🚀 go-pugleaf

-----BEGIN PGP SIGNED MESSAGE-----

In article <41g14g$efj@bmtlh10.bnr.ca>, Linh Dang <linhd@bnr.ca> wrote:

>No, the idea of time-sharing system is that the OS can STOP non-I/O
>processes quickly enough when I/O arrived. Since I/O incoming
>interval is a LOT longer than the cpu-time-slice, you won't loose I/O
>data.

the CPU should not have to context switch at all to handle incoming
I/O; I/O should be handled in the kernel by an ISR.  ISRs are normally
executed in a system-wide context not associated with any process.

it's the braindead windows mentality that makes processes responsible
for their own I/O.

strictly speaking, any system that allows interrupts is preemptive, in
that an interrupt can preempt the running process in order to service
the interrupt.  i make the distinction between preemptive and
nonpreemptive based on whether an interrupt can cause rescheduling.
in preemptive multitasking, it can; in nonpreemptive multitasking, it
cannot.  very few systems used polled I/O anymore, precisely because
in most cases it is incredibly inefficient, and risks data loss.

standard UNIX may reschedule any process at any time that it is about
to transition to user mode from kernel mode.  this occurs after the
servicing of any interrupt or trap. 



-----BEGIN PGP SIGNATURE-----
Version: 2.6.2

iQCVAwUBMDz6rRbgfSHT/piVAQFzOwQAh75Z7WMxARrqJfVNjpqJS6lGM3Zgmwug
Z74BpCbri3PVHU3/NZpRKNGpCQwq1KhAbjIFpeW/Y2p+RjCqZn6ozsQMoL7ge1YK
O0C4R8A4m0xV80+U1vXwgM0cbej27gdW74Wv2EYk6HdJy1rCHCvcIz97/GtXtd42
RlqCpQoPUsc=
=YTef
-----END PGP SIGNATURE-----