IBM Support

I/O Performance Degradation using Resource Groups on Power11

Newsletters


Abstract

A performance issue has been identified that can significantly impact I/O throughput for shared processor logical partitions (LPARs) in systems that have utilized user defined resource groups (RGs). In certain configurations, I/O performance may degrade by up to approximately 10x.
This document provides a high-level explanation of the issue, when it occurs, its impact, and available mitigation and resolution steps.

Content

Summary of the Issue

Under specific conditions, delayed dispatching of shared processor partition virtual processors may occur resulting in delayed handling of I/O interrupts.  These interrupts are critical for timely handling of I/O operations.

Conditions That Trigger the Issue

The issue occurs during dynamic processor reassignment between resource groups. Resource groups can only be impacted when shared processor partitions are actively running within the potentially impacted resource group during the following operations:

  • When a user defined resource group is created, the default resource group may be impacted.

  • When processors are added to a user defined resource group, the default resource group may be impacted.

  • When processors are removed from a user defined resource group, the user defined resource group may be impacted.

Impact

  • Significant degradation of I/O performance for shared processor partitions within the impacted resource groups.
  • Performance issues persist until corrective action is taken.
  • The issue is isolated to shared processor partitions in the default and user defined resource group and does not affect dedicated processor partitions.

Recovery Options

If the issue is encountered, the following recovery actions have been validated to restore normal performance:

Option 1: Restart Shared Processor Partitions

  • Power off all shared processor partitions within the impacted resource group.  All the partitions must be in the powered off state before proceeding with partition power on.
  • Power on the shared processor partitions.
  • Result: Performance returns to normal

Option 2: Live Partition Mobility (LPM)

  • Migrate all shared processor partitions in the impacted resource group to another system.  All the partitions must be migrated from the impacted resource group before proceeding to migrate back.
  • Power off all shared processor VIOS partitions within the impacted resource group.  All the VIOS partitions must be in the powered off state before proceeding with VIOS partition power on.
  • Optionally migrate the shared processor partitions back to the original system.
  • Result: Performance returns to normal

Important:
For both options, all shared processor partitions in the impacted resource group must be handled collectively for recovery to be effective.

Recommendations

  • If the system is not currently impacted, avoid creating user defined resource groups or making dynamic resource group changes (adding/removing processors).
  • Plan to apply the upcoming firmware update once available to prevent the issue.  Applying the firmware update will not automatically recover a system that has encountered the issue and one of the recovery options listed must be performed.

Fix Availability

The fix is included in the following firmware releases:

  • FW1110.30 (released 2026/06/26) and later

[{"Type":"MASTER","Line of Business":{"code":"","label":""},"Business Unit":{"code":"BU054","label":"Systems w\/TPS"},"Product":{"code":"SGUS3Y","label":"Power Hypervisor (PHYP)"},"ARM Category":[],"Platform":[{"code":"PF025","label":"Platform Independent"}],"Version":""}]

Document Information

Modified date:
07 July 2026

UID

ibm17274074