System requirements for low-end UX

[davidecavestro]

While still far from having a fine-tuned config to share, I certainly have Immich running on a NAS equipped with just 4Gb of RAM and a dual-core Celeron N4505: both face recognition and typesense are enabled.

So I would like to share my experience and collect feedback, so that other low-end users give Immich a try, as it really works like a charm even on devices like mine.

The first time I considered running Immich on this hardware, I read the official docs and decided to give up: minimum requirements were not met.

Then - after a lot of headaches with other software - I gave it a try and: KABOOM! I started with ML and search disabled, then enabled them.

Now my configuration is still not fine-tuned, but for sure docker is running Immich like this:

This is an excerpt of my compose.yml:

version: "3.8"

services:
  immich-server:
    ...
    deploy:
      resources:        
        limits:
          cpus: '1'
          memory: '1024M'

  immich-microservices:
    ...
    deploy:
      resources:        
        limits:
          cpus: '0.6'
          memory: '768M'

  immich-machine-learning:
    ...
    deploy:
      resources:
        limits:
          cpus: '1.2'
          memory: '1300M'

  typesense:
    ...
    deploy:
      resources:
        limits:  
          cpus: '0.4'
          memory: '512M'

  # no limits in place for remaining services

But please note that I also set some global cgroup limits to 3.2Gb of memory and 1.8 CPU cores for the whole docker stuff, so that - in case of spikes - single limits are further bounded by global ones:

#!/bin/bash

CG_NAME=docker

MEMORY=3200
CPU=180
# Create the cgroups in case they don't exist yet
mkdir -p /sys/fs/cgroup/memory/$CG_NAME
mkdir -p /sys/fs/cgroup/cpu/$CG_NAME

# Set the limits
echo $(($MEMORY * 1024 * 1024)) > /sys/fs/cgroup/memory/$CG_NAME/memory.limit_in_bytes

echo 100000 > /sys/fs/cgroup/cpu/$CG_NAME/cpu.cfs_period_us
echo $(($CPU * 1000)) > /sys/fs/cgroup/cpu/$CG_NAME/cpu.cfs_quota_us

At the moment this system is also running a jellyfin instance and some scheduled rclone copy tasks.

TBH I cannot assure containers are never killed this way: maybe it could happen if I launch a lot of concurrent batch jobs.
THat said I'm sure this way it just works properly for me, and I will refine it if needed.

Anyone tried something similar?
What was the oucome?

[alextran1502]

Thank you for sharing your experiences. We are looking forward to even further optimization of Immich to serve use case such as yours!

[dro159]

@davidecavestro
Thank you for this. I have an Intel NUC with the same processor and can definitely use this tuning. Just a question, the bash script above, was that all you ran to set global cgroup limits? I tried it and getting access denied when the script tries to echo the limits.

[davidecavestro]

Just a question, the bash script above, was that all you ran to set global cgroup limits? I tried it and getting access denied when the script tries to echo the limits.

I confirm I only run this script for cgroup limits, but as root user.
You can try with sudo depending on your system. Mine is a TOS 5 so I have a sort-of-admin-user that acts as root.
Moreover the commands probably need some changes in case you have cgroups2.

[dro159]

Thank you! the problem was cgroups2 being used in my os by default.

For those using cgroups2, here is the modified bash script in case anyone needs it. This worked for me.

#!/bin/bash

CG_NAME=docker

MEMORY=3200
CPU=180
# Create the cgroup in case it doesn't exist yet
mkdir -p /sys/fs/cgroup/$CG_NAME

# Set the limits
echo $(($MEMORY * 1024 * 1024)) > /sys/fs/cgroup/$CG_NAME/memory.max
echo $(($CPU * 1000)) 100000 > /sys/fs/cgroup/$CG_NAME/cpu.max

[bt-88]

Thanks for sharing. I will give this a try as I'm experiencing problems when attempting to run Immich on my QNAP TS-251+ NAS. It's a Intel® Celeron® J1900 4-core 2-2.4Ghz CPU with 8GB of DDR3L memory. The machine learning service seems to eat all resources as soon as I upload my first batch of photos, ultimately suffocating all containers.

[zoro273]

@bt-88 how did it go ? I have a similar machine as yours and was keen to give it a go !