[pve-devel] qemu-server : cgroups implementation through systemd-run

[Alexandre DERUMIER]

>>yes, I do not want to do that. I think it does not really makes sense to 
>>restrict cpuset for 
>>containers if we can restrict overall cpu usage using cpulimit? 



I think it depend if application running inside container can do multithreading.

for example, you have a physical serveur with 20cores.

you can to assign 2cpus.

so 10%.

that mean that container will have 20cores running at 10%.

If an application in container is monothreaded (1 php script for example), I'll use 10%speed of 1core.



Now, same config with settings 2cpu set.

1 monothreaded application, will use 100% of 1core.  
So 10x faster than previously


That's an huge difference.





----- Mail original -----
De: "dietmar" <dietmar at proxmox.com>
À: "aderumier" <aderumier at odiso.com>
Cc: "pve-devel" <pve-devel at pve.proxmox.com>
Envoyé: Samedi 30 Mai 2015 12:49:19
Objet: Re: [pve-devel] qemu-server : cgroups implementation through systemd-run

> >>You think I should call it 'cpulimit' instead of 'cpus'? 
> 
> I mean, if you go in a container, do a "top", you'll always see the full 
> number of cpus of the host. 
> 
> in openvz : 
> cfs_quota_us = cpulimit. 
> cpus = number of cores assign in a container. 
> 
> 
> It's possible to really fix the number of cpus inside the container with 
> cpu.cpuset. 
> (The only problem is that we need to specify cpuids we want to assign, so it's 
> not easy to have a good auto loadbalacing with multiple containers) 

Yes, I do not want to do that. I think it does not really makes sense to 
restrict cpuset for 
containers if we can restrict overall cpu usage using cpulimit?