Virtual machine performance severely degrades after vMotion to new hosts with Sub-NUMA Clustering
search cancel

Virtual machine performance severely degrades after vMotion to new hosts with Sub-NUMA Clustering

book

Article ID: 443262

calendar_today

Updated On:

Products

VMware vSphere ESXi

Issue/Introduction

  • A virtual machine running memory-intensive workloads, such as Microsoft SQL Server, experiences severe performance degradation after migrating via vMotion from a legacy cluster host to a newly provisioned host.
  • Query execution times may increase exponentially (e.g., from 4 seconds to over 8 minutes).

Environment

VMware ESXi 8.0.3

Cause

The primary performance degradation is caused by Sub-NUMA Clustering (SNC) being enabled by default in the destination host's hardware BIOS. SNC splits a single physical CPU socket into multiple smaller logical NUMA domains.

When a virtual machine is sized larger than these divided logical domains, it is forced to span across multiple NUMA nodes, inducing severe remote memory access latency.

Resolution

  1. Schedule a maintenance window and place the destination ESXi host into Maintenance Mode.

  2. Reboot the host and enter the hardware vendor's UEFI/BIOS configuration menu.

  3. Navigate to the CPU and Memory configuration settings.

  4. Locate the Sub-NUMA Clustering (SNC) option and change the value to Disabled.

  5. Save the configuration changes and reboot the ESXi host.

  6. Exit Maintenance Mode and verify virtual machine performance.

Additional Information

For secondary causes related to performance degradation after migration, verify that the host hardware BIOS power profile is set to Maximum Performance rather than Balanced or Power Saving mode.

Rightsizing virtual machines on ESXi 8.0: vCPU, memory, and CPU topology guidance

Using Virtual NUMA Controls for Optimizing Performance of Compute-intensive Workloads

Powered by Wolken Software