[PVE-User] Automated load balancing script(s)

[David Black]

Thank you for all the info and helpful suggestions Alexandre!

I didn't know about the /cluster/resources so will look to using just that vs. other sources.   And yes, at the moment I'm using total bogomips as a relative measure of available host CPU.   None of my hosts have hyperthreading either, but it could be considered as for instance, a logical CPU adds 20% of a physical CPU's bogomips to the total.

Dave

----- Original Message -----
> Hi,No, I don't think someone have had time to implement this and it's
> not for now on the roadmap.
> 
> But help is always welcome :)
> 
> 
> 1 thing that must be tricky, it's that we don't have a master server
> on a proxmox cluster.
> So, one way is that each server can monitor his load, and migrate to
> other servers with less load.
> 
> An interesting paper here with some algos :
> http://research.ijcaonline.org/volume46/number6/pxc3879263.pdf
> 
> 
> Other things to check:
> 
> for cluster with differents hosts models/cpus:
> - How to know that a host is more or less powerfull than other ?
> (bogomips ?)
> - don't allow live migration on a server with less cpu/cores than a
> guest vm.
> 
> -We also need to allow/disallow a vm to be auto migrated on a
> differents hosts.
> -Avoid to migrate a vm for some spike load. (use average value on
> last X minutes)
> 
> It could be implemented as a daemon on the proxmox host.
> 
> You can get vm and host stats through pve api
> #pvesh get /cluster/resources
> 
> 
> Hope It can help you.
> 
> Don't hesitate to send mail to pve-devel mailing list.
> 
> Regards,
> 
> Alexandre
> 
> ----- Mail original -----
> 
> De: "David Black" <proxmox at 1000.relay.net>
> À: pve-user at pve.proxmox.com
> Envoyé: Mardi 4 Septembre 2012 19:49:28
> Objet: [PVE-User] Automated load balancing script(s)
> 
> 
> I'm having a great time with Proxmox VE 2.1 in a three-node cluster
> with Solaris and Linux-based NFS storage, and find the only feature
> I really miss is some kind of automated load balancing - live
> migration of VMs to optimize performance.
> 
> This past weekend I wrote scripts to collect node and VM stats, and
> am at the point where it's time to implement an algorithm that takes
> the available data and chooses when and what to migrate where. That
> may be the most difficult part, but there are simple ways to start
> (then get fancier later).
> 
> Before I continue, has anyone already done something similar, and is
> this kind of feature on the Proxmox VE roadmap?
> 
> Thanks!
> Dave
> 
> 
> _______________________________________________
> pve-user mailing list
> pve-user at pve.proxmox.com
> http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
>