On Wed, Sep 17, 2008 at 10:55 AM, Bob Friesenhahn <
[EMAIL PROTECTED]> wrote:

> Each write to a vdev causes a write to all the disks in that vdev. This
> means that the write consumes at least one IO from each of the involved
> disks.  That write does not cause any activity in the non-involved vdevs.


So I infer from the previous and next paragraph that a read operation on a
raidz2 vdev incurs a read operation on each disk member of the vdev.

> As a general rule, multi-user performance should improve linearly with the
> number of vdevs since each vdev contributes a disk's share of IOPS.
>  However, in the special case where the vdevs are mirrors, the read
> performance can be doubled since the read may be done independently from
> each side of the mirror, thereby doubling the available read IOPS.


I will also infer that this differs from traditional RAID5/6 setups in that
parity is only read when the array is degraded due to disk failure or
rebuilding an array. A zpool therefore has read operation overhead
associated with what is refered to as checksumming.
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss

Reply via email to