Notes on Designing Large Scale Systems

Key Challenges

• Ensuring system responsiveness under heavy load.
• Handling massive data volumes and concurrent user requests.
• Database as the primary bottleneck.

Techniques to Address Challenges

1. Database Replication
   • Creating multiple copies (replicas) of a database distributed across servers.
   • Ensures high availability and uninterrupted service.
   • Scales read capacity.

Types of Replication

• Leader-Follower Replication

  • One database acts as the leader (master), others as followers (slaves).
  • Write operations directed to the leader; changes propagated to followers.
  • Read operations distributed across leader and followers to enhance capacity.

• Leader-Leader Replication

  • Multiple databases act as leaders, each can accept read and write operations.
  • Potential for conflicts between leaders requires conflict resolution mechanisms.

Replication Methods

• Asynchronous Replication

  • Changes propagated to replicas in the background.
  • Faster but risks temporary data inconsistencies.

• Synchronous Replication

  • Changes committed to both leader and followers simultaneously.
  • Guarantees consistency but may impact write performance.

Conflict Resolution Techniques for Leader-Leader Replication

• Timestamp-Based Resolution

  • Most recent update wins.

• Last Write Wins (LWW)

  • Most recent write regardless of timestamp wins.

• Custom Conflict Resolution

  • Application-specific rules for resolving conflicts.

Database Sharding

• Distributes data across multiple servers for horizontal scaling.
• Splits large tables, locating parts of data in each shard.

Sharding Criteria

• Data can be sharded based on criteria like:
  • Customer IDs: e.g., users 1-1000 in shard 1, 1001-2000 in shard 2.
  • Geographical Region: e.g., US users in one shard.

Shard Keys

• Determine data assignment to shards.
  • Range-Based Sharding: Data partitioned based on ranges of the shard key.
  • Hash-Based Sharding: Uses a hash function applied to the shard key for even data distribution, but less efficient for range queries.

Differences in Sharding Implementation

• SQL Databases: Typically lack out-of-the-box sharding; require custom logic.
• NoSQL Databases: Many (e.g., MongoDB) provide built-in sharding support.

Summary

• Replication: Ensures high availability; scales read capacity.
• Sharding: Enables horizontal scaling; distributes data.
• Common to use both techniques in combination for optimal performance.