On Tue, 2009-02-03 at 22:34 -0800, Howard Chu wrote:
> Dan Williams wrote:
> > On Wed, 2009-01-14 at 12:51 -0800, Howard Chu wrote:
> >> Howard Chu wrote:
> >>> I think you need to acknowledge that single AP deployments are a very 
> >>> common
> >>> occurrence. Yes, there are offices and hotels and other large buildings 
> >>> that
> >>> deploy multiple APs throughout a campus, where roaming is an issue. But 
> >>> you're
> >>> trying to force NM to act as a one-size-fits-all solution and that frankly
> >>> isn't reality.
> >>>
> >> Perhaps a workable approach would be to check for multiple APs with the 
> >> same
> >> SSID at the beginning of an association; if more than one is seen, then 
> >> keep
> >> background scanning enabled...
> >
> > One thought I had to fix your case, where scanning interrupts a stream,
> > is to use a sliding window to look at the number frames
> > transmitted/received, and if there is a significant number of them in
> > the past, oh, 30 seconds, postpone the scan for a bit.  If there's a
> > postponed scan and the number of tx/rx frames drops below a certain
> > number, let the scan proceed.  Thus, as long as your download or VOIP
> > conversation or streaming movie is going, the scan might get postponed,
> > but when it stops, the scan can occur.
> 
> That sounds promising. Is getting the frame count a cheap operation? Then 
> it's 
> just a matter of figuring out what the threshold between "busy streaming" and 
> "not" is.

Not entirely sure how cheap it is; ifconfig reads /proc/net/dev directly
which isn't the best solution.  We might be able to use what "ethtool
-S" does, but some drivers don't support that yet.  There may also be a
netlink interface to do this.

Dan


_______________________________________________
NetworkManager-list mailing list
[email protected]
http://mail.gnome.org/mailman/listinfo/networkmanager-list

Reply via email to