Trade-offs & Alternatives

📖 3 min read 📄 Part 7 of 10

Design Facebook Newsfeed - Tradeoffs and Alternatives

Core Decisions

1. Feed Ranking: Chronological vs Algorithmic

Chronological (Simple)

Pros: Simple, transparent, real-time Cons: Miss important content, spam visible, poor engagement

Algorithmic (Chosen)

Pros: Personalized, higher engagement, better content quality Cons: Complex, ML infrastructure needed, less transparent

Decision: Algorithmic ranking with ML models

  • Higher user engagement
  • Better content discovery
  • Spam filtering
  • Personalization

2. Fan-out: Push vs Pull vs Hybrid

Push Model

Pros: Fast reads, simple Cons: Slow writes for celebrities, storage overhead

Pull Model

Pros: Fast writes, no storage overhead Cons: Slow reads, complex merge

Hybrid (Chosen)

Pros: Optimized for different user types Cons: Complex implementation

Decision: Hybrid approach

  • <1K friends: Push
  • 1K-100K: Partial push

  • 100K: Pull

3. Database: SQL vs NoSQL

MySQL (Chosen for Users)

Pros: ACID, strong consistency, complex queries Cons: Vertical scaling limits, sharding complex

Cassandra (Chosen for Posts/Feeds)

Pros: Horizontal scalability, high write throughput Cons: Eventual consistency, limited queries

Decision: Polyglot persistence

  • MySQL for users (consistency critical)
  • Cassandra for posts/feeds (high volume)
  • TAO for social graph (graph queries)

4. Caching: Aggressive vs Conservative

Aggressive (Chosen)

Pros: Lower database load, faster response, better UX Cons: Higher cache cost, stale data, complex invalidation

Decision: Aggressive caching (95%+ hit rate)

  • 100TB Memcached cluster
  • Multi-level caching
  • Read-heavy workload (200:1)

5. Real-time Updates: WebSocket vs Polling

WebSocket (Chosen)

Pros: Real-time, efficient, bidirectional Cons: Complex scaling, connection management

Long Polling

Pros: Simple, works everywhere Cons: High latency, resource intensive

Decision: WebSocket with polling fallback

  • Real-time feed updates
  • Live reactions and comments
  • Fallback for restricted networks

Alternative Architectures

Monolith vs Microservices

Microservices (Chosen)

Pros: Independent scaling, technology flexibility, fault isolation Cons: Operational complexity, network latency, distributed debugging

Decision: Microservices (200+ services)

  • Independent scaling
  • Team autonomy
  • Better fault isolation

Self-Hosted vs Cloud

Self-Hosted (Chosen)

Pros: Full control, lower cost at scale, custom optimizations Cons: High capex, operational complexity

Decision: Self-hosted data centers

  • Cost-effective at Facebook's scale
  • Custom hardware optimizations
  • Full control over infrastructure

Performance vs Cost

Replication Factor

High (5 replicas - Chosen)

Pros: High availability, better read performance Cons: 5x storage cost, higher write latency

Decision: 5 replicas for balance

  • 1 master + 5 read replicas
  • High availability (99.99%)
  • Read scaling

Data Retention

Long (5 years - Chosen)

Pros: Complete history, better analytics Cons: High storage cost, slower queries

Decision: Tiered storage

  • Hot: 0-30 days (NVMe SSD)
  • Warm: 30-365 days (SATA SSD)
  • Cold: >365 days (S3 Glacier)

Consistency vs Availability

CP vs AP (CAP Theorem)

Strong Consistency (User data)

  • User authentication
  • Privacy settings
  • Friend relationships

Eventual Consistency (Feed data)

  • Feed delivery (5-10s lag acceptable)
  • Reaction counts
  • Comment counts

Decision: Hybrid based on criticality

  • Strong for critical operations
  • Eventual for feeds and metrics

This analysis provides the foundation for making informed architectural decisions for Facebook's newsfeed system.