[Gluster-users] Gluster native client very slow

[Olivier Desport]

Le 06/09/2013 15:22, Olivier Desport a écrit :
>
>
>
>
> Le 06/09/2013 15:15, Vijay Bellur a écrit :
>> On 09/06/2013 05:46 PM, Olivier Desport wrote:
>>> Hello,
>>>
>>> I'm testing a two nodes glusterfs distributed cluster (version 3.3.1-1)
>>> on Debian 7. The two nodes write on the same iscsi volume on a SAN.
>>>
>>> When I try to write an 1G file with dd, I have the following results :
>>>
>>> NFS : 107 Mbytes/s
>>> Gluster client : 8 Mbytes/sec
>>>
>>
>> What is the block size being used with dd? With smaller block sizes, 
>> write throughput on native client would be low. If a bigger block 
>> size like 128 KB is used, the throughput would be better.
>>
>> -Vijay
> The block size is 1k :
>
> dd if=/dev/zero of=/data/cms/500m-client-gluster-2 bs=1k count=500000
>
>

I tried with bs=128k and the write speed is 222 MBytes/s !


>
>>
>>> My /etc/fstab on the client :
>>>
>>> /etc/glusterfs/cms.vol /data/cms glusterfs defaults 0 0
>>>
>>> I'd like to use the gluster protocol because it can give me HA. Can you
>>> help me to increase the perfs ?
>>>
>>> Thanks.
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> Gluster-users mailing list
>>> Gluster-users at gluster.org
>>> http://supercolony.gluster.org/mailman/listinfo/gluster-users
>>>
>>
>
>
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://supercolony.gluster.org/mailman/listinfo/gluster-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20130906/d5cdb4d9/attachment.html>