High Cache Usage in Ubuntu 22.04.1 VM - Is it my configuration or Nextcloud?

[echumley]

Hey all! I've tried looking this up but no other forum posts I could find had the exact same situation.

Problem: Proxmox reports high RAM usage of my Ubuntu Server 22.04.1 VM running Nextcloud AIO, but on the server it shows that's in fact the high cache value that's being reported. It's always sitting at ~80-90% no matter how much RAM I allocate even though the only RAM actually being used is chilling at ~3GB. This only happens over time and when the VM is rebooted, the overall RAM usage starts low and climbs back to 80% over the course of a few hours.

Setup
======
Server: Dell Poweredge R630 with 24C/48T, 128GB RAM, and 8 1.2TB drives in RAIDz-2
Proxmox Version: Proxmox VE v8.0.3
VM Configuration: Almost stock template with the qemu-guest-agent.service installed, cloned, and Nextcloud AIO installed on Docker. Memory ballooning is enabled.

Useful Screenshots
=================
VM htop sorted by highest memory usage:


VM free -h:


Proxmox Node arc_summary -s arc & free -h:


Other issues that might be related:
Extremely slow upload speed to Nextcloud. Probably unrelated.

Anything helps as this is stumping me. This is a test of an "in production" server so I'd like to figure out how to get the VM to release that cache and free it up for other services. Thanks y'all!

EDIT: The high cache usage seems to start whenever a user starts to upload a file.

 

[gurubert]

What you see is normal Linux kernel behaviour. It tries to use all available RAM for the filesystem cache.

 

[echumley]

What you see is normal Linux kernel behaviour. It tries to use all available RAM for the filesystem cache.

This is what I've been told, but it's certainly odd that this is the first time I've ever experienced it. I've ran this version of Ubuntu Server and Nextcloud before and this is the first time it's done this. Thanks for the reply!

 

[LnxBil]

What you see is normal Linux kernel behaviour. It tries to use all available RAM for the filesystem cache.

It's the same in every OS, yet those report the memory as free instead of "as cache" / "used". This is especially the case for Windows and therefore people learned to interpret it that way, which is technically wrong but yield a LOT of posts here on the forum (or in more general non-windows forums).

If you don't want to cache so much, reduce the memory, not increase it. Normally, everything is cached (even non-sync writes) after it is read until you hit a high watermark level and then older entries are remove. Therefore you may have bogus transfer speeds with unusual high start speed only to drop at some point. In the background, the cache gets full and is then flushed out.