode Full HD h264 online media source to SD output with no success .
I'm using rented VPS that have `Intel xeon E3-1240 v5`
The video transcoding start fast at 1x speed "Im using -r option" but after few seconds it drop to 0.02x with ~ 0.6fps

this is transcoding parameters I use

ffmpeg -re  -i  https://someurl/test.ts  -i  logo.png  -filter_complex "crop=iw:ih-100:iw:-100:keep_aspect=1,scale=w=854:h=480,overlay=x=(main_w-overlay_w)-20:y=40" -bufsize 1216k -c:v libx264 -preset ultrafast -qscale 0 -vb 290k -c:a aac -ab 35k -f flv "rtmp://rtmp-api.facebook.com:80/rtmp/ssssssssssss"

do I need to add/remove any more tweaking codes , and do you thing is't CPU throttling (form my host ) prevent me from having needed CPU resource ?

[pandy]

Are you able to run same command on local computer (you can redirect output to null or file)?

[jagabo]

Running that ffmpeg command line with local files (input and output on SSD), I get about 25 fps. For a quick comparison, I ran a similar encoding using AviSynth (same crop, resize, overlay) and the x264 cli encoder (no audio handling) and it encoded at 93 fps. Using the veryfast preset instead of ultrafast encoded at 91 fps and gave much better quality.

[pandy]

Running that ffmpeg command line with local files (input and output on SSD), I get about 25 fps. For a quick comparison, I ran a similar encoding using AviSynth (same crop, resize, overlay) and the x264 cli encoder (no audio handling) and it encoded at 93 fps. Using the veryfast preset instead of ultrafast encoded at 91 fps and gave much better quality.

With '-re' processing is performed with native frame speed - such framerate encoding drop as described by OP means probably some issue - not sure if VBV bufer size and expected bitrate are unrealistic and they forcing some weird libx264 behaviour . I saw some strange behaviour on virtual machines with real time processing and IMHO they are (VM) not reliable unless specially configured.

[www-data]

I'm doing rebroadcasting of live stream my current home connection can't allow me to do the same commands (since I only have 1Mbit Internet connection ),
My suspension is on Video filter I use could increase CPU usage or the VPS single Virtual cores , didn't have the power to handle Full HD stream (even I face same problem with SD stream now ) frame rate dropped a lot
This is my last code :-

ffmpeg -thread_queue_size 2612 -re -i https://someurl/test.ts -i /root/small.png \
-filter_complex crop=iw:ih-33:iw:-33:keep_aspect=1,scale=w=854:h=480,overlay=x=(main _w-overlay_w):y=(main_w-overlay_w)-12 \
-x264-params nal-hrd=cbr -vb 280k -minrate 280k -maxrate 280k -bufsize 300k -c:v libx264 -c:a aac -ab 38k -ar 44100 -f flv rtmp://localhost/hls/live

Here I tried my best decres CPU usage
BTW I have tow VPS do you agree with the idea that I do video filter on first VPS and send the output to the second one (VIA Nginx RTMP ) , to reduce CPU usage .

I ran a similar encoding using AviSynth (same crop, resize, overlay) and the x264 cli encoder (no audio handling) and it encoded at 93 fps. Using the veryfast preset instead of ultrafast encoded at 91 fps and gave much better quality.

thanks for the value info , I didn't try AviSynth before even x264 directly , my VPS uses Linux Debian 64bit with 512MB of RAM I don't know if these tools runs the same way on Linux

--- Update I completely remove all video filters and the transcoding continued for ~ half hour with no frame dropping



This is my code and ffmpeg output

root@:~# ffmpeg -thread_queue_size 2612 -re -i http://someurl/test.ts -x264-params "nal-hrd=cbr" -vb 280k -minrate 280k -maxrate 280k -bufsize 300k -c:v libx264 -c:a aac -ab 38k -ar 44100 -f flv rtmp://localhost/hls/live

ffmpeg version N-88046-g0cb8369bce-static https://johnvansickle.com/ffmpeg/ Copyright (c) 2000-2017 the FFmpeg developers
built with gcc 6.4.0 (Debian 6.4.0-8) 20171010
configuration: --enable-gpl --enable-version3 --enable-static --disable-debug --disable-ffplay --disable-indev=sndio --disable-outdev=sndio --cc=gcc-6 --enable-fontconfig --enable-frei0r --enable-gnutls --enable-gray --enable-libfribidi --enable-libass --enable-libvmaf --enable-libfreetype --enable-libmp3lame --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libopenjpeg --enable-librubberband --enable-librtmp --enable-libsoxr --enable-libspeex --enable-libvorbis --enable-libopus --enable-libtheora --enable-libvidstab --enable-libvo-amrwbenc --enable-libvpx --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxvid --enable-libzimg
libavutil 56. 0.100 / 56. 0.100
libavcodec 58. 0.100 / 58. 0.100
libavformat 58. 0.100 / 58. 0.100
libavdevice 58. 0.100 / 58. 0.100
libavfilter 7. 0.100 / 7. 0.100
libswscale 5. 0.100 / 5. 0.100
libswresample 3. 0.100 / 3. 0.100
libpostproc 55. 0.100 / 55. 0.100
Input #0, mpegts, from 'https://someurl/test.ts':
Duration: N/A, start: 9655.635011, bitrate: N/A
Program 1
Metadata:
service_name : Service01
service_provider: FFmpeg
Stream #0:0[0x100]: Video: h264 (High) ([27][0][0][0] / 0x001B), yuv420p(progressive), 640x360 [SAR 1:1 DAR 16:9], 25 fps, 25 tbr, 90k tbn, 50 tbc
Stream #0:1[0x101]: Audio: aac (LC) ([15][0][0][0] / 0x000F), 44100 Hz, stereo, fltp, 73 kb/s
Stream mapping:
Stream #0:0 -> #0:0 (h264 (native) -> h264 (libx264))
Stream #0:1 -> #0:1 (aac (native) -> aac (native))
Press [q] to stop, [?] for help
[libx264 @ 0x52ed2c0] using SAR=1/1
[libx264 @ 0x52ed2c0] using cpu capabilities: MMX2 SSE2Fast SSSE3 SSE4.2 AVX FMA3 BMI2 AVX2
[libx264 @ 0x52ed2c0] profile High, level 3.0
[libx264 @ 0x52ed2c0] 264 - core 152 r19 ba24899 - H.264/MPEG-4 AVC codec - Copyleft 2003-2017 - http://www.videolan.org/x264.html - options: cabac=1 ref=3 deblock=1:0:0 analyse=0x3:0x113 me=hex subme=7 psy=1 psy_rd=1.00:0.00 mixed_ref=1 me_range=16 chroma_me=1 trellis=1 8x8dct=1 cqm=0 deadzone=21,11 fast_pskip=1 chroma_qp_offset=-2 threads=1 lookahead_threads=1 sliced_threads=0 nr=0 decimate=1 interlaced=0 bluray_compat=0 constrained_intra=0 bframes=3 b_pyramid=2 b_adapt=1 b_bias=0 direct=1 weightb=1 open_gop=0 weightp=2 keyint=250 keyint_min=25 scenecut=40 intra_refresh=0 rc_lookahead=40 rc=cbr mbtree=1 bitrate=280 ratetol=1.0 qcomp=0.60 qpmin=0 qpmax=69 qpstep=4 vbv_maxrate=280 vbv_bufsize=300 nal_hrd=cbr filler=1 ip_ratio=1.40 aq=1:1.00
Output #0, flv, to 'rtmp://localhost/hls/live':
Metadata:
encoder : Lavf58.0.100
Stream #0:0: Video: h264 (libx264) ([7][0][0][0] / 0x0007), yuv420p(progressive), 640x360 [SAR 1:1 DAR 16:9], q=-1--1, 280 kb/s, 25 fps, 1k tbn, 25 tbc
Metadata:
encoder : Lavc58.0.100 libx264
Side data:
cpb: bitrate max/min/avg: 280000/0/280000 buffer size: 300000 vbv_delay: -1
Stream #0:1: Audio: aac (LC) ([10][0][0][0] / 0x000A), 44100 Hz, stereo, fltp, 38 kb/s
Metadata:
encoder : Lavc58.0.100 aac
frame= 7205 fps= 25 q=34.0 size= 11488kB time=00:04:48.01 bitrate= 326.7kbits/s speed=0.999x

[pandy]

cbr is not a good idea IMHO, focus on quality and guide encoder to limit peak bitrate - it looks like crf should be around 28 - 32 to match expected bitrate - quality will be low but... not sure about specified thread buffer size - this is Kib or something else? native aac has some limitations. filters in ffmpeg are mostly single thread type... they may be explanation of the slow encoding - particularly overlay may be an issue - check color space, perhaps you can use https://www.ffmpeg.org/ffmpeg-filters.html#movie-1 instead two inputs, you may try to experiment with threading type.

[jagabo]

Running that ffmpeg command line with local files (input and output on SSD), I get about 25 fps. For a quick comparison, I ran a similar encoding using AviSynth (same crop, resize, overlay) and the x264 cli encoder (no audio handling) and it encoded at 93 fps. Using the veryfast preset instead of ultrafast encoded at 91 fps and gave much better quality.

With '-re' processing is performed with native frame speed.

Ah, I see. The goal is to have real time processing, not faster or slower.

Removing the realtime constraint, ffmpeg ran the full command line at 150 fps. 130 fps with veryfast instead of ultrafast. So the processing in ffmpeg isn't the problem.

I thought the problem might have something to do with source frames being read out-of-order. But that doesn't seem likely since the ultrafast preset uses no b-frames and only one reference frame.

[www-data]

Are you able to run same command on local computer (you can redirect output to null or file)?

I use the first on my laptop ant reach 0.43x after few minutes


when I use my second code it drooped to 0.74x with 16% CPU usage

ffmpeg -thread_queue_size 2612 -re -i https://someurl/test.ts -i small.png \
-filter_complex crop=iw:ih-33:iw:-33:keep_aspect=1,scale=w=854:h=480,overlay=x=(main _w-overlay_w):y=(main_w-overlay_w)-12 \
-x264-params nal-hrd=cbr -vb 280k -minrate 280k -maxrate 280k -bufsize 300k -c:v libx264 -c:a aac -ab 38k -ar 44100 -f flv NULL

cbr is not a good idea IMHO, focus on quality and guide encoder to limit peak bitrate - it looks like crf should be around 28 - 32 to match expected bitrate - quality will be low but... not sure about specified thread buffer size - this is Kib or something else? native aac has some limitations.

cbr is the only parameter that gived me better pic quality and less CPU usage I didn't try cfr alone before (without bitrate )
removing the audio didn't solve the issue (I tried with -c:a copy ) and didn't even decrease CPU usage either .

filters in ffmpeg are mostly single thread type... they may be explanation of the slow encoding - particularly overlay may be an issue - check color space, perhaps you can use https://www.ffmpeg.org/ffmpeg-filters.html#movie-1 instead two inputs, you may try to experiment with threading type.

But not with my laptop CPU (core i5 3210m) I test it with the SD source and cpu usage were ~16% only .
I didn't try Movie filter method yet as I din't see many people use it before , but I'll give it try .

[poisondeathray]

Where is https://someurl/test.ts from ? Your local home connection to the VPS ?

Maybe your bottleneck is your 1Mbit bandwidth. Next, test a local file input instead of https, locally on your laptop

[www-data]

Where is https://someurl/test.ts from ? Your local home connection to the VPS ?
Maybe your bottleneck is your 1Mbit bandwidth. Next, test a local file input instead of https, locally on your laptop

It's from live video source from Internet it has 600 kbs bitrate ,, anyway I did like what you said when I use local file framerate didn't drops at all it at 1x all the time even I notice q value showed on ffmpeg command at 29.0 , CPU usage were at 25% (I was keep Opera browser open while encoding )

[poisondeathray]

Likely you have bandwidth bottleneck. You can't really do anything in terms of hardware, or commandline, or even expensive VPS for that - it won't help since that's not the problem. It has more to do with the "live video source" bandwidth - it's not feeding your laptop or the VPS fast enough

You can also test VPS, with another URL (a fast server) to rule out VPS bandwidth issue. (But it' s unlikely)

In the 1st post it said "Full HD h264" online media, was that 600kbps too ? Pretty low bitrate for "Full HD"

[www-data]

. It has more to do with the "live video source" bandwidth - it's not feeding your laptop or the VPS fast enough
In the 1st post it said "Full HD h264" online media, was that 600kbps too ? Pretty low bitrate for "Full HD"

My VPS have no network issue I reach 20MB/s download speed and it have 100Mbit/s upload line , - I have two sources one with ~ 5Mbits bitrate speed (The Full HD) and the other source with ~600Kbit/s (SD ) and I don't think the source dropping connection as I was able to download the SD source with my home connection (IDM) and the file were played with correct framerate .
Lat CPU usage on my VPS was always at 100% when I use filter (both codes) but when I use the transcoding without any filters I was able to continue the transcoding for more then hour with few dropping farms from time to time (I guess that from the source ) since my VPS CPU usage were less than 50% .

[poisondeathray]

Where is someurl ? Where is the live stream being sent from?

Laptop some url => 11fps
Laptop local file => 25fps no drop. Perfect.

Both same commandline, same hardware otherwise

This means problem is the source "some url" bandwidth either server up , or down for you locally . At least for your laptop.

Lat CPU usage on my VPS was always at 100% when I use filter (both codes) but when I use the transcoding without any filters I was able to continue the transcoding for more then hour with few dropping farms from time to time (I guess that from the source ) since my VPS CPU usage were less than 50% .

Then the VPS is also bottleneck if it's pegged at 100%. Pay for a better VPS. Pay for better service, or complain to the host

But if you are using "some url" to feed the VPS, then it has 2 bottlenecks