Over the past couple of weeks I have been upgrading my home network to
CAT6/gigabit (I now understand why no one likes to pull cable). Almost
everything is working as expected with one exception. iperf reports
significantly different speeds between my server (mutilate) and a
workstation (bend) on the network depending on which one is the iperf
server and which is the client.
The server is running CentOS 4.3 and the workstation is running FC4.
Both systems have all current updates applied. The network cards are
3com 3c2000-t's and they connect through a 3com OfficeConnect gigabit
switch. The switch says the connections are gigabit and ethtool says
the same on the workstation (ethtool doesn't like the sk98lin module on
the CentOS box).
If I make the workstation, bend, the iperf server I see pretty much what
I expect:
[dave@mutilate ~]# iperf -c bend
------------------------------------------------------------
Client connecting to bend, TCP port 5001
TCP window size: 27.5 KByte (default)
------------------------------------------------------------
[ 3] local 192.168.255.254 port 33315 connected with 192.168.0.187 port
5001
[ 3] 0.0-10.0 sec 814 MBytes 683 Mbits/sec
However, if I reverse the roles and make bend the client and mutilate
the server I see:
[dave@bend ~]# iperf -c mutilate
------------------------------------------------------------
Client connecting to mutilate, TCP port 5001
TCP window size: 27.4 KByte (default)
------------------------------------------------------------
[ 3] local 192.168.0.187 port 39728 connected with 192.168.255.254 port
5001
[ 3] 0.0-10.0 sec 232 MBytes 194 Mbits/sec
If I go to an old PIII/733 box I put on the network for this testing I see:
[dave@spindle ~]# iperf -c mutilate
------------------------------------------------------------
Client connecting to mutilate, TCP port 5001
TCP window size: 27.5 KByte (default)
------------------------------------------------------------
[ 3] local 192.168.0.188 port 33183 connected with 192.168.255.254 port
5001
[ 3] 0.0-10.0 sec 490 MBytes 411 Mbits/sec
somewhat slower but not bad given the old hardware.
Anyone have a reason why iperf says the connection slows down when bend
is the server but not when bend is the client? Could a bad cable cause
this or is it the skge module on bend (spindle and mutilate both run
CentOS 4.3 with the sk98lin module and iperf says the speed is about the
same in both directions between them)? A bad cable seems odd given the
speed is fine in one direction.
Thanks,
Dave
--
Politics, n. Strife of interests masquerading as a contest of principles.
-- Ambrose Bierce
