On Sun, 2011-04-24 at 23:06 +0200, Tomasz Torcz wrote: > On Sun, Apr 24, 2011 at 07:10:48PM +0200, Kevin Kofler wrote: > > Ben Boeckel wrote: > > > One thing I liked a lot with my ifconfig scripts/wpa_supplicant pairing > > > is that when wireless is spotty, the network doesn't keep going up and > > > down. Instead, applications see lots of dropped packets. When > > > reauthentication can take 5 to 10s (or more), assuming that the > > > connection is steady when its just spotty can result in better behavior. > > > Also nice when quickly swapping ethernet cables. A "network is gone" > > > event gets different reactions from applications (particularly those > > > that are NM-aware which makes those applications MUCH more annoying to > > > deal with in these cases) than "some packets were lost". An option to > > > "persist connections despite something probably not actually existing" > > > would be nice for situations like this. > > > > I've found NM to actually be quite tolerant of spotty wireless connections. > > In fact, usually, it's me who triggers a reconnect (or if possible, a > > connect to a different access point, e.g. when I'm at the university in a > > shared building with the business university (WU), I try switching from > > eduroam to eduroam-wu when reception of my university's eduroam is poor), NM > > just happily stays "connected" even with 100% packet loss. > > Well, I have opposite experience with my wired connection. It takes only > about 5 flip-flop (carrier on/carrier off) in 10 seconds for NM to consider > connection down. When carrier state changes happen, NM sets the carrier state internally, but won't do anything about it for 4 seconds. If you get another carrier change within that 4 seconds, NM pushes the action off for another 4 seconds. If you get another, then it pushes it off for another 4 seconds. So basically, whenever the device settles down and stops spamming carrier changes, NM won't do anything for 4 seconds. The next question, what's causing your carrier to flip-flop int he first place? Dan -- devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxxx https://admin.fedoraproject.org/mailman/listinfo/devel
