Virtual machines may show a large number of interrupts pinned at vCPU0 on ESXi 6.x and ESXi 7.0 (prior to Update 2)

Products

• VMware vSphere ESXi

Issue/Introduction

Virtual machines may observe a disproportionate number of interrupts focused on vCPU0 (CPU0 within the Guest OS). Additional symptoms are (and not limited to):

• Windows virtual machines produce a Blue Screen (STOP) error
• Linux OS virtual machines may experience performance degradation within the virtual storage stack
• Fail-over solutions within the Operating System may trigger a false fail-over action as interrupts may slow down operations that depend on CPU0 for failover status
• CPU usage is observed to be 100% specifically with vCPU0 leading up to the event

The conditions for this situation are more frequent (but not limited to):

1. large vCPU counts exceeding 48 vCPUs on a virtual machine
2. Paravirtual SCSI (PVSCSI) controllers present on the virtual machine
3. VMXNET3 adapters present on the virtual machine
4. Guest OS activity is sensitive to IO workloads where excessive interrupts may interfere
5. For virtual machines running a Windows OS, you may observe a high %DPC (Processor Information\% DPC Time) within Performance Monitor.

Environment

VMware vSphere 6.x
VMware vSphere 7.0.x

Resolution

VMware recommends to upgrade to at least vSphere ESXi 7.0 Update 2 to allow enhanced performance improvements, as well as code fixes that address the above behaviour. There is no fix available in vSphere 6.x.

For virtual machines with a VMXNET3 adapter allocated, it is recommended to enable RSS per KB article: Poor network performance or high network latency on Windows virtual machines (2008925)