Only one server seems to disagree with switch

[dalchri]

Over the weekend, our mysql database server started communicating with the rest of the network very slowly:

LAN IP address: 192.168.0.8
NIC: National Semiconductor DP83820 Copper Gigabit Adapter
Switch: TRENDnet TEG-S160TX

I have tried the following:

1) I swapped cables with a 2nd server on the same switch that is working fine. The problem with the mysql database server persists. This elminates a particular port or cable as the source of the problem.

2) Powered the switch off and back on. The problem persists.

3) Cleaned the switch. The problem persists.

4) Hooked the mysql server up to a different switch. The problem was solved.

While the mysql server was exhibiting the problem, ping did not reveal any significant increase in delay time. No lost packets. No increase in load on the CPU. However, trying to copy a file from the mysql server (through SAMBA) also took an excessively long time. The connection to the switch indicates 1GB/s.

The server is perfectly fine working off a different switch. The troublesome switch has several other servers working perfectly fine, and even off the original port that the mysql server was connected to.

Any ideas other than don't use that server with that switch anymore?

Any logs I can look into to troubleshoot basic network connectivity?

The switch isn't managed. Is there any feedback I can get from the switch?

Thanks in advance for any suggestions for further troubleshooting!

 

[sleipnir214]

When you did you ping testing, did you perform the tests with packets of different sizes? In particular, did ping testing show degraded performance with large packets?

Want the best answers? Ask the best questions!

TANSTAAFL!!

 

[dalchri]

Thank you for the suggestion. I was using the default 32 byte packet size. I will have to try the test again later with 65500 byte packets.

If this does show a degraded time but no dropped packets what should I try next?

If this shows dropped packets what should I try next?

If there is no degradation at all what should I try next?

Thanks for the idea!

 

[dalchri]

Using larger packet sizes does result in timeouts. Some ports of the switch seem to do worse than others. The strange thing is that I only get dropped packets if I start out with a large packet size. If I work my way up in size to determine where the threshold is, I don't seem to encounter the problem.

I should probably point out that the alternate switch I am using that works fine is not a Gigabit switch and only supports connections up to 100 Mb/s.

I suppose my next test will be to swap the NIC with a different server.

 

[kordaff]

Is that a switch with removable blades? At one place I used to work, we would have a blade die every few weeks - seems the consistently high loads they carried caused an overheating issue which seemed to lead to the occasional premature failure of the blades. Lots of switches and lots of blades and we had a regular problem...

Any excessive heat, even just intermittently could cause the port on the switch or the nic in the server to fail

Kordaff <--hates 60 degree server rooms lol - where's my gloves and coat??