As previously announced the m68k porters met this weekend in Kiel this year.
It was really great to meet the persons in reality with whom you’re working for several years now!
You can find a short summary of the results here.
2 thoughts on “re: Killing Servers with Virtualisation and Swap”
Comments are closed.
I guess your solution doesn't work for the exact fact mentioned at the start of your post. Now if some server starts trashing, it will still eat the disks (the whole array in this case) and other virtual machines will grind to a halt, so in my eyes it seems like the issue hasn't changed.
Assume that one VM starts swapping a lot in a setup where all swap areas are on one LVM on one RAID. All disks will be busy in moving their heads, slowing down all other VMs as well. Right?
So, with the different setup you'll gain a) flexibility (you can either choose dedicated swap partitions for some VMs or still remain on your big LVM/RAID) and b) it doesn't slow down all other VMs that much anymore. Yes, it will slow down as well, but not as much as when it's swapping across all disks.
When you setup for example one swap partition on a single partition outside of the LVM/RAID for use in a VM and this VM swaps, it will slow down itself because of its head movements. This will give time for the LVM/RAID to move all other heads of that array, including the swapping disk of the VM. This will slow down the RAID of course to some degree, whereas it can read/write from all other 7 or 11 disks in full speed.
It would be nice, though, if someone could test this, of course. But I'm confident that this setup would give better performance.