Azure Cosmos Database

August 13, 2023 7:00 PM PDT

---

This document outlines the key features and comparisons of Azure Cosmos Database, focusing on its architecture, performance, and use cases.

Presenter: R, Tech Lead

---

System Design Presentation - Azure Cosmos Database

Key Features

• Fully Managed: Azure Cosmos Database is a fully managed database service.
• High Resiliency: Offers 99.999% uptime.
• Low Latency: Achieves less than 10ms read/write latency at P99.
• Elastic Scale:
  • Throughput: Supports from 100 requests per second to over a trillion requests per second.
  • Storage: Scales from 50GB to petabytes.
• Global Distribution: Can be deployed across all Azure regions.
• Tunable Consistency: Options include eventual, consistent-prefix, session, bounded staleness, and strong consistency.

Architecture

• Data Structure: Utilizes unique data structures for indexing.
  • B+ Tree: A traditional choice for low write throughput, but can slow down with many indices.
  • BW Tree: Provides lock-free updates using compare and swap, improving write throughput.
• Partitioning:
  • Horizontal partitioning with a maximum of 50GB per partition.
  • Each replica set contains four replicas with one leader.
  • Forwarder propagates changes to the leader.

Conflict Resolution

• Strong consistency eliminates write conflicts.
• Delta records are written on SSD to improve write throughput.
• Conflict detection occurs during the compare and swap process.

Comparison with DynamoDB

• Indexing:
  • DynamoDB offers an optional sort key and supports a limited number of global secondary indices.
  • Cosmos DB supports indexing on all JSON fields.
• Query Capabilities: Cosmos DB provides more types of queries across different database types.

Pros and Cons

• NoSQL Databases:

  • Pros: Flexible schema design, horizontal scaling, performance tuning for specific workloads, favors availability over consistency.
  • Cons: Limited query capabilities (e.g., no joins).

• SQL Databases:

  • Pros: ACID compliance, structured schema, supports complex queries including joins.
  • Cons: Performance bottlenecks with large datasets.

Use Cases

• SQL:

  • E-commerce, financial systems, content management systems.

• NoSQL:

  • Social media, graph databases, chat applications, big data applications (e.g., Cassandra for Netflix), Internet of Things, high write-throughput scenarios (e.g., Philips Hue using DynamoDB).

• Hybrid Cases:

  • Gaming industry (e.g., Redis for leaderboards), e-commerce with personalized recommendations.