In IBM PowerVM on systems based on the IBM POWER architecture, capped vs uncapped partitions define how an LPAR behaves when it comes to using CPU beyond its guaranteed entitlement in a shared processor environment.

🔹 1. Key Concept: Entitled Capacity

Every micro-partitioned LPAR has:

• Entitled Capacity (EC) → guaranteed minimum CPU (e.g., 0.5 cores)

The difference between capped and uncapped is:
👉 Can the LPAR exceed this entitlement or not?

🔹 2. Capped Partition

🧠 Definition

A capped LPAR:

• Cannot use more CPU than its entitlement, even if idle CPU is available

⚙️ Behavior

• Strict upper limit = entitlement
• Hypervisor enforces a hard cap

📊 Example

• Entitlement = 1.0 core
• System has free CPU

👉 LPAR still uses only 1.0 core max

✅ Advantages

• Predictable performance
• Strong resource isolation
• Prevents noisy neighbor issues

❌ Disadvantages

• Wasted CPU capacity if system is underutilized
• No burst capability

🔹 3. Uncapped Partition

🧠 Definition

An uncapped LPAR:

• Can use more CPU than its entitlement if spare capacity exists

⚙️ Behavior

• Guaranteed minimum = entitlement
• Can burst above EC
• Controlled by weight (priority)

📊 Example

• Entitlement = 1.0 core
• System has free CPU

👉 LPAR may use:

• 1.5, 2.0, or more cores (depending on availability)

⚖️ Role of Weight

• Determines share of extra CPU
• Higher weight → more CPU during contention

✅ Advantages

• High utilization
• Better performance for bursty workloads
• Efficient consolidation

❌ Disadvantages

• Less predictable performance
• Possible contention

🔹 4. How Hypervisor Handles Both

Using the POWER hypervisor (PHYP):

Step 1: Guarantee phase

• All LPARs get their entitlement

Step 2: Sharing phase

• Remaining CPU distributed only to uncapped LPARs

Step 3: Enforcement

• Capped LPARs are strictly limited
• Uncapped LPARs compete based on weight

🔹 5. Comparison Table

🔹 6. Workload Suitability

✅ Capped Partitions

• Financial systems
• Real-time OLTP
• Strict SLA environments

✅ Uncapped Partitions

• Application servers
• Dev/test environments
• Analytics workloads

🔹 7. Interaction with Hardware

On processors like the IBM POWER10 processor:

• SMT threads enable fine-grained sharing
• Fast scheduling allows quick CPU redistribution

🔹 8. Key Insight

Capped partitions prioritize predictability, while uncapped partitions prioritize efficiency and performance scaling.

🔑 Summary

🧠 Bottom Line

In IBM PowerVM:

• Capped LPARs ensure strict CPU limits and stable performance
• Uncapped LPARs maximize resource usage by leveraging idle CPU

👉 Most environments use a mix:

• Capped for critical workloads
• Uncapped for scalable or non-critical workload