On Fri, 9 Aug 2002, Robert Davies wrote:
> On Thursday 08 August 2002 18:48, you wrote:
> > On Wed, 7 Aug 2002, Robert Davies wrote:
>
> > > Even with DMA, the card generally has to generate an Interrupt to say a
> > > packet needs servicing. There are cards which permit polling rather than
> > > interrupts, and a driver was under development to use that feature. If
> > > you're interested a search of Kernel Traffic or the LWN kernel pages,
> > > might show up an explanation how a non-interrupt driven driver would
> > > work.
> >
> > That would be almost stupid for high performance networking ;-)
>
> I'm not sure what you mean there, but it's not stupid, if you know you have
> data streaming in, why not process it directly in busy times, rather than
> wait for an ISR to copy in data and schedule a tasklet to process it?
You're right...under certain circumstances. I don't generally like the
idea of polling based drivers but I accept that sometimes they can be a
good idea when you know you're under heavy load. On my non-heavily-loaded
non-datacenter simple desktop boxen I would rather use interrupts thanks.
;-)
> Yes Linux interrupts are low overhead, but they are still interrupting the
> processor and requiring some processor state to be sved.
Let's be clear, Intel is shit. Certain other CPUs provide different
sets of global registers for each trap level and fast context switching,
which effectively does away with a lot of debate - has anyone tried and
compared such polling drivers on a SPARC based system?
Jon.
--------------------------------------------------------------------
http://www.lug.org.uk http://www.linuxportal.co.uk
http://www.linuxjob.co.uk http://www.linuxshop.co.uk
--------------------------------------------------------------------
This archive was generated by hypermail 2.1.3 : Sat 10 Aug 2002 - 11:59:47 BST