On Sat, 2014-07-12 at 12:13 +1000, Finn Thain wrote:
> On Fri, 11 Jul 2014, Ben Hutchings wrote:
>
> > I noticed that the default I/O scheduler on m68k is noop, which is
> > really bad for spinning rust. This is because the default must be
> > built-in, and you put these overrides in the m68k config:
> >
> > CONFIG_IOSCHED_DEADLINE=m
> > CONFIG_IOSCHED_CFQ=m
> >
> > I think this is a mistake and that cfq should be reverted to built-in so
> > it can be the default, as on other architectures. Any objection to me
> > doing that?
>
> I think you may be right about the default, though I suspect that emulated
> VMs might do better with the no-op module and rely on the host I/O
> scheduler. (I didn't benchmark anything.)
Maybe.
> For physical MC680x0 hardware with an IDE SSD, perhaps with IDE-SCSI
> converter, you'd probably want to use the no-op module also.
It does not make sense to use the basic 'elevator' algorithm to limit
movement of disk heads on an SSD, but Linux is generally able to
recognise an SSD and disable that. But there is more to I/O scheduling
than the elevator.
Ben.
--
Ben Hutchings
We get into the habit of living before acquiring the habit of thinking.
- Albert Camus
Attachment:
signature.asc
Description: This is a digitally signed message part