WDS application notes

From: Jerritt Collord (collord_at_collord.net)
Date: 2002-04-07 20:01:19 UTC

A few PersonalTelco.net folks sat down yesterday (afternoon to be more precise, post-patch) with the new driver and tested WDS. A great feature! Here's some notes--mostly simple and imprecise testing.

The architecture we played with mainly was like this:

NET ---   box0    hostap       sta0

WDS| sta1 | sta2
------------ sta3
| box1 hostap

Each normal wlan0 interface was given private IPs, the WDS link between box0 and box1 was configured manually, each wlan0-apX interface was given private IPs and proper routing was established manually. No bridging was attempted at this time. staX were of a variety of different cards, drivers and OSs.

Box1 had no problem having multiple STAs associated and balanced bandwidth to each as expected. Numbers we saw were (tested on a 3M file with wget/http):
  one client getting ~60 KB/s transfer
  two clients each getting ~24 KB/s TCP transfer

With pretty good results, we reduced the beacon frame interval on both boxX from the default of ten per second to one per second:   one client getting ~70 KB/s TCP transfer   two clients each getting ~35 KB/s transfer   three clients each getting ~25 KB/s transfer

It seems the WDS backchannel is favored over client links somehow. (Now running the iperf tool. Also please note KB above vs. Kb.) An iperf server was established on sta3.

   from a machine on NET above, a UDP bitrate of ~ 1Mb/s    from sta2, a UDP bitrate of ~100Kb/s
   similar ratios for TCP tests, similar also with changes in the client/server roles of the relevant boxen

Changing the ESSID of wlan0 in box0 resulted in box1 spewing "(not our SSID)" messages to console... probably some devel-time debugging left in? Changing other_ap_policy to 2 stopped this.

We were unable to get autom_ap_wds functionality to display signs of working... deleteing manually the connection to that MAC and then enabling the discovery didn't seem to do much.

The long interface names are a problem for some utilities we noticed... mostly that in displaying the device name some things assume a smaller width and truncate it... perhaps s/wlan0/w0/ ?



This archive was generated by hypermail 2.1.4.