iperf 1.7.0 observation / question (fwd)

[Mitch K. <mi...@nc...>]

>Jeff King
>Wave Three
>435-787-0555 x103 (office)
>
>
>I have a question 
>regarding some iperf 1.7.0 testing I have been doing between 
>two directly connected Windows machines (one Win2K and one 
>WinXP).  They are both connected with 100M full duplex 
>connectivity.  Same result with multiple switches and 
>directly connected with Xover cable.
>
>Test 1) Single UDP stream
>
>Machine 1)
>
>C:\iperf s u i 1 
>
>Machine 2)
>
>C:\iperf c 10.2.50.2 u b 10m
>
>Result:  This single one way UDP stream works fine 
>with either machine in the server or client role; 
>little or no data loss.
>
>Test 2) UDP Test in tradeoff mode
>
>Machine 1)
>
>C:\iperf s u i 1
>
>Machine 2)
>
>C:\iperf c 10.2.50.2 u b 10m -r
>
>Result:  This works fine, little or no data loss.
>
>Test 3) UDP Test in Dualtest mode
>
>Machine 1)
>
>C:\iperf s u i 1
>
>Machine 2)
>
>C:\iperf c 10.2.50.2 u b 10m d 
>
>Result:  This results in significant data loss using 
>default 1470 byte datagram independent of what machine 
>is client or server role. 
>
>Test 4) UDP Test in Dualtest mode with large datagram
>
>Machine 1)
>
>C:\iperf s u i 1  l 10000
>
>Machine 2)
>
>C:\iperf c 10.2.50.2 u b 10m d l 10000
>
>Result:  This results in little or no data loss in 
>either direction.
>
>
>This does not appear to be a speed or duplex problem 
>as both ends are in 100/full mode.  I see a similar 
>problem when increasing the bandwidth to 20m but then 
>I have to increase the datagram size even larger to get 
>low error rate.
>
>Do you have suggestions as to what you think the problem 
>may be as I would think that smaller datagrams would 
>not exhibit such a problem?
>
>Thanks for your feedback.
>
>Regards,
>
>Jeff King
>Wave Three
>435-787-0555 x103 (office)
>
--
Mitch Kutzko | mi...@da... | mi...@nc... | 217-333-1199
http://hobbes.ncsa.uiuc.edu/