Do you have tried to bench with something like fio ?

----- Mail original ----- 

De: "Florent Bautista" <[email protected]> 
À: "Alexandre DERUMIER" <[email protected]> 
Cc: [email protected] 
Envoyé: Vendredi 14 Novembre 2014 12:30:21 
Objet: Re: [ceph-users] RBD read performance in Giant ? 

Hi, 

I don't think this is the problem, because : 
- I use kvm RBD disk for my VM, with cache=none option 
- after upgrade, I didn't restart my VM to see performance drop (so rbd 
cache may not have changed from firefly) 

I put this in my config, restarted VM but nothing changed. 

On 11/14/2014 12:20 PM, Alexandre DERUMIER wrote: 
> can you try to disable rbd cache ? (which is enable by default in giant) 
> 
> [client] 
> rbd_cache = false 
> 
> 
> ----- Mail original ----- 
> 
> De: "Florent Bautista" <[email protected]> 
> À: [email protected] 
> Envoyé: Vendredi 14 Novembre 2014 11:54:19 
> Objet: [ceph-users] RBD read performance in Giant ? 
> 
> Hi all, 
> 
> On a testing cluster, I upgraded from Firefly to Giant. 
> 
> Without changing anything, read performance on a RBD volume (in a VM) has 
> been divided by 4 ! 
> 
> With Firefly, in the VM (/dev/vda is a Virtio RBD device) : 
> 
> hdparm -Tt /dev/vda 
> 
> /dev/vda: 
> Timing cached reads: 20568 MB in 1.99 seconds = 10314.48 MB/sec 
> Timing buffered disk reads: 1190 MB in 3.00 seconds = 396.49 MB/sec 
> 
> 
> With Giant : 
> 
> hdparm -Tt /dev/vda 
> 
> /dev/vda: 
> Timing cached reads: 20340 MB in 1.99 seconds = 10199.48 MB/sec 
> Timing buffered disk reads: 318 MB in 3.00 seconds = 105.85 MB/sec 
> 
> 
> Write performance is the same than before, no change. 
> 
> How can we explain this ? 
> 
> Every system is running Debian Wheezy. 
> 
> Thank you a lot. 
> 
> _______________________________________________ 
> ceph-users mailing list 
> [email protected] 
> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com 
_______________________________________________
ceph-users mailing list
[email protected]
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

Reply via email to