[Prism54-devel] [Bug 87] prism54 lost link with high traffic

bugzilla-daemon at mcgrof.com bugzilla-daemon at mcgrof.com
Thu Nov 11 04:09:58 UTC 2004


http://prism54.org/cgi-bin/bugzilla/show_bug.cgi?id=87





------- Additional Comments From hwertz at avalon.net  2004-11-11 04:09 -------
     My WG511 also bails under load.  With what I've done below, I transferred
like 35GB with no problems though.  I was using NFS over TCP with 4KB rsize and
wsize which seemed to help avoid troubles somewhat.. now I'm just using the
default UDP with 32KB NFS packet size and it still works great.  With 70-74dBm
signal, 24Mbit bitrate I can get a full 2.4MB/sec through the link!  I'll get up
to like 4MB/sec if the laptop's right next to the access point.  I'm running a
WRT54G access point.

     1) I built the prism54-1.2 driver off the site, it seems to work better
than the one in 2.6.8.1 (which is supposedly also version 1.2 8-).

     2) I had found that if I generate any broadcast traffic it seems to kick
the card back on when it does hang, so I don't get those 20 second dead times
under load.  Also for me when the card went down *entirely* it appears it was
during these 20 second dead times that the card just didn't come back up.  I
just did arping -q -b -I eth0 1.2.3.4, since that shouldn't really generate
significant traffic (pinging the broadcast address works too but is more chatty.)

     3) Also, I found the (much shorter) hangs I was getting virtually stopped
when I did a iwconfig eth0 retry lifetime 100ms.  Conversely, they get much
worse if I either increase the retry lifetime or any of the retry limits.  This
suggests pretty strongly to me it's a firmware bug being hit like you guys say.

     This doesn't affect it hanging or not, but iwpriv eth0 s_frameburst 1300
 also seems to kick the throughput up nicely for me 8-).  I do have a shell
script that monitors the signal strength and drops or raises bitrate
accordingly, since the firmware likes to drop it off unnecessarily.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.


More information about the Prism54-devel mailing list