[pve-devel] backup ceph high iops and slow

[Dmitry Petuhov]

I think that skipping free space isn't main issue: backup of used space 
is even slower.

17.10.2014 10:31, Alexandre DERUMIER пишет:
> I have talked with ceph and qemu dev about this. (mainly from drive-mirror, but it's almost the same).
>
> It's missing some feature in rbd block driver to skip zero block on read.
>
> for restore no problem.
>
> For drive-mirror write to target, it's also missing a feature to skip zero-block write.
>
>
> ----- Mail original -----
>
> De: "Dmitry Petuhov" <mityapetuhov at gmail.com>
> À: pve-devel at pve.proxmox.com
> Envoyé: Vendredi 17 Octobre 2014 08:04:04
> Objet: Re: [pve-devel] backup ceph high iops and slow
>
> 16.10.2014 22:33, VELARTIS Philipp Dürhammer пишет:
>> Why do backups with ceph make so high iops?
>> I get around 600 iops for 40mb/sec which is by the way very slow for a backup.
>> When I make a disk clone from local to ceph I get 120mb/sec (which is the network limit from the old proxmox nodes) and only around 100-120 iops which is the normal for a seq read with 120mb/sec...
>> Is the vzdump making problems?
> It's not vzdump's issue, but qemu's. See
> http://forum.proxmox.com/threads/19341-Ceph-backup
>


-- 
С уважением,
   Дмитрий Петухов
   Ведущий системный администратор
   Отдел информационных технологий
   ЗАО "Электро-ком"
   phone: +79209755610
   email: d.petuhov at electro-com.ru