[sheepdog-users] qemu-img conversion timing and redundancy level

Liu Yuan namei.unix at gmail.com
Tue Feb 25 10:16:54 CET 2014


On Tue, Feb 25, 2014 at 08:35:01AM +0100, Valerio Pachera wrote:
> Hi,  I just wish to share this data with you.
> 
> I converted a disk on a cluster that has only 2 nodes using -c 2 and -c 2:1.
> Because there are only 2 nodes, -c 2:1 don't generate parity chunks.
> I was expecting -c 2:1 conversion to be faster than -c 2, but it seems
> to be a little bit slower.
> 
> qcow2 -> -c 2:1
> real    1m18.664s
> user    0m1.352s
> sys     0m6.760s
> 
> qcow2 -> -c 2
> real    1m7.526s
> user    0m1.204s
> sys     0m7.176s
> 
> raw -> -c 2:1
> real    1m10.800s
> user    0m1.304s
> sys     0m6.964s
> 
> raw -> -c 2
> real    1m7.893s
> user    0m1.300s
> sys     0m7.196s

This is understandable because erasure code path is a little bit longer than
replication path even though we don't generate parity.

Thanks
Yuan



More information about the sheepdog-users mailing list