Trunked VLAN problem in ESXi 3.5

roadgeek · ‎04-02-2010

Hi,

I've just recently upgraded from VMWare Server 2 to ESXi 3.5. The conversion process has been pretty painless, except for one part: my network setup. In my lab, I have my internal network (no VLAN) and my storage VLAN (vlan 5) physically connected to vmnic0. esxcfg-vswitch reads as follows:

esxcfg-vswitch -l

Switch Name Num Ports Used Ports Configured Ports MTU Uplinks

vSwitch0 64 4 64 1500 vmnic0

PortGroup Name VLAN ID Used Ports Uplinks

172.16.5.0-vlan5 5 0 vmnic0

172.16.100.0 0 0 vmnic0

VM Network 0 0 vmnic0

Management Network 0 1 vmnic0

Hosts connected to the 172.16.100.0 port group work great. Hosts connected to 172.16.5.0-vlan5 work, except for when they receive a large packet; any "full size" packet is sent and hits the ESXi server, but never makes it to the guest. Here are the ethernet headers of one such packet:

20:05:07.152191 00:01:02:3c:0b:43 > 00:50:bf:00:00:41, ethertype 802.1Q (0x8100), length 1518: vlan 5, p 0, ethertype IPv4,

172.16.5.4.http > 172.16.5.1.63171: . 1:1449(1448) ack 168 win 429 <nop,nop,timestamp 1289495062 3072115662>

172.16.5.4 is a HTTP server that is NOT a virtual machine; 172.16.5.1 is my client VM, running Centos5.

On my Centos5 client, my ethernet interface's MTU is set to 1500. When I go and change it to 1496, my connectivity issues disappear.

My best guess is that the vSwitch is dropping the packet because it thinks it is too large, and when I decrease the MTU of my interface by four bytes (the size of the 802.1Q tag) I effectively work around the issue. However:

1) All of my other physical hosts on this VLAN have MTUs of 1500 and work without problems, and;

2) This exact same setup worked fine with VMWare Server 2

I've done a lot of Googling on this and haven't come up with much, so my question is what am I doing wrong or not understanding correctly? Is my only option to either drop the MTU on all of my VLAN 5 connected systems, or drop the MTU of the vSwitch to 1496?

Thanks for your help!

Jackobli · ‎04-03-2010

What an exciting question, I hope someone other can help. I really would like to read the solution.

In the meantime I got some additional questions:

- Do you really use ESXi 3.5 (and which update, 5?) or ESXi 4.0?

- How did you convert the guests? Standalone converter? Downgrade the virtual HW?

- What kind of vNIC did you (or the conversion process) set for the guest (CentOS)? e1000?

- Did you update the vmware tools to the ESXi tools version?

roadgeek · ‎04-04-2010

I had to use ESX 3.5U5 since my system only has 100Mb interfaces. I ordered a pair of GigE cards but they have not arrived yet. When they do, I'll try this again with 4.0.

- How did you convert the guests? Standalone converter? Downgrade the virtual HW?

I used the VMWare Converter. Pretty painless; point it at the source .VMX file, and it take care of the conversion and uploading it to the new server.

- What kind of vNIC did you (or the conversion process) set for the guest (CentOS)? e1000?

I'm using the standard pcnet32 adapter, once I hit this problem I haven't played much with the guest configuration.

- Did you update the vmware tools to the ESXi tools version?

I don't use VMWare Tools on these systems.

Since posting my original question, I actually managed to increase the MTU of my vSwitch and interface to 1504, interestingly enough. 1504 was the max size; I couldn't increase it beyond that. However, I'm still seeing the exact same problems. My only other thought is that perhaps the TCP checksum offloading is getting in the way? Below is the output of my configuration. Thanks for any help.

esxcfg-nics -l
Name PCI Driver Link Speed Duplex MTU Description
vmnic1 03:06.00 e100 Up 100Mbps Full 1500 Intel Corporation 82557/8/9
vmnic0 03:05.00 e100 Up 100Mbps Full 1504 Intel Corporation 82557/8/9

esxcfg-vmknic -l
Interface Port Group IP Address Netmask Broadcast MAC Address MTU TSO MSS Enabled
vmk0 Management Network 172.16.100.15 255.255.255.0 172.16.100.255 00:50:56:76:8c:f9 1504 40960 true

esxcfg-vswitch -l
Switch Name Num Ports Used Ports Configured Ports MTU Uplinks
vSwitch0 64 7 64 1504 vmnic0

PortGroup Name VLAN ID Used Ports Uplinks
172.16.100.0 0 0 vmnic0
VM Network 0 0 vmnic0
172.16.5.0-vlan5 5 3 vmnic0
Management Network 0 1 vmnic0
- Do you really use ESXi 3.5 (and which update, 5?) or ESXi 4.0?

roadgeek · ‎04-20-2010

So to follow up on this problem: I ended up purchasing a pair of Intel Pro Desktop 10/100/1000 PCI network adapters to replace the old interface cards. The old cards were listed as "supported" for ESXi 3.5, all of my research indicates that several others had similar problems that they were never able to resolve.

Once I installed the new cards, I upgraded the server to ESXi 4.0 and have been enjoying the new platform ever since without any type of networking problems.

So the resolution to this issue is: ditch the ethernet pro 100 cards and upgrade to new, supported hardware.