More mysterious Tx excessive retries
jimc at math.ucla.edu
Wed Apr 24 15:44:34 EST 2002
The continuing saga of the Dell Truemobile 1150 (Agere firmware 6.16)
vs. Linksys WPC11 (Intersil firmware 1.03). This is with driver 0.10.
I've had another thought. Suppose the reason the Linksys card under Linux
cannot talk to the Cisco-Aironet access points is that it never acks
(at MAC level) any packets. Suppose the reason the Dell card thinks it
has to retry so many times is that it never gets the acks. Suppose the
Agere firmware does the obvious thing, and acks as soon as it has the
complete packet on the card with a correct checksum. But the Linksys
firmware doesn't ack until the driver confirms that it has the packet.
The Windows driver would know how to do this, so the Linksys card works
fine under Windows. But the Linux driver doesn't...
In a somewhat ineffectual attempt to shed light on this hypothesis, I did
"ifconfig eth1 ; ping -c 1 fafnir1 ; ifconfig eth1" and similarly on the
partner (Fafnir) except sleeping. I had tcpdump running. Here are the
differences between the before and after ifconfig statistics. The errors
are all TX excessive retries.
Dell: RX packets 1 TX packets 0 TX errors 1 all others 0
Linksys: RX packets 1 TX packets 1 all others 0
Retry after bouncing the Linksys card (it needed an arp)
Dell: RX packets 2 TX packets 0 TX errors 1 all others 0
Linksys: RX packets 2 TX packets 2 all others 0
scp 1.28 Mb after compression; Dell card is sending
Dell: RX packets 544 TX packets 0 TX errors 973
Linksys: RX packets 971 TX packets 546 TX errors 14
Expected number of packets (at 2304 bytes each): 555
My conclusion is that isolated packets are received only once, possibly
because the Linksys card suppresses re-sent packets (that it never
acknowledged) by itself, but massed packets can be delivered as many
as twice, no more. This doesn't tell us a whole lot.
James F. Carter Voice 310 825 2897 FAX 310 206 6673
UCLA-Mathnet; 6115 MSA; 405 Hilgard Ave.; Los Angeles, CA, USA 90095-1555
Email: jimc at math.ucla.edu http://www.math.ucla.edu/~jimc (q.v. for PGP key)
More information about the wireless