Host PSODs After ESXi Upgrade environment with sysContexts enabled for the VM's
search cancel

Host PSODs After ESXi Upgrade environment with sysContexts enabled for the VM's

book

Article ID: 382347

calendar_today

Updated On:

Products

VMware vSphere ESX 7.x VMware Telco Cloud Infrastructure

Issue/Introduction

The issue stems from specific advanced configurations and NUMA-related settings in the customer's environment following the ESXi upgrade from 7.0U3o to 7.0U3q as part of the TCI 2.2 upgrade. Key contributing factors include:

  1. NUMA Configuration Conflicts: All low-latency VMs are pinned to a single NUMA node, leading to resource contention and inefficiencies in core assignment when multiple VMs are active.
  2. Use of Deprecated sysContexts: Advanced configurations (sched.cpu.latencySensitivity.sysContexts) introduced in earlier vSphere versions are still in use, though the feature has been deprecated and no longer provides any performance benefits.
  3. preferHT Configuration: This setting impacts NUMA client sizing and VM placement, potentially exacerbating issues on ESXi 7.x due to changes in how vLLC (Virtual Last Level Cache) sizing is handled.
  4. NSX/ENS Absence: Lack of NSX-T and ENS capabilities renders certain configurations redundant, including sysContexts.

Environment

2.2 and newer versions

7.0 and newer version

Cause

PSOD on the ESXi host

Resolution

 

  • Immediate Recommendations:

    • Remove sysContexts: If not providing measurable benefits, remove sched.cpu.latencySensitivity.sysContexts from the VM configurations as a simple mitigation.
    • Test Without preferHT: Consider removing the preferHT configuration for low-latency VMs and evaluate the impact on NUMA client sizing and performance.

  • Long-Term Considerations:

    • Verify if NSX-T and ENS can be implemented to modernize the environment and remove reliance on deprecated configurations.

 

Additional Information

Powered by Wolken Software