Designing Data Models in Real Products

Learn how senior engineers translate abstract product requirements into rock-solid, production-grade data structure architectures.

Case Study 1: Real-Time Chat Messenger (WhatsApp / Telegram)

Requirements:

1. Messages must be displayed in strict chronological order.
2. Fast random retrieval of messages by message ID.
3. New incoming messages append to the bottom instantly.
4. Smooth reverse scrolling (infinite pagination into chat history).

Optimal Architecture:

• LinkedHashMap<String, Message>:
  • Hash Table component: O(1) lookup by messageId for updating delivery/read receipts.
  • Doubly Linked List component: Preserves strict chronological arrival sequence.
• SQLite Database with B+ Tree Index on (conversation_id, timestamp):
  • Enables instant range pagination: WHERE conversation_id = ? ORDER BY timestamp DESC LIMIT 50.

Case Study 2: Collaborative Text Editor (Google Docs / Figma)

Requirements:

1. Multiple users edit the same document concurrently.
2. Inserting or deleting characters at position K must not corrupt other users' cursor offsets.

Why Standard Arrays or Strings Fail:

A standard string/array requires O(n) shifts on every keystroke. Across millions of collaborative updates, this freezes the application.

The Solution: Rope / Piece Table / CRDT

• Rope: A binary tree where leaves hold string fragments. Inserting in the middle splits a leaf and reconnects pointers in O(log n) time without copying massive arrays.
• Conflict-Free Replicated Data Types (CRDTs): Node-based graph structures where every character has a unique cryptographic identifier, allowing concurrent edits to merge deterministically.

Case Study 3: Infinite Feed with Paging (Instagram / Twitter)

Requirements:

1. Display feed posts.
2. Deduplicate sponsored ads and organic posts.
3. Cache locally for offline viewing.

Optimal Architecture:

• ArrayDeque<Post> in RAM for the active UI viewport list.
• HashSet<String> for O(1) post ID deduplication.
• Room SQLite Cache with LRU eviction strategy.

Summary

• Production systems rarely rely on a single data structure; they compose complementary structures (e.g., Hash Tables + Doubly Linked Lists).
• Choose data models based on exact I/O patterns: point lookups, ordering, range scans, and concurrency.