1. When the scan list contains 6 or more assess points, the scan list is not fully developed on the first scan after a cold boot. It usually takes 2 scans, sometimes 3, to fully develop a long scan list. The scans are fast, taking only about 0.4 seconds. An early version of NM, fired off two quick scans before trying to make a wireless connection. This no longer appears to be the case which causes some interesting problems when the scan list is long.
2. Wireless drivers typically scan based on one or more of consecutive missed beacons, successive retries, signal to noise ratio. ipw2200 uses only the count of consecutive missed beacons. ipw2200 roaming is in two stages. In stage 1, ipw2200 remains associated, examines the scan list, and tries to find a better assess point. If it finds one, it passes to stage 2 in which it disassociates, and runs the test for a better access point again. If it finds a better access point, it tries to associate with it, other wise it remains in the disassociated state. If the module option associate=0 is set, ipw2200 will not make any further attempts to connect. If the default option associate=1 is set, ipw2200 will begin to scan (about once every 0.4 seconds), and will try to connect to a suitable access point in its scan list. Since this behavior duplicates code in NM, I recommend setting associate=0 so that NM alone generates scan requests. 3. When I walk between my office and the next building, ipw2200 roams and switches access points in less than a second and NM is not even aware of the change. 4. If I walk out of access point range, ipw2200 first roams and then disassociates. Within a few seconds, NM manager will pick up this change, deactivate wireless (which includes releasing the dhcp lease), and will start to search for an access point. If no access point is found within the timeout period, NM will show the no-connection icon. If you walk back into range of an access point, NM will not attempt to re-connect. A forced connection using the NM menu is required (select the wireless network you want to connect to). 5. Without standardized driver reporting of signal to noise ratio, missed beacons, or successive retries, NM is limited in what it can do with respect to roaming or filtering out weak access points from its scan list. 6. The proposed capwap standard will allow an access point with a single bssid, to broadcast a public essid and respond to up to 16 hidden essids. At our campus convention center, we have access points that are broadcasting an unencrypted guest essid and also the campus encrypted, hidden essid, both with a single bssid. To deal better with this scenario, a boolean 'broadcast' key could be added to Gconf. The value should be set within the scan processing function, not later. When NM tries to resolve hidden essids by matching bssids, NM should only look at the networks in Gconf which are hidden (broadcast value false). -- Bill Moss Alumni Distinguished Professor Mathematical Sciences Clemson University _______________________________________________ NetworkManager-list mailing list [email protected] http://mail.gnome.org/mailman/listinfo/networkmanager-list
