Diagnosing packetloss

[sub_mesa]

Dear forum,

I've been trying to diagnose a packetloss problem on one of my FreeBSD servers. Its running 6.2-RELEASE-p7 amd64 and the issue is with the nge0 interface.

nge0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
options=13<RXCSUM,TXCSUM,VLAN_HWTAGGING>
inet X.X.X.X netmask 0xffffff00 broadcast X.X.X.255
inet X.X.X.X netmask 0xffffff80 broadcast X.X.X.255
inet X.X.X.X netmask 0xffffff80 broadcast X.X.X.255
inet X.X.X.X netmask 0xffffff80 broadcast X.X.X.255
ether X:X:X:X:X:X
media: Ethernet autoselect (100baseTX <full-duplex>)
status: active

(some data masked due to privacy reasons)

The first step i tried was disabling hardware checksum offloads, as sometimes these do not work correctly. However, the follow command appears to have no effect at all:

ifconfig nge0 -rxcsum -txcsum

It doesn't yield any error message, but after the command ifconfig still shows rxcsum and txcsum to be enabled. What am i doing wrong here? The interface is using aliases because its hosting several jails. Could this be related to my issue?

The packetloss seems to be about 12%. It prevents me from updating the portstree for example, although SSH does work, with some 4-5 second of pause sometimes because of some lost packets. It appears the size of the packets may be relevant; smaller packets usually get through. But this is just a guess.

Anyone can point me in the right direction for diagnosing this?

[vivek]

Run mtr on your server ip from home or office Linux / BSD workstation. It will point out if you are dropping packets location. Also, make sure port on switch and card are both set at correct speed including cabling.

[Alt]

I think you dont need to touch chksum here. Try to look what says these commands:
netstat -w 1 -I nfe0
netstat -m
And more: is there high load on server? what says `top`?