I recently deployed the new Cisco UCS platform emulator (UCSPE) 2.1 to get a first impression of the changes in the new UCSM 2.1 software. I ran it without problems in VMware Fusion on my MacBook Pro, but couldn’t get it to run properly on the ESXi 5.0 host in my home lab. Somehow, the network connectivity seemed to be broken and I kept getting the following error message on the console: “
eth2: Bus master arbitration failure, status 9cf3“. Since I had had some power problems in my home lab over the weekend, I first thought that something might have gone wrong either with my host, or with my storage. However, after re-downloading the ova from Cisco and redeploying it I still had the same issue. A bit of googling turned up the following post:
This forum thread had nothing to do with UCSPE, but with a Veeam virtual appliance running on ESXi. One of the messages in the thread suggested that the problem might have something to do with the adapter type “Flexible”. The proposed fix is to replace the adapters with adapters of type “E1000”.
So I figured that I’d give this a try, because my UCSPE was broken anyway, and I could always deploy a fresh copy from the OVA template. So I removed the three network adapters from the VM and created three new ones of type e1000. After this operation I booted UCSPE again and I could connect to it!
I still have the feeling that what I did was just a workaround and didn’t really address the root cause. It may be time to apply some patches to my ESXi host. However, for now I am happy, because I can start playing with UCSPE and UCS Central in my lab!
Update: After patching my ESXi host to ESXi 5.0 update 2 the original problem went away. I deployed a new UCSPE and network connectivity worked without having to change the adapter type from “flexible” to “e1000”. It looks like I was hitting a bug after all.