Workload balancing in an IBM Parallel Sysplex is achieved through a combination of Workload Manager (WLM), XCF group awareness, shared data via Coupling Facility (CF), and dynamic routing decisions. Unlike typical load balancers that just distribute requests, Sysplex balancing is policy-driven, state-aware, and continuously adaptive at the operating system level.

1. Core idea: goal-based balancing (not simple round-robin)

In a sysplex, workload balancing is:

Goal-oriented distribution of work across multiple z/OS systems based on service-level objectives (SLA), not just equal CPU sharing.

WLM decides:

• Which system should run new work
• When to move or delay work
• How to prioritize competing workloads

2. Main component: Workload Manager (WLM)

WLM is the central intelligence for workload balancing.

It:

• Classifies incoming work (batch, transaction, DB2, CICS, etc.)
• Assigns it a service class (priority + performance goal)
• Monitors system health in real time
• Directs work to the “best” sysplex member

3. Types of workload balancing

A. Sysplex Distributor (TCP/IP level balancing)

Used for network-based workloads.

• Incoming TCP/IP connections arrive at a distributor stack
• XCF + WLM decide target LPAR
• Connection is routed to best system

👉 Used for web, API, and socket-based workloads

B. Transaction-level balancing (CICS, IMS, etc.)

• Workload is routed at transaction start
• WLM chooses best-performing system
• Keeps affinity when needed

C. Batch workload balancing

• JES (Job Entry Subsystem) uses WLM recommendations
• Jobs are routed to less busy systems

4. How WLM makes decisions (core logic)

WLM continuously evaluates:

A. System capacity

• CPU utilization
• I/O load
• Memory pressure

B. Service class performance

• Are SLAs being met?
• Response time vs target

C. Queue depth

• How much work is waiting per system

D. Work importance

• Critical vs non-critical workloads

5. Role of XCF in balancing

XCF provides:

• Membership awareness (which systems are active)
• Cross-system communication
• State sharing between WLM components

👉 Without XCF, WLM cannot coordinate sysplex-wide decisions.

6. Role of Coupling Facility (CF)

CF contributes indirectly by enabling:

• Shared lock management (DB2, CICS)
• Cache coherency (data sharing)
• Global state consistency

👉 This ensures workload can move safely between systems without data corruption.

7. Dynamic work routing process (step-by-step)

Step 1: Work arrives

• Network request or job submission enters sysplex

Step 2: Classification

• WLM assigns service class and goals

Step 3: Sysplex-wide evaluation

WLM checks:

• All system loads
• Response time history
• Resource contention

Step 4: Target selection

WLM selects system with:

• Best ability to meet SLA
• Lowest contention risk
• Proper resource availability

Step 5: Dispatch

• Work is routed to chosen LPAR
• Execution begins locally

Step 6: Continuous adjustment

WLM continuously:

• Monitors performance
• Shifts new work to other systems if needed

8. Key balancing strategies

A. Goal-based scheduling

Instead of equal distribution:

• “90% transactions must finish under 0.5s”
• WLM ensures sysplex meets this target

B. Affinity management

Some workloads must stay on same system:

• Database sessions
• Cached state transactions

WLM respects:

• Soft affinity (preferred system)
• Hard affinity (required system)

C. Dynamic rebalancing

If one system becomes overloaded:

• New work is redirected elsewhere
• Existing work continues locally

9. Feedback loop (critical concept)

Sysplex balancing is a closed-loop control system:

1. Measure performance
2. Compare with goals
3. Adjust routing decisions
4. Repeat continuously

👉 This happens every few seconds in real time.

10. Why sysplex balancing is different from cloud load balancing

11. Simple mental model

Think of sysplex workload balancing as:

A real-time intelligent scheduler that constantly moves incoming work to the best-performing system based on business goals, system health, and shared state visibility across all nodes.

Key takeaway

Workload balancing in Parallel Sysplex occurs through:

• WLM goal-driven decision making
• XCF-based system coordination
• CF-enabled shared state consistency
• Dynamic routing at transaction and workload levels
• Continuous feedback-based adjustment