Interpretation of trace results between client-server

[news2010a]

Client/server environment. Server is connected to a Cisco hub. I took traces and I see about 150 packets retransmitted by the server side. Client displays 14 packets retransmitted. Performance is lower than expected. The server network interface card is logging Tx errors.

Questions:

Does the retransmission could be generated by a busy hub?

I know that errors in interfaces can be caused by a faulty port, bad cable or speed/duplex mismatch.

Someone claimed that a busy, congested hub could be capable of generating such errors and retransmissions. Can someone confirm that claim?

[paolo bevilacqua]

Hi,

yes to all questions.

[sundar.palaniappan]

The tx errors you are seeing on the client/server may be a result of high number of collisions on the LAN. That's definitely possible with a hub in place as it would generate large amount of broadcasts during busy traffic periods. Try replacing the hub with a switch to break the collision domain and the errors might cease.

HTH

Sundar

[news2010a]

Positive. The errors went away once I put the switch in place.

Note that I set the switch port to 100Mbs full-duplex and the server side NIC to 100Mbs full-duplex.

Out of pure curiosity, is there any way I that could determine whether the errors were generated just because the hub was busy, or perhaps there was a negotiation issue? Very curious about that...

[sundar.palaniappan]

The hub must have been busy at that time. But, I just want to emphasize that wasn't the reason why you saw those errors. It's due to the fact the hub indiscriminately sends out the broadcast out every port and that would have resulted in more collisions of ethernet frames due to the half-duplex nature of the ports.

HTH

Sundar