Issues with Intel AF DA 10G adapter

kurtbunker · ‎03-09-2010

Anyone out there using Dell R710 servers with Intel AF DA dual port SFP+ cards on ESX 4 U1? I am having several different types of problems on this hardware with Nexus 5000/2000 series switching. The problem has also appeared on R900 servers recently with the same Intel adapters. I have seen several updates for issues with the Intel ixgbe driver and recently I see that VMware moved the supported driver to 2.0.44.14.4 on ESX 4 Update 1. The AF DA cards have some Cold boot issues and informal word from Intel is there may be a recall, but I still don't have details on the depth of the recall.

The Problem is wide spread from a certain VMNIC in a team losing Cisco CDP capability and VLANs to a flapping NIC going up and down dozens of times a day. All point back to the adapter or its driver. My gut told me to take the Cisco TAC route due to the strong VMware team they have, but they were not very helpful beyond verifying the Switch was OK. I am open to any suggestions and have been looking deeper for things like VMDQ/netqueue configuration.

I can go much deeper into details in areas like BIOS updates, patch levels for ESX etc. if anyone has a similar environment that is sucessfully running teamed 10G AF DA NICs.

Thanks in advance.

Kurt

thing2 · ‎03-31-2010

Hi,

Same hardware as us, exact same problem

😕

regards

Steven

kurtbunker · ‎03-31-2010

I believe we found either the solution or a partial one to our issue. I worked with some Intel engineers that responded to a post I put out the my local VMUG and they said that they are using roughly the same setup and you have to use IP hashing or the ixgbe driver has poor performance and potential instabilty. The switches are Nexus so we setup VPC (Virtual Port Channeling) on one of my clients environments and it seems OK so far. we are going to spawn dozens of clones and try to push it a little harder this week. More when I have a definitive answer.

Kurt

kurtbunker · ‎03-31-2010

Here is another thread where the Intel Engineers got involved....

http://communities.vmware.com/message/1490242#1490242