[LMStone]

Originally Posted by SpEnTBoY

RAID5 writes are slow by comparison to other RAID options especially if there are large numbers of small files. For each write the parity calculation chews up time in overall I/O cost. The more you have to write and the smaller the files, the poorer your performance using any RAID that employs a parity operation on store (as part of the write operation)

Regards,
Lonny

If you use fast disks the performance hit for smaller files is mitigated somewhat, especially if the controller has a large cache (battery-backed one hopes) and does efficient write sorting.

For larger files, using a RAID5 or RAID6 array with six or more disks can actually improve performance, since a single large file write is now broken up into several smaller stripes, all of which are written simultaneously.

In any event, mounting /opt/zimbra/backup on a physically separate disk system will help mitigate disk I/O bottlenecks negatively impacting the end-users' experience.

(For that matter, putting amavis's temp files on a separate partition mounted with noatimes will also help improve performance).

Hope that helps,
Mark