[Prism54-users] Re: [Prism54-devel] Prism54 development update

[Denis Vlasenko]

On Monday 05 September 2005 00:03, Maximilian Engelhardt wrote:
> On Mi, 2005-08-31 at 16:32 +0300, Denis Vlasenko wrote:
> > This could (should) be achieved with...
> > 
> > > - being more resistant to possible future protocol and hardware changes - if 
> > > Prism54 didn't rely on FullMAC that much, support for the 3886 and 3887 would 
> > > have been much more easy (probably as simple as cutting down the firmware 
> > > code and offloading operations to the host).
> >            ^^^^^^^^^^^^^^^^^^^^^
> > exactly this!
> > 
> > Actually, what is the minimum needed from a wireless card?
> > 
> > * ability to tx an arbitrary packet at given rate/modulation
> >   (automatic retry and/or rate fallback is nice but not 100% reqd)
> > * ability to autonomously tx very low level control stuff
> >   like sending ACKs (it's time critical)
> > * tune into given channel
> > * rx packets
> > * DMA packets to/from host via PCI / transfer data over USB / whatever
> > * anything else?
> > 
> > Beacons, association, even WEP and scanning - all can be done by host OS.
> > This also will give us nice well debugged common 802.11 stack.
> > 
> > Thus bare minimum firmware can be rather dumb (and thus easier to write/debug).
> > Maybe we (wireless crowd, not just Prism54 project) should aim
> > exactly to such "bare minimum fw" approach?
> 
> I disagree with this. I think the firmware should to as much as possible
> to avoid CPU load. On the one hand this gives greater performance, on
> the other hand it can perhaps save some power on laptops. And if we have
> a fully open source firmware we will be able to implement everything we
> want.

I agree that WEP/AES in firmware for bulk data may make some difference
and may make sense, other things would not.

Do you seriously expect using CPU for sending 10 beacons per second
will even be visible in you favorite system load meter?
Maybe on 100MHz PentiumI it (barely) will be, but on anything
more sane it would not.

After all, people had difficulties detecting impact of HZ=1000
versus HZ=100, and that's nearly thousand additional interrupts
per second. Here we talk about tens of interrupts per sec.
--
vda