Intel NIC seems to corrupt packets

[jarlt]

Hi all,

I tested several ESX installations (4.0 and 4.0 upd 1) and always ran into the same issue with the Intel Pro 1000GT NIC. When using this NIC for the virtual machine (Ubuntu9.10 64 bit - have the same issue with WinXP 32) the large packets from the virual machine to other stations are corrupted. Several packets contain blocks of Zeros at random positions and therfore damage copied files. The packets captured with TCPdump on the Ubuntu show 2.9k packets with all the correct content. Likely the TCP dump captures before packet fragmentation therefore the packets are larger than the MTU of 1500. The capture at a PC directly connected to the Intel NIC of the VMware shows the same packets (but fragmented to 1500) with some parts Zero'ed out. (two screenshots are attached that show the same packet within Ubuntu and on the directly connected PC).

I changed the NIC with another Intel Pro1000GT, re-installed ESX and Ubuntu - without success. The onboard NICof the Asus Mainboard (Nvidia) does not show the same problem. Also installed it on another PC with similar Asus mainboard and Intel NIC - still the problem.

Anyone with a good idea?

Regards

Juergen

[jarlt]

One addition, I started the Win32 in ESX and applied it to the Intel NIC. The filetransfer to a locally attached PC got corrupted again - not with Zeros but with other incorrect content. In the Wireshark of the Virtual Windows machine there were still packets of 2974 (even 32k) bytes even though the MTU of the "VMware Accelerated AMD PCNet Adapter" is set to 1500 byte. Not sure if the Virtual machine is sending with a larger framesize and the packets are then incorrectly fragmented or if the capture driver just sits in a layer that only sees the reassembled packets. For sure I can see that the content of the packtes differ which causes the corruptions.