Availability

What is Availability?

Availability measures the proportion of time a system is operational and accessible. It's typically expressed as a percentage, often called "nines" of availability.

The "Nines" of Availability

Note: Five nines (99.999%) is considered "high availability" and is the target for critical systems like banking, healthcare, and emergency services.

Calculating Availability

Single Component

Availability = MTBF / (MTBF + MTTR)

Where:

• MTBF: Mean Time Between Failures
• MTTR: Mean Time To Recovery

Systems in Sequence

When components are in series (all must work):

Total Availability = A₁ × A₂ × A₃ × ... × Aₙ

Example: Three services at 99.9% each:

0.999 × 0.999 × 0.999 = 99.7%

Systems in Parallel (Redundancy)

When components are redundant (any one can work):

Total Availability = 1 - (1 - A)ⁿ

Example: Two servers at 99% each:

1 - (1 - 0.99)² = 1 - 0.0001 = 99.99%

Strategies for High Availability

1. Redundancy

• Active-Passive: Standby takes over on failure
• Active-Active: Multiple instances handle requests simultaneously
• N+1 Redundancy: One extra component beyond minimum required

2. Replication

• Database replication: Multiple copies of data
• Multi-region deployment: Survive regional outages
• Cross-zone redundancy: Survive datacenter failures

3. Load Balancing

• Distribute traffic across healthy instances
• Automatic health checks and failover
• Geographic load balancing for global availability

4. Fault Isolation

• Bulkhead pattern: Isolate failures to prevent cascade
• Circuit breaker: Stop calling failing services
• Graceful degradation: Provide reduced functionality instead of complete failure

Availability Patterns

Failover

Active-Passive (Cold Standby):

• Primary handles all traffic
• Secondary activated on primary failure
• Simple but causes brief downtime during switch

Active-Active (Hot Standby):

• Multiple instances handle traffic
• No downtime on single failure
• More complex to implement

Health Checks

Systems should continuously monitor component health:

• Liveness probes: Is the service running?
• Readiness probes: Can it handle requests?
• Deep health checks: Are dependencies healthy?

Availability vs Other Properties

Real-World Examples

High Availability Systems

Downtime Incidents

• 2017 AWS S3 outage: Took down thousands of websites for 4 hours
• 2021 Facebook outage: 6-hour global outage due to BGP misconfiguration
• 2020 Google outage: 45-minute outage affecting Gmail, YouTube, etc.

Designing for Availability

Questions to Ask

1. What's the acceptable downtime for this system?
2. What's the cost of downtime (revenue, reputation)?
3. What failure scenarios must we handle?
4. What's the budget for redundancy?

Best Practices

• Define SLAs/SLOs: Set clear availability targets
• Plan for failure: Assume components will fail
• Test failover: Regularly test backup systems
• Monitor everything: Detect issues before users do
• Automate recovery: Reduce MTTR with automation

Interview Tips

• Know the nines: Memorize the availability percentages and their downtime
• Calculate compound availability: Understand how components in series/parallel affect total availability
• Discuss trade-offs: Availability often trades off with consistency (CAP theorem)
• Mention real examples: Reference AWS, Google, etc. availability strategies
• Consider cost: Higher availability = higher cost

Summary

Availability is crucial for user trust and business continuity. Achieving high availability requires:

1. Redundancy at every layer
2. Automated failover mechanisms
3. Continuous monitoring and alerting
4. Regular testing of failure scenarios
5. Clear SLAs and recovery procedures