On Thu, 6 Feb 1997, Boris Boesler wrote:

> Hi!
> 
>  I don't have a SA. On the SA the behavios should be worse if
> my guess is correct.
> 
>  Reformulated guess: During hard disc accesses the ARM is WAITING
> 	in a loop (polling) till the drive has ready.
> 
>  fsck is a good example for this. Timings are 5-10% usr and the rest
> in in system (on a 50MHz 710)
> 
>  I know that the driver for the Connect32 uses polling but I didn't
> know that the performance is that low.

You probably missed a message the other week about the amount of
difference that having cumana and powertec SCSI driver that used
interrupts made ...
Without interrupts the kernel sits in a polling loop after sending a SCSI
command waiting for data. Thus the kernel will be waiting during all drive
seeks etc.
With interrupts the command is posted and the kernel will get on when
running other processes while the driver processes the command and does
any seeks, fills drive caches etc. and thes will interrupt the kernel when
it is ready to start transferring data.

Cheers,
				mark