Reported RAM usage for Linux VM and Node

[bellocarico]

I believe the current RAM usage for Linux VM not to be accurate. I would expect the reported value (under Summary) to be take from free's used column



Where instead it seems like a different calculation is made (used+buffer+cache)


It is normal for Linux to buffer and cache as much as it needs/can but we all know by now that's not real memory demand.
So I would not personally report the buffer/cache (so short yellow bar only) but if you still think there's a value in this can the colors be changed for each of them? e.g.




The very same principle applies to the node.What you call "RAM usage"

It's rather "RAM allocated" which can easily raise above the amount of RAM installed on the system (oversubscribed). Perhaps this second chart should be split into two like Ram usage (actual demand considering balooning if any) and RAM allocated.

If you do go the way I'm suggesting the node should also report on allocated CPUs that can be oversubscribed as well.


Thanks

 

[Dunuin]

So I would not personally report the buffer/cache (so short yellow bar only) but if you still think there's a value in this can the colors be changed for each of them? e.g.

Some color-coding would be nice. But you really want to know the total used RAM including cached/buffered data.
1.) ZFSs ARC cache doesn‘t count as buff/cache and is userspace. So the “used“ part of the bar should be divided in two colors for ARC and non-ARC in case ZFS is used
2.) you want to know when the free (not the available) RAM drops below 20%, because thats when ballooning (hardcoded) and KSM (default but changeable) will kick in

For VMs this is sadly not possible as PVE only gets the guests RAM usage from the qemu-guest-agent and that is just reporting used, total and free RAM. It won‘t give PVE any information on how much RAM is used for caching, so PVE can't know how much that VM got as "availabe" or “buff/cache“.

 

[Neobin]

Such a nice color-separation would be awesome, especially for the node memory usage and even more in combination with ZFS.

For reference (partially):
https://bugzilla.proxmox.com/show_bug.cgi?id=1454