diff --git a/jellyfin/readme.md b/jellyfin/readme.md
index e9abbca..f54ed16 100644
--- a/jellyfin/readme.md
+++ b/jellyfin/readme.md
@@ -2,21 +2,207 @@
###### guide-by-example
-
+
+
+
# Purpose & Overview
-Stream movies/tv-shows/music to a browser, or a [large selection of devices and services.](https://jellyfin.org/clients/)
+Stream movies, shows, music to a phone, a TV, a browser, ...
+Something like your own Netflix.
-* [Official site](https://jellyfin.org/)
+* [The Official site](https://jellyfin.org/)
* [Github](https://github.com/jellyfin/jellyfin)
* [DockerHub](https://hub.docker.com/r/jellyfin/jellyfin/)
-Jellyfin if a free media system, an alternative to proprietary Plex.
-The core server side is written in C#, web client in Javascript,
-and a number of other clients written in various languages and frameworks.
+Jellyfin is free and opensource without any limitation.
-Starting point for me was [this viggy96 repo](https://github.com/viggy96/container_config)
+The server can be installed on windows, or on linux or run as a docker container.
+It's written in C#, web client in Javascript
+and a number of other [clients](https://jellyfin.org/clients/) written in
+various languages and frameworks.
+
+
+
+Video - core concepts
+
+
+
+* a **video file** is a bunch of pictures - **frames**,
+ packed in to one file.
+* To save disk space and bandwidth its **compressed** using a video standard/codec.
+ * **H.262** - also called MPEG-2 - stuff of the past
+ * **H.264** - the most common now, also called **AVC** or MPEG-4
+ * **H.265** - also called **HEVC**, fast spreading, 50% improved over H.264,
+ but fees for it got more expensive with convoluted patent pools
+ * **AV1** - the future, no royalty fees, more improvements.
+ The successor to VP9, which was developed by google.
+* Ways to transcode
+ * **Software** - cpu does the job, uses some library, it is **very cpu heavy**
+ a phone doing a software playback would either stutter, or be through
+ the entire battery in 30 minutes.
+ * **Hardware** - there is a dedicated hardware - `ASIC` - a tiny part
+ of a cpu/gpu/soc that is designed for just one thing -
+ to transcode a specific video standard. It is extremely efficient at it.
+* `standard` vs `codec`
+ * **Video Compression Standard** - set of specifications to follow, like a recipe,
+ examples are H.264, H.265, AV1,..
+ * **Codec** - The actual software or hardware implementation of a standard,
+ a cook that follows a recipe and gives you a finished meal.
+ Examples are `x264`, `x265`, `libaom`, or hardware ASICs on a gpu
+ or a cpu like nvidia `NVENC` or `intel quick sync`.
+* Other terminology
+ * **Bitrate** - how much data per seconds flow, for example 2Mbps or 20Mbps,
+ higher bitrate means better quality and larger files/bandwidth needs
+ * **Decode** - taking a compressed video file and turning it into a viewable format.
+ * **Encode** - compressing raw video in to a specific video format
+ * **Transcode** - converting a video format in to a different format,
+ consists of both decode and encode steps
+ * **Direct Play** - No transcoding required. The device that is trying to play
+ the video is able to do so with its hardware.
+ * **HDR** - High Dynamic Range. A new technology to improve contrast and visible
+ details but lacks wide hardware support yet.
+ The opposite is SDR - Standard Dynamic Range.
+ * **ffmpeg** - a command line application, open source,
+ contains codec libraries to be a single tool for all video manipulation.
+ Is often the one under the hood of any media related software, doing the work.
+
+
+
+---
+---
+
+
+Jellyfin transcoding
+
+
+
+Jellyfin streams media files to client applications.
+In most cases it's **direct play**, no transcoding required and any machine
+can do dozens of **concurent streams** in direct play.
+
+But there are cases **when a client does not support codec** of the video file.
+Example is Firefox playing a movie encoded in HEVC(x265).
+Jellyfin knows which codecs client supports and if needed
+it **transcodes video on the fly**, as it's being streamed.
+
+Another reason can be if you set hard limit on upstream bandwith use.
+In `Dashboard` > `Playback` > `Streaming` setting lets say
+[2Mbps.](https://i.imgur.com/7hSf4e1.png)
+
+Out of the box, jellyfin is set to **use cpu for software-cpu transcoding**.
+It will work fine for one or two simple 1080p concurent streams,
+unless the cpu is really underpowered/old.
+In any case it can put [a heavy load](https://i.imgur.com/rjNFN1J.png)
+on the CPU.
+
+### Hardware acceleration
+
+* [official docs](https://jellyfin.org/docs/general/administration/hardware-acceleration/)
+
+Ideally jellyfin is running on something that can offer hardware accelerated
+transcoding, using iGPU which greatly improves performance and power consumption.
+
+* **Intel** - QSV - intel quicksync. Go-To recommendation and ideal for most users,
+ high quality, fast, reliable drivers, cheap and low power consumption as an igpu.
+ Most used so huge userbase.
+* **AMD** - VAAPI when in docker - kinda shunned as historicly it had the worst
+ encode quality. [AMD improved some](https://youtu.be/H0pCpNT4b-Q),
+ but so did the competition so it's still not as good. But in real use
+ most people would likely not be able to tell difference.
+ AV1 encode supported by their newest stuff finally concludes the discussion
+ as it's praised for great quality.
+* **nvidia** - NVENC - quality not as good as intel, but better than amd,
+ driver limit of 8 encoding sessions. Not an ideal solution as a pcie card will
+ consume some extra power.
+* **Intel Arc** - QSV - relatively new hardware, should be great,
+ but be careful about idle power consumption,
+ some reviews post \~30W idle for arc while \~5W for amd/nvidia.
+
+But to repeat:
+Very likely majority of media will be in h.264 and h.265 and **overwhelming
+majority of devices can play these straight up - direct play**. So no need to
+overthink transcoding hardware much.
+That is unless you want to go 4K + HDR.
+
+### 4K and HDR
+
+
+
+With 4K resolution often comes **HDR** - High Dynamic Range.
+If its played on HDR device all is fine and theres better contrast and details.
+But non-HDR devices need transcoding with tonemapping.
+**Tonemapping** is what converts HDR content to **SDR** - Standard Dynamic Range.
+Without tonemapping the colors would be heavily desaturated - washed out.
+
+Some **clients**, like Findroid on android use mpv for playback, which somehow
+deals with the HDR to SDR conversion on its own.
+Might be better since its direct play, but might be worse because of putting load
+on the device.
+
+### Testing various hardware
+
+Archlinux straight on metal, docker Jellyfin.
+Testing [10x FHD streams](https://i.imgur.com/nP71y0E.png),
+and 4K+HDR+tonemapping till see stutter
+Testing by running movies, x265 encoded, in edge under linux.
+`Throttle Transcodes` is turned off so that the movie is transcoded in full,
+not just 3min segments.
+
+Results
+
+* ryzen **7700X**
+ * 10x streams FHD - [pass](https://i.imgur.com/NAyfXmG.png)
+ * 4K+HDR+tonemapping - not tested
+* intel **n200**
+ * 10x streams FHD - [pass](https://i.imgur.com/nP71y0E.png)
+ * 4K+HDR+tonemapping - not tested
+* ryzen **8600G**
+ * 10x streams FHD - [pass](https://i.imgur.com/9R60Spw.png)
+ except [weird green artifacts](https://i.imgur.com/s4lSJRI.png)
+ for ~30sec when starting all 10 streams at once
+ * 4K+HDR+tonemapping - not tested
+ * AV1 encoding - [worked](https://i.imgur.com/ZaQRIAc.png)
+* intel **i5-12600k**
+ * 10x streams FHD - [pass](https://i.imgur.com/BGXOeGS.png)
+ * 4K+HDR+tonemapping - 4 streams
+* ryzen **4350GE**
+ * 10x streams FHD - 6 max, otherwise stutter
+ * 4K+HDR+tonemapping - 1x stream [barely](https://i.imgur.com/8iairvO.png)
+
+
+Software used for monitoring the cpu and gpu usage
+
+* [btop](https://github.com/aristocratos/btop)
+* `intel_gpu_top` from [intel-gpu-tools](https://archlinux.org/packages/extra/x86_64/intel-gpu-tools/) package
+* `radeontop` from [radeontop](https://archlinux.org/packages/extra/x86_64/radeontop/) package
+
+
+
+---
+---
+
+# Clients
+
+Jellyfin's open source nature allows developers to create unofficial client apps
+for various platforms.
+
+* [Awesome Jellyfin Clients List](https://github.com/awesome-jellyfin/awesome-jellyfin/blob/main/CLIENTS.md)
+
+The ones I tried and find good to know about.
+
+### [Findroid](https://github.com/jarnedemeulemeester/findroid)
+
+Feels bit better in user interface and control.
+Does not jump subtitles like the official client, also seems its always directplay
+even 4k+HDR, have to investigate how the fuck does it do.
+Uses mpv, but does it mean theres some big load on a phone when compared
+to the official app?
+It ignores max streaming bandwith limit when server streams over the internet.
+
+### [Finamp](https://github.com/jmshrv/finamp)
+
+Aimed for music. Did not tested much yet.
# Files and directory structure
@@ -45,14 +231,15 @@ Starting point for me was [this viggy96 repo](https://github.com/viggy96/contain
You only need to provide the two files.
The directories are created by docker compose on the first run.
-# compose
+# Compose
+
+* [the official documentation](https://jellyfin.org/docs/general/installation/container)
A relatively simple compose.
The only atypical thing is the **passthrough** of the graphic card
-for hardware accelerated transcoding.
-In the `devices` section a passthrough of a graphic card is done,
-`/dev/dri/renderD128` refering to the first gpu of the system
+for hardware accelerated transcoding.
+It's in the `devices` section `/dev/dri/renderD128`
In `group_add` section permissions are set.
You want to execute the command: `getent group render | cut -d: -f3`
to get the correct group number for you system and set it in.
@@ -74,13 +261,14 @@ services:
group_add:
- "989" # match output of: getent group render | cut -d: -f3
volumes:
- - ./jellyfin_config:/config
- ./jellyfin_cache:/cache
- - /mnt/bigdisk/serialy:/media/tv:ro
+ - ./jellyfin_config:/config
+ - /mnt/bigdisk/serialy:/media/shows:ro
- /mnt/bigdisk/filmy_2:/media/movies:ro
ports:
- "8096:8096" # webGUI
- "1900:1900/udp" # autodiscovery on local networks
+ - "7359:7359/udp" # autodiscovery on local networks
networks:
default:
@@ -93,6 +281,11 @@ networks:
# GENERAL
DOCKER_MY_NETWORK=caddy_net
TZ=Europe/Bratislava
+
+# JELLYFIN
+
+# url or ip address
+JELLYFIN_PublishedServerUrl=https://tv.example.com
```
**All containers must be on the same network**.
@@ -116,133 +309,153 @@ tv.{$MY_DOMAIN} {
# The first run
-
+Click through the basic setup.
-Click through basic setup.
-
-WORK IN PROGRESS
-
-WORK IN PROGRESS
-
-WORK IN PROGRESS
-
-# Transcoding
-
-### The basics
-
-* a **video file** is just a bunch of pictures - **frames**,
- somehow packed in to one file.
-* To save up disk space and bandwidth its **compressed** using some video
- standard/codec.
- * MPEG-2 - stuff of the past
- * **H.264** - the most common now
- * **H.265** - also called **HEVC**, fast spreading, 50% improved over H.264
- * **AV1** - the future, open codec - no licencing fees, more improvements
-* Ways to transcode
- * **Software** - cpu does the job, uses some library, it is **very cpu heavy**
- a phone doing a software playback would either stutter, or be through
- the entire battery in 30 minutes.
- * **Hardware** - there is a dedicated hardware - a tiny part of a cpu/gpu/soc
- that is designed for just one thing - to transcode a specific video standard.
- That means it is **extremely efficient** at it.
-* Terminology
- * **Decode** - taking a compressed video file and turning it into a viewable format.
- * **Encode** - compressing raw video in to a specific video format
- * **Transcode** - converting a video format in to a different format,
- consists of both decode and encode steps
+* Change the server's name
+ `Dashboard` > `General` > `Server name`
+*
-Ideally you deploy jellyfin somewhere with an igpu to get hardware accelerated
-transcoding, but it is far from required.
-For most people, majority of media will be in H.264 or H.265 which will be
-**direct play** - no transcoding required on most devices.
-Even if theres occasional need to transcode, average cpu can do one or two streams.
+
+Intel specific setup
-If you plan to serve more people and have larger library you should
-definitly plan to have something with igpu
+* [the official documentation](https://jellyfin.org/docs/general/administration/hardware-acceleration/intel/#configure-with-linux-virtualization)
-#### HDR
+Assuming an intel cpu with an igpu of the last \~8 years.
+The `compose.yml` + `.env` should just work.
-The issue starts with 4k content, of which majority also uses
-HDR - High Dynamic Range. This is for benefit of HDR TVs, monitors, phones,...
-To play on non-HDR devices transcoding is always required and not just typical
-transcoding, but also tonemapping as trancoding HDR content without it will make colors seem
-heavily desaturated - washed out.
+### Hardware Accelerated Transcoding.
-* Not all devices like phones, PCs - browsers, TVs, streaming boxes,...
- have build in support for all these standard.
-* If video is in H.265 but firefox on linux cant decode it,
- jellyfin detects this and transcodes it to something that can be played.
+`Dashboard` > `Playback` > `Transcoding`
-### intel igpu
+* `Hardware acceleration` - **Intel QuickSync (QSV)**
+* `QSV Device` - `/dev/dri/renderD128`
+* `Enable hardware decoding for`
+ * Check everything **except** `AV1` and `HEVC RExt 12bit`
+ * if 12th gen+ check everything
+* `Enable hardware encoding` - check
+* `Enable VPP Tone mapping` - check
+* `Enable Tone mapping` - check
+* `Encoding preset` - `Auto`
+* `Throttle Transcodes` - check
+* `Delete segments` - check
-to get "Enable Intel Low-Power H.264 hardware encoder" working
+To be able to use `Enable Intel Low-Power H.264 hardware encoder`
-* `echo "options i915 enable_guc=2" > /etc/modprobe.d/i915.conf`
+* On the docker host
+ `echo "options i915 enable_guc=2" > /etc/modprobe.d/i915.conf`
* `mkinitcpio -P`
* reboot
-The above compose basic setup worked for me
+
-* ryzen 1700, headless, without any gpu
-* modern intel cpus with igpu - n200, i5-125600k
-* modern amd ryzens with igpu - 7700x, 5500GT
+---
+---
-but how to setup things might change over time so one should check
-[the official documentation](https://jellyfin.org/docs/general/administration/hardware-acceleration/intel#configure-with-linux-virtualization)
+
+AMD specific setup
+* [the official documentation](https://jellyfin.org/docs/general/administration/hardware-acceleration/amd#configure-with-linux-virtualization)
+* [some videos](https://youtu.be/H0pCpNT4b-Q) with [some takes](https://youtu.be/UNJLDS5gC7o)
+ on [AMD transcoding](https://youtu.be/pnvp9DtqVjo)
+Assuming an amd cpu with vega or RDNA.
+The `compose.yml` + `.env` should just work.
-# Specifics of my setup
+`Dashboard` > `Playback` > `Transcoding`
-* no real long term use
-* findroid app does not jump subtitles like official one
-* amd cpu and no gpu, so no experience with hw transcoding
-* media files are stored and shared on trunas scale VM
- and mounted directly on the docker host using [systemd mounts](https://forum.manjaro.org/t/root-tip-systemd-mount-unit-samples/1191),
- instead of fstab or autofs.
+* `Hardware acceleration` - **Video Acceleration API (VAAPI)**
+* `QSV Device` - `/dev/dri/renderD128`
+* `Enable hardware decoding for`
+ * Check everything **except** `AV1` and `HEVC RExt 12bit`
+ * if 7000+ check everything
+* `Enable hardware encoding` - check
+ * `Allow encoding in AV1 format` - check if 8000G+
+* `Enable Tone mapping` - check, but needs work
+* `Encoding preset` - `Auto`
+* `Throttle Transcodes` - check
+* `Delete segments` - check
- `/etc/systemd/system/mnt-bigdisk.mount`
- ```ini
- [Unit]
- Description=12TB truenas mount
+Tone mapping
- [Mount]
- What=//10.0.19.11/Dataset-01
- Where=/mnt/bigdisk
- Type=cifs
- Options=ro,username=ja,password=qq,file_mode=0700,dir_mode=0700,uid=1000
- DirectoryMode=0700
+* for my arch host installing these packages
+* `sudo pacman -S xf86-video-amdgpu libva-mesa-driver mesa-vdpau vulkan-radeon vulkan-tools`
+* reboot
- [Install]
- WantedBy=multi-user.target
- ```
+
- `/etc/systemd/system/mnt-bigdisk.automount`
- ```ini
- [Unit]
- Description=12TB truenas mount
+---
+---
- [Automount]
- Where=/mnt/bigdisk
+
+Mounting network shares
- [Install]
- WantedBy=multi-user.target
- ```
+If the media files are stored and an smb share and should be mounted directly
+on the docker host using [systemd mounts](https://forum.manjaro.org/t/root-tip-systemd-mount-unit-samples/1191),
+instead of fstab or autofs.
- to automount on boot - `sudo systemctl enable mnt-bigdisk.automount`
+`/etc/systemd/system/mnt-bigdisk.mount`
+```ini
+[Unit]
+Description=12TB truenas mount
+
+[Mount]
+What=//10.0.19.11/Dataset-01
+Where=/mnt/bigdisk
+Type=cifs
+Options=ro,username=ja,password=qq,file_mode=0700,dir_mode=0700,uid=1000
+DirectoryMode=0700
+
+[Install]
+WantedBy=multi-user.target
+```
+
+`/etc/systemd/system/mnt-bigdisk.automount`
+```ini
+[Unit]
+Description=12TB truenas mount
+
+[Automount]
+Where=/mnt/bigdisk
+
+[Install]
+WantedBy=multi-user.target
+```
+
+to automount on boot - `sudo systemctl enable mnt-bigdisk.automount`
+
+
+
+---
+---
# Troubleshooting
+#### Playback failed due to a fatal player error
+
+
+
+* docker host - go to `jellyfin/jellyfin_config/log/`
+* the last file in the folder is the last playback attempt
+* paste that output in to chatgpt
+
+#### Autodiscovery not working
+
+to check if autodiscovery port is open and server is responding:
+
+* `echo -n 'Who is JellyfinServer?' | nc -u -b 10.0.19.4 7359`
+
+#### Unable to connect to the selected server right now
+
*We're unable to connect to the selected server right now. Please ensure it is running and try again.*
-If you encounter this, try opening the url in browsers private window.
-If it works then clear the cookies in your browser.
+* try opening the url in browsers private window
+* if that works then clear the cookies in your browser
-*No playback at all but GUI works fine*
+#### No playback at all but GUI works fine
Might be no access to network share, for example if dockerhost boots up faster
than NAS.