📚 How to properly set buff/cache consumption. Linux starts to swap memory, even that there is a lot lot "available" memory. For qemu kvm virtual host.

How to properly set buff/cache consumption. Linux starts to swap memory, even that there is a lot lot "available" memory?

I have following free -h output

total: 125g

used: 54g

free: 1g

shared: 1,3g

buff/cache/: 70g

available: 67g

swap total: 151

swap used: 10gb

​

​

cat /proc/meminfo

buffers: 274496kb

cached: 69439892kb

​

​

I tried to set vm.swappiness=1

and tried vfs_cache_pressure=150

​

But I didn't understand vfs_cache_pressure meaning. Some sources say that the higher the value the faster the system reclaims cache, other sources say the other way around.

​

1) How do I determine what process (virtual mahine in my case) uses a lot of buff/cache?

2) What are the best ways to make Linux kernel to prefer reclaim cached memory before using swap?

I don't want to disable swap altogether because it can lead to OOM Killer.

And when swap is used - it leads to huge performance drops for some of my applications (related to machine learning)

3) What are kernel options that I can change to improve performance of disk caching?

4) What are the ways to change qemu kvm options to lessen use of cache?

Maybe disable writeback cache for virtual disks?