backend

CRDTs vs OTs: Real-Time Collaboration Algorithms

February 7, 2026

CRDTs vs OTs: Real-Time Collaboration Algorithms

What Are They?

CRDT (Conflict-free Replicated Data Type) and OT (Operational Transformation) are algorithms for building real-time collaborative applications where multiple users edit the same data simultaneously.

• OT: Transform operations based on concurrent changes (Google Docs approach) • CRDT: Data structures that automatically merge without conflicts (Figma approach)

The Problem They Solve

When multiple users edit the same document simultaneously:

  1. User A types "Hello" at position 0
  2. User B types "World" at position 0
  3. Both send their changes at the same time

Without coordination: Data corruption, lost edits, inconsistent states

With OT/CRDT: All users see the same final result

Operational Transformation (OT)

How It Works

OT transforms operations against each other to preserve intent:

Initial: "ABC"

User A: Insert "X" at position 1 → "AXBC"
User B: Insert "Y" at position 2 → "ABXC"

Without OT: Conflicts and data loss
With OT: Transform B's operation → Insert "Y" at position 3
Result: "AXYBC" (both edits preserved)

OT Challenges

  • Complexity: N operations require O(N²) transformations
  • Central server: Typically needs a server to order operations
  • Edge cases: Many subtle bugs in transformation functions

CRDTs (Conflict-free Replicated Data Types)

How It Works

CRDTs use special data structures where all operations are commutative—order doesn't matter:

Key insight: Every character gets a unique, ordered ID

User A: Insert "X" with ID (A, 1) after position 0
User B: Insert "Y" with ID (B, 1) after position 0

Both merge: IDs determine final order
Result: Consistent on all replicas

Types of CRDTs

| Type | Description | Example | | ---------------- | --------------------------- | ------------------ | | G-Counter | Grow-only counter | View counts, likes | | PN-Counter | Increment/decrement counter | Inventory | | G-Set | Grow-only set | Add tags | | OR-Set | Observed-remove set | Shopping cart | | LWW-Register | Last-writer-wins | User profile | | RGA/YATA | Sequence/text | Collaborative text |

Comparison: CRDTs vs OTs

| Aspect | OT | CRDT | | ---------------- | --------------------------- | ---------------------- | | Architecture | Usually centralized | Peer-to-peer friendly | | Complexity | Transform functions complex | Data structure complex | | Ordering | Requires operation ordering | Order-independent | | Memory | Lower (ops discarded) | Higher (tombstones) | | Latency | May need server round-trip | Local-first possible | | Offline | Harder to support | Natural support | | Used By | Google Docs, Etherpad | Figma, Yjs, Automerge |

When to Use What

Choose OT When:

  • ✅ Centralized server architecture
  • ✅ Memory is a concern (no tombstones)
  • ✅ Simpler data structures (mostly text)
  • ✅ Always-online assumption

Choose CRDTs When:

  • ✅ Peer-to-peer or offline-first apps
  • ✅ Complex data types needed
  • ✅ Decentralized architecture
  • ✅ Eventually consistent is acceptable

Real-World Implementations

| Product | Algorithm | Notes | | --------------- | --------- | -------------------------------- | | Google Docs | OT | Server-based, optimized for text | | Figma | CRDT | Custom implementation for design | | Notion | OT | Hybrid approach | | Apple Notes | CRDT | Uses CloudKit for sync | | Yjs | CRDT | Popular open-source library | | Automerge | CRDT | JSON-like CRDT library |

Interview Q&A

Q: Explain the difference between CRDTs and OT in simple terms.

A:

  • OT: When conflicts happen, transform operations to make them compatible. Like a traffic controller merging lanes.
  • CRDT: Design data structures so conflicts can't happen. Like giving everyone their own lane.

Q: What's the main advantage of CRDTs over OT?

A: CRDTs work offline and peer-to-peer without a central server. Operations can be applied in any order and still converge to the same state. OT typically requires a central server to order operations.

Q: What's the main disadvantage of CRDTs?

A: Memory overhead. CRDTs often use tombstones (marking deleted items instead of removing them) to handle deletions, which can grow unbounded. Garbage collection is complex.

Q: How would you implement a collaborative counter?

A: Use a G-Counter (grow-only) or PN-Counter CRDT:

  • Each node maintains its own count
  • Merge by taking max for each node
  • Sum all nodes for total value

Q: Why is OT considered harder to implement correctly?

A: OT transformation functions have many edge cases. For N operation types, you need N² transformation pairs. Subtle bugs cause divergence. CRDTs avoid this by making merge commutative by design.

Key Takeaways

✅ OT transforms concurrent operations; CRDTs use conflict-free data structures
✅ OT is typically centralized; CRDTs support peer-to-peer
✅ CRDTs have higher memory usage (tombstones)
✅ CRDTs naturally support offline-first
✅ Both achieve eventual consistency
✅ Choose based on architecture needs (centralized vs distributed)