This isn't a strictly b43 thing, but since so many of the developers regularly use iperf for testing I thought I'd get some of your expertise on something which has been baffling me for two weeks, and which appears to make no sense.

When testing a metropolitan-Ethernet link the following was repeatably observed. (Repeatably means over several test days with different test laptops and each test repeated 5+ times in either direction and bidirectionaly) This is supposed to be 100Mbps bridged Ethernet so numbers around 90Mbps are fine.

Unidirectionally the links check out fine:
A->B 90Mbps
B->A 90Mbps

Bidirectionally one of them is slow:
A<-=>B (A: iperf -s   B: iperf -c A -d)      A->B 86Mbps    B-A>  55Mbps
B<-=>A (A: iperf -c B -d B: iperf -s) A->B 86Mbps B-A> 55Mbps (yes, that's right, with client/server flipped the B->A side is still the slow one)

Thinking this clearly indicated a lack of bandwidth on the return leg (B->A) we've been working with the carrier. However, today, for the sake of further testing, I moved the testset to the local LAN separated only by gigabit switches and both A and B on 100Mbps FDX ports. I got the same test results.

I then downloaded tcpperf, which claims to be a very simple test used for simulation modeling (perfect!) from http://wand.cs.waikato.ac.nz/~stj2/nsc/software.html and ran it. tcpperf showed 90Mbps A->B, B->A and A<->B

This leaves me with two baffling questions:
1. What am I doing wrong with iperf?
2. Why is it the B->A path is always the one that suffers in the bidirectional test no matter who is acting as client and who is acting as server?

I spent hours yesterday and today googling everything iperf and bidirectional, and then just plain iperf (which is how I found tcpperf) and got no information.
Any ideas?

Thanks in advance,

Ehud
PS iperf -r (instead of -d) performs the same as the unidirectional tests above.

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

_______________________________________________
Bcm43xx-dev mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/bcm43xx-dev

Reply via email to