Hi,
On 03/01/2017 10:55 PM, Andreas Steinel wrote:
Hi Gilberto,
ah yes, now it's clear.
Yes, this would be a good thing, but it is very hard to do that in
terms of "finding a metric". There is memory pressure, cpu load and
possible local storage (for non-shared-storage migrations), but should
be doable. I think you can file a feature request to get this
implemented.
It is somewhat on our radar, but yes as you said the metric finding is a
problem which could be not easy to get right.
We thought about starting with static loads manually assigned to each
VM/CT, and a capacity for the hosts, this would balance the load of to
get started VMs over the cluster.
If this will get advanced I'd add some dynamic metric aggregation.
For the metrics, I'd say CPU and Memory and eventually Network load
could be possible inputs in our control loop.
The main problems/pitfalls would be:
* the cluster acts as a non linear system, not the easiest to control
(automatically) :)
* never ever update control loop faster than your systems (cluster, or
local nodes) sample and delay time
(Totzeit, in German not sure if I got the correct translation), this
would just make the system unstable sooner or later.
* the worst case delay time would be (highly) probably the one of a
migration, during that time we may even generate more load (i.e. on a
network metric) dependent of the cluster/node setup
* as our delay time is quite huge we need to look at reducing
oscillation, it should be nonexistent.
* the very very different setups in which Proxmox VE gets deployed
So as a result of this local storage is not an ideal metric, imo.
It act as an PT2 element so we increase delay time further (we need to
read and copy all disks), a better solution for this is to move this
property to storage technology, i.e. use Ceph, Sheepdog, GlusterFS, etc..
Anyway, nothing fixed planned, I thought just a little about it and we
consider it in general.
cheers,
Thomas
Best
Andreas
On Wed, Mar 1, 2017 at 10:35 PM, Gilberto Nunes
wrote:
Hello Andreas...
Perhaps I don't make myself clear...
With Load Balance I mean when in a node cluster there is a high load, a VM
could to automatic migrated to other node with less load... Something like
that...
Sorry if I messed up the concepts...
Perhaps you can enlight me.
Regards
2017-03-01 18:25 GMT-03:00 Andreas Steinel :
Hi Gilberto,
What do you want to load balance? PVE is a IaaS architecture and
normally you would implement balancing on a SaaS architecture, so one
layer higher (towards the application).
Best,
Andreas
On Wed, Mar 1, 2017 at 12:31 PM, Gilberto Nunes
wrote:
Hello friends
Is there any plan to implement Load Balance in next releases of Proxmox
VE??
I think this is could be a nice feature to add in main stream version
What about?
--
Obrigado
Cordialmente
Gilberto Ferreira
Consultoria em Servidores e Serviços Linux | Virtualização Proxmox |
Zentyal Server | Zimbra Mail Server
(47) 3025-5907
(47) 99676-7530
Skype: konnectati
www.konnectati.com.br
___
pve-devel mailing list
pve-devel@pve.proxmox.com
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
___
pve-devel mailing list
pve-devel@pve.proxmox.com
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
--
Obrigado
Cordialmente
Gilberto Ferreira
Consultoria em Servidores e Serviços Linux | Virtualização Proxmox |
Zentyal Server | Zimbra Mail Server
(47) 3025-5907
(47) 99676-7530
Skype: konnectati
www.konnectati.com.br
___
pve-devel mailing list
pve-devel@pve.proxmox.com
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
___
pve-devel mailing list
pve-devel@pve.proxmox.com
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
___
pve-devel mailing list
pve-devel@pve.proxmox.com
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel