Odd behavior...
This is all using iperf 2.0.2 on Linux 2.4.20 (client side in the trouble
case), and 2.6.16 (server side in the trouble case).
I am trying to troubleshoot a link which is showing VERY directional
differences. When I run the client on one side we get a very consistent
10.4 Mbs per stream reported at the server, and a more variable and slightly
higher value at the client. If I run 50 streams, I get over 500 Mbs. When
I reverse the direction, I get 340Mbs with one stream.
So, I have two questions:
1) The real iperf related question is this: When I am seeing the limit
(10.4 Mbs), I am seeing significant differences in what the client reports
and what the server reports. They report the same number of bytes, but when
I ask for a 30 second test, the server will report something like 38.5
seconds and the very consistent 10.4, where the client shows some
variability and might report 32.2 seconds, and 12.5 Mbs. Why the difference
between the client and server? How can the client finish sooner than the
server? Should that be telling me something?
2) Has anyone ever seen something like that before? We have done due
diligence on setting the window (-w 8M) and verifying that Linux is
configured to allow that. Our networking people assure me there is no
traffic shaping being done in the network.
Thanks,
Bill
---------------------------------------------------------------
William E. Allcock
Argonne National Laboratory
Bldg 221, Office B-139
9700 South Cass Ave
Argonne, IL 60439-4844
Email: allcock --at-- mcs.anl.gov
Office Phone: +1-630-252-7573
Office Fax: +1-630-252-1997
Cell Phone: +1-630-854-2842