We are setup with standalone Dell R420s running ESXi 5.1 799733 (free license). We have two onboard Broadcom 5720 NICs and a Broadcom PCI card with two 5720 NICs. NetQueue is disabled:
esxcli system settings kernel list | grep -i netqueue
netNetqueueEnabled Bool Enable/Disable NetQueue support. FALSE FALSE TRUE
We are having sporadic issues with these systems. All host CPUs climb to 100% usage and response times on the VMs are extremely slow. Our "fix" to the issue is to move our standby network adapter to active which then lowers the CPU back to a normal level. We then move one of the active network adapters back to standby after a few minutes. This first started happening back in April. Since then we have updated the Broadcom drivers to tg3-3.129d.v50.1 which was supposed to address this issue.
This past week this issue has occurred two more times and both times on systems with the new drivers installed. The logs on the host do not show any abnormalities and we cannot duplicate the error to troubleshoot. One of the Linux VMs does have the following error message: e1000: eth0: e1000_clean_tx_irq: Detected Tx Unit Hang, but I'm still leaning on this being a host issue of some sort.
This appears to be something on the host itself, the performance monitor shows very little traffic on the active adapter and the networking team doesn't see anything on their end. I have used iperf to force traffic between two Linux VMs with no issues. Dell does not see any issues on the hardware side. I do see Broadcom has a newer driver from May, tg3-3.129d.v50.1, but I don’t feel confident that this is the solution after updating these drivers once.
Any help/suggestions would be appreciated.