Round-robin read
Round-robin reads alternate between copies.
This results in equal utilization for reads even when there is more than one I/O outstanding.
Which is right for me?
Each of the scheduling policies provide benefits, as well as drawbacks.
When deciding on a method of mirroring, you need to take into consideration how critical the data is, as well as performance.
The trade off is performance versus availability.
In general, a mirrored logical volume is slower than an unmirrored logical volume, because you have to write the data in two or three places.
The exception can be a mirrored LV in a high-read environment.
If your application does mostly reads, and you are using parallel or parallel/round robin scheduling, reads may complete faster
because the I/Os are spread across multiple disks, which can occur simultaneously if the disks are on separate controllers.
One of the parallel scheduling policies usually provides the best performance in a write intensive environment, because writes can proceed in parallel.
However, there is some additional overhead,
and mirrored logical volumes are usually slower than comparable unmirrored logical volumes in a write intensive environment.
Sequential scheduling provides the worst performance, but provides the best chance of
recovering data in the event of a system crash in the middle of a write operation.
Sequential scheduling makes it more likely that you have at least one good copy (the primary copy) of a logical partition after a crash.
This is discussed further in the next visual.
Synchronizing
When turning on mirroring for an existing logical volume, the copies have to be synchronized
so the new copy contains a perfect image of the existing copy at that point in time.
This can be done by using the -k option on the mklvcopy command at the time
mirroring is turned on or with the syncvg command at a later time. Until the copies are synchronized, the new copy is marked stale.