I think I know why Andrea's new elevator code is so much
than 2.2.x's disk sorting. The general scheme of the
to keep low the latency introduced to old requests by new
being added. In doing this, it prevents merging with other
to the same area of the disk. In actuality this "latency
really increasing latency, because a missed merge means it
take two I/O operations to move data which could have been
a single request. Also, since the elevator will move the
towards the end of the I/O queue, it is almost certain that
to other sectors will happen first, thus an unnecessary disk
I have no idea why the elevator make LVM/IDE dbench runs
for him, as it seems quite wrong fundamentally.
Will study this more soon.