IBM Intel PRO/1000 GT Dual Port Server Adapter problem

[ScottHolman]

I have 4 IBM System x3655 servers that each contain the following NICs

2 x integrated Broadcom

1 x Pro/1000 PT dual port

1 x Pro/1000 GT dual port

All bought from IBM, so are IBM badged, and all hardware is on the HCL.

The Broadcom and PT work flawlessly in each host. However, I'm not able to push any traffic through the either port on the GT on any of the hosts. The hosts are 3.0.1 and fully patched.

Strangely the Intel site has the GT down as a desktop NIC and the IBM site as a server NIC. I guess it's been tweaked somewhat by IBM to improve performance.

Each server's lspci output shows the GT as:

Ethernet controller: Intel Corporation 8254NXX Gigabit Ethernet Controller (rev 03)

The NIC is in the same PCI slot in each server.

Each server has the GT using the e1000 driver version 7.0.33

So, where do I go next? I've tried looking for an IBM NIC driver, but can find one. Also contemplated firmware upgrades.

I'm just about to place a call with IBM, but they are always slow so I thought I'd put this up here, as I've always gotten a better response.

has anyone experienced problems like this with the GT? If so how did you resolve it?

[conyards]

You may have a faulty connection somewhere here, whats the out put from

esxcfg-nics -l

esxcfg-vswitch -l

Do you have any error messages in the Kernel log? I recall an issue with the e1000 that generate TX_HANG errors to the kernel log. patch ESX-1006511 is supposed to address thi issue.

Simon

[ScottHolman]

Thanks Simon, but I think you're on the wrong track there. The output is good from esxcfg-nics & vswitch. Everything as expected. The NICs are recognised, configured, and happily sitting in the same vswitch as the broadcom and the intel pro/1000 PT.

Remember the PT is using the e1000 driver of the same version across the same four hosts as the GTs, but the GT is the one that isn't working on any of the hosts using the same driver are the PT. Plus, all boxes are fully patched.

Just to make I've checked the logs for TX_HANG and e1000 error/warnings and there's nothing.

I've downloaded the latest driver from IBM for this NIC and was considering, as there it contains a Linux driver, to use that for this NIC. The IBM website says that the NIC support VMware, but has no specific driver for it.

Any other ideas guys?

[ScottHolman]

If anyone is interested IBM came back to me saying:

VMware is in the process of putting a notice at the top of their I/O compatibility matrix that states something to the effect that:

"Not all cards will function in every server"

In other words, just because the matrix says that the card will work it doesn't necessarily mean that it will work with all server hardware. We have to fallback to the ServerProven website in this instance and say that these cards are not qualified to work with VMware.

This is the position that System X has also taken for the Intel PRO/1000 Quad port PT cards. If they are not on the ServerProven website, they are not supported by IBM.

==================================================

The ServerProven site for the x3655 is:

http://www-03.ibm.com/servers/eserver/serverproven/compat/us/xseries/x3655.html#LAN

This lists the GT & the PT as not being certified by IBM to work on the x3655 with 3.0.x

The PT is listed as been certified on 2.5.x. However, the PT in my box is working, this may well invalidate my support contract.

The only Intel PRO/1000 that IBM have tested and have proven to work in the x3655 is the PF model, which happens to be a single port adaptor and no use to me. There are also a couple of single port Broadcoms on the list too, but again this is not use to me.

It just goes to prove that you have to be very careful when choosing components for your ESX hosts. Just because it's on the HCL doesn't mean it will definitely work in your server. Always dig a little deeper. Make sure your chosen hardware manufacturer has proved that the components work in that box.