Inquiry About Potential Changes to Auto Balloon Functionality

[sloong]

Dear PVE Team,

I hope this message finds you well.

I'd like to inquire about plans for future improvements to the Auto Balloon feature in PVE.

Going through the code, I get that the current implementation seems to make memory adjustments based on host memory usage, with the logic being:
1. keep host memory at 80% occupied target
2. the total amount of adjustable memory is physical memory * 0.8 - the total amount of all memory that cannot be adjusted
3. vm's balloon = adjustable memory/total_shares*shares. it's fixed value.
This works well with a stable memory footprint, but I think Hyper-V's, based on the vm's real memory usage allocation would be more efficient.
For example, certain situations that require a little memory to keep running most of the time, and a lot of memory for a short period of time in certain situations.
In this way, memory allocation for each VM can be managed more precisely, especially when some VMs are under low load, which can effectively minimize memory waste and adjust memory in time when the load is higher.

Please consider the above suggestions.

Thank you for your hard work and for considering user feedback!

I look forward to your response.

Best regards,

 

[Falk R.]

That all sounds quite nice, but your method also has its weak points. Some software uses an extremely large amount of RAM as a cache and the RAM is therefore always occupied and often active. If you take some RAM away from these cache monsters, they often still run just as fast, but with your method you would never take RAM away from them.

I only see balooning as a feature for absolute emergencies anyway. Most of the time you have a cleanly sized system and balooning never comes into play there.