In IBM systems (and distributed systems in general), shared disk and shared nothing are two fundamentally different ways of organizing how nodes access data in a cluster.

At a high level:

Shared disk = many systems share the same storage
Shared nothing = each system owns its own storage

1. Shared Disk Architecture

What it means

• Multiple nodes access the same physical storage
• Data is stored centrally (e.g., SAN, shared volumes)

IBM example

• IBM Z Parallel Sysplex
• IBM z/OS with shared datasets / Db2 data sharing
• IBM Power with shared SAN storage (e.g., PowerHA setups)

How it works

• All nodes can read/write the same data
• Coordination is required to avoid conflicts:
  • Global locking
  • Cache coherence
  • Transaction control

Advantages

(A) Strong consistency

• Single copy of data
• Easier to maintain correctness

(B) Simplified data management

• No need for replication logic

(C) Fast failover

• Another node can immediately access same data

Challenges

(A) Coordination overhead

• Locking and synchronization needed

(B) Storage contention

• Many nodes hitting same disks

(C) Scalability limits

• Central storage can become bottleneck

2. Shared Nothing Architecture

What it means

• Each node has its own CPU, memory, and storage
• No shared disk between nodes

IBM context

• IBM Power running distributed databases (e.g., sharded systems)
• Linux clusters on Power
• Big data platforms (Hadoop, Spark)

How it works

• Data is partitioned across nodes
• Each node handles its own subset of data
• Communication happens over network

Advantages

(A) High scalability

• Add more nodes → more storage + compute

(B) No central bottleneck

• No shared disk contention

(C) Fault isolation

• Failure affects only part of data

Challenges

(A) Data distribution complexity

• Requires partitioning/sharding

(B) Consistency challenges

• Needs replication + coordination

(C) More network communication

• Nodes must exchange data frequently

3. Key differences

4. IBM Z vs distributed clusters (important contrast)

IBM Z Parallel Sysplex (Shared Disk)

• Uses Coupling Facility for:
  • Global locking
  • Cache coherence
• Provides:
  • Strong consistency
  • Near-linear scaling (to a point)

Typical distributed systems (Shared Nothing)

• Use:
  • Replication
  • Consensus protocols
• Provide:
  • Massive scalability
  • Eventual or configurable consistency

5. When each is used

Shared Disk is best for:

• Banking systems
• OLTP workloads
• Systems needing strict consistency

Shared Nothing is best for:

• Big data analytics
• Cloud-native systems
• Large-scale distributed processing

6. Hybrid approaches (modern trend)

Many systems combine both:

• Shared disk for critical data
• Shared nothing for scale-out workloads

7. Simple analogy

Shared Disk:

• Multiple people working on the same document in a shared drive
• Need rules to avoid overwriting each other

Shared Nothing:

• Each person has their own document copy
• They share updates when needed

Key takeaway

Shared disk architecture allows multiple IBM systems to access the same data with strong consistency using coordination mechanisms, while shared nothing architecture distributes data across nodes for higher scalability but requires more complex data management and communication.