TCP Badchecksum transmits
Cabling, Ethernet, Hardware, Hubs, IPv4, Network testing, Ping, Protocol analysis, Routers, Switches, TCP
We've got a small network of 15 computer with one server over TCP/IP (WinXP Pro and W2003 Server). The physical network consists of one main switch with two switches and one hub connected to it. The problem we're having is that 5 computers on one of the switches suffers from performance degredation and sometimes hangs, when copying large files or when they have much traffic. The switches are new, but the cabling is fairly old (Cat 5). When I did some network tests with pathping the packet loss is zero, but when I captured packets with Ethereal I found out that there's a lot of badchecksum packets transmitted from the computers that I have problem with. The packets consists of many types of badchecksum, for example framesize, authentication and many more. I've tried to replace the link-cable to that switch, but that didn't deminish the badchecksum-count. Do you have any suggestions to what could be the problem?
Software/Hardware used:
Software/Hardware used:
ASKED:
September 13, 2005 10:47 AM
UPDATED:
September 14, 2005 2:53 AM
Answer Wiki:
A likely possibility is a bad card or interface. A jabbering card can mess everyone else up on that switch, (I know the switch is supposed to limit where the frames go, but I have seen this happen). A flakey interface or connection can cause errors for all packets passing thru it.
Can you substitute another switch? This would help isolate the problem to the switch or the clients/cables. Some of the newer cheap switches have a tendency to accumulate errors, even with minimal traffic. We had this with our internet connection. We used a new 5 port switch to link the firewall to the external router. Within a month, our ISP called me to say his router was accumulating errors. It turned out to be the cheap switch.
If the problems are more wide spread you may want to check out the configuration of the topology.
Are the switches and hub managed?
Do the failures have a pattern linked to the topology? For example, if you are having problems with a system on the hub talking to a system on a switch, you may have issues with speed and duplex. Hubs work only at half duplex and most switches want to run at full duplex. Make sure your switch to switch links are set to 100 Mbit full duplex on both ends. Make sure the switch connection to the hub is running at half duplex.
The clients are more likely to connect correctly automatically but check them anyway.
We had some reliability issues on our network and when I tracked it down, I found a cisco switch talking half duplex to a bay switch running full duplex and a cisco running full duplex to hp switches running at half duplex. When I matched everything up instead of letting them "negotiate" the connection, the errors went away.
rt
To see all answers submitted to the Answer Wiki: View Answer History.
I’m inclined to agree with Astronomer. It’s most likely a bad switch interface or network card on the workstation.
Simplest test would be to move the workstation patch cable to another port and see if it makes any difference.
You didn’t say if your switch was manageable or not, but if it is, then look and see what sort of errors (and in which direction) are showing on the port.
The other thing that might be relevant could be other hardware problems. The problem with software-only sniffers (like Ethereal and others) is that the NDIS network driver does not propagate physical layer errors up to where they are visible. If you are suffering things at that level, then it will just flood the network and you’ll never see anything but bad performance.
I also agree about what Astronomer said about matching speed, duplex manually – takes a little more time, but removes SO many problems.
Bob
Thank you for the replies. I’ve been on the same track about the errors. The switches are non-managed, but I can give it a go on the workstations and configure them manually.