Modern Cloud Architectures - Redundancy

Redundancy is a key design principle in modern cloud architectures that improves fault tolerance, availability, and performance.

Why Use Redundancy?

• Performance: Distribute workload across multiple replicas to improve response time
• Error Detection: Compare results when replicas disagree
• Error Recovery: Switch to backup resources when primary fails
• Fault Tolerance: System continues functioning despite component failures

Importance of Fault Models

The effectiveness of redundancy depends on how individual replicas fail:

• For independent crash faults, the availability of a system with n replicas is:

  Availability = 1-p^n
  

  Where p is the probability of individual failure

• Example: 5 servers each with 90% uptime → overall availability = 1-(0.10)^5 = 99.999%

This only holds if failures are truly independent, which requires consideration of common failure modes.

Redundancy by Replication

Replication involves maintaining multiple copies of:

• Data
• Services
• Infrastructure components

Data Replication

• Synchronous Replication: Write operations complete only after all replicas are updated

  • Ensures consistency but increases latency
  • Used for critical data where consistency is paramount

• Asynchronous Replication: Primary replica acknowledges writes before secondaries are updated

  • Better performance but may lose data if primary fails before replication
  • Used when performance is prioritized over consistency

• Quorum-based Replication: Write operations complete when a majority of replicas acknowledge

  • Balances availability and consistency

Service Replication

• Active-Passive Replication:

  • One active instance handles all requests
  • Passive instances ready to take over if active fails
  • Lower resource utilization but potential downtime during failover

• Active-Active Replication:

  • Multiple active instances handle requests simultaneously
  • No downtime during instance failure
  • Requires more complex state management

Infrastructure Redundancy

Modern cloud data centers implement redundancy at multiple levels:

Hardware Redundancy

• Geographic Redundancy:

  • Data centers distributed across multiple regions
  • Mitigates regional outages from natural disasters, power grid failures
  • Data typically replicated across regions

• Server Redundancy:

  • Servers deployed in clusters with automatic failover
  • If one server fails, another takes over seamlessly

• Storage Redundancy:

  • Data replicated across multiple devices and technologies
  • RAID configurations protect against disk failures

Network Redundancy

1. Server-level Redundancy:

   • Redundant Network Interface Cards (NICs)
   • Dual or more power supplies

2. Network-level Redundancy:

   • Redundant switches, routers, firewalls, load balancers

3. Link and Path-level Redundancy:

   • Link aggregation (multiple links between devices)
   • Spanning Tree Protocol to prevent network loops
   • Load balancing across multiple paths

Network topologies designed for redundancy:

• Hierarchical/3-tier topology
• Fat-tree/clos topology

Power Redundancy

• Multiple power feeds from different utility substations
• Uninterruptible Power Supplies (UPS) for temporary outages
• Backup generators for medium/long-term outages
• Power Distribution Units with dual inputs

Cooling Redundancy

• N+1 configuration (one extra cooling unit than required)
• Multiple cooling technologies
• Redundant cooling loops (pipes, heat exchangers, pumps)
• Hot/cold aisle containment

Redundancy Challenges

• Cost: Redundant systems require additional hardware and management
• Complexity: More components mean more potential failure points
• Consistency: Maintaining consistent state across replicas
• Testing: Verifying redundancy actually works as expected