What is the impact of uncapped partitions on fairness?

What is the impact of uncapped partitions on fairness?

In IBM PowerVM, uncapped partitions improve utilization and flexibility—but they also make CPU fairness conditional rather than absolute.


🚀 Core Idea

  • Capped LPAR → fixed CPU limit (strict fairness)
  • Uncapped LPAR → can borrow unused CPU from the shared pool

👉 Fairness shifts from “equal limits” → “guaranteed minimum + weighted sharing of excess”


⚙️ How Fairness Works with Uncapped Partitions

1. Entitlement-Based Fairness (Baseline)

  • Every LPAR gets its entitled capacity first

👉 This ensures:

  • No LPAR is starved
  • Minimum fairness is always preserved

2. Weight-Based Fairness (Beyond Entitlement)

  • Extra CPU is distributed using uncapped weights

👉 Example:

  • LPAR A (weight 200)
  • LPAR B (weight 100)

➡️ A gets ~2× more of the spare CPU than B


3. Opportunistic Behavior

  • Uncapped LPARs can temporarily consume:
    • Idle CPU from others

👉 Result:

  • High efficiency
  • But fairness becomes dynamic, not fixed

📊 Impact on Fairness

✅ Positive Effects

Better Resource Utilization

  • Idle CPU doesn’t go to waste
  • Active workloads scale up dynamically

Proportional Fairness

  • CPU distributed according to:
    • Entitlement
    • Weight

⚠️ Negative Effects

Relative Fairness (Not Absolute)

  • LPARs may not get equal CPU even if similar
  • Depends on:
    • Weights
    • Activity of other LPARs

Potential Dominance by High-Weight LPARs

  • High-weight partitions can consume most spare CPU

👉 Lower-weight LPARs:

  • May see reduced performance during contention

Performance Variability (Jitter)

  • CPU availability fluctuates based on:
    • System load
    • Other LPAR activity

👉 Impacts:

  • Latency-sensitive workloads

🔄 Behavior Under Different Conditions

System StateFairness Behavior
Low utilizationVery flexible (everyone can burst)
Moderate loadWeighted sharing of extra CPU
High contentionFairness reverts to entitlement
Extreme loadStrict minimum guarantees only

⚡ Example Scenario

LPAREntitlementWeightDemand
A1 CPU200High
B1 CPU100High
C1 CPU100Low

👉 If C is idle:

  • A gets the largest share of extra CPU
  • B gets some
  • Fairness = proportional, not equal

🔗 Interaction with I/O

  • If high-weight LPAR dominates CPU:
    • Others may experience:
      • Increased I/O latency
  • Especially important for:
    • Virtual I/O Server (VIOS)

👉 VIOS must have:

  • Sufficient entitlement to maintain fairness system-wide

🧠 Key Insight

Uncapped partitions redefine fairness as:

“Guaranteed minimum for all + proportional sharing of what remains”

—not equal distribution.


🎯 When to Use Uncapped Mode

Best for:

  • Mixed workloads
  • Cloud / consolidation environments
  • Applications that can scale with CPU availability

Use Caution for:

  • Real-time systems
  • Strict SLA workloads
  • Latency-sensitive databases

🔑 Final Takeaway

  • Uncapped partitions improve efficiency but make fairness dynamic
  • True fairness depends on:
    • Entitlements (guarantee)
    • Weights (priority)
    • System load (contention level) 
Looking for servers Rental ?

Call Our Expert :


  • (call for rental enquiries)

Email us :