Unit
- High Availability (HA): A system property ensuring continuous operation for a defined period, typically measured by uptime percentage. The commonly used measure is the number of nines: 99.9% (three nines) allows 8.76 hours downtime per year, while 99.999% (five nines) permits only 5.26 minutes.
- Uptime vs. Downtime: Uptime is the duration a system operates correctly; downtime is the period it fails. Planned downtime for maintenance is often excluded from SLA calculations, but high availability design aims to eliminate even planned downtime through techniques like rolling updates.
- SLA (Service Level Agreement): A contract specifying guaranteed availability levels. Typical SLAs range from 99.9% to 99.999%. Failing to meet SLA targets often incurs financial penalties.
- Fault Tolerance vs. High Availability: Fault tolerance implies the system continues operating without interruption when a component fails, while high availability focuses on minimizing downtime. Fault tolerance is a subset of high availability.
- Single Point of Failure (SPOF): A component whose failure causes the entire system to fail. Identifying and eliminating SPOFs is a primary goal in designing HA systems.
- Redundancy: Duplicating critical components to provide backup in case of failure. Redundancy can be active-passive or active-active, and can be applied at hardware, network, and software levels.
- MTBF and MTTR: Mean Time Between Failures (MTBF) measures reliability; Mean Time To Repair (MTTR) measures how quickly a system can be restored. Reducing MTTR is often more cost-effective than increasing MTBF.
- Five Nines Goal: The gold standard for high availability is 99.999% uptime, equating to about 5 minutes of downtime per year. Achieving this requires automated failover, redundant infrastructure, and rigorous testing.
- Load Balancing: Distributing incoming traffic across multiple servers to prevent any single server from becoming overwhelmed. Load balancers can detect server failures and redirect traffic to healthy instances, thereby reducing downtime and improving resource utilization in HA architectures.
- Failover Clustering: A group of independent servers that work together to maintain high availability. If the primary node fails, a secondary node automatically takes over without user intervention. This approach minimizes downtime by ensuring seamless transition and is commonly used in database and application clusters.
- Geographic Redundancy: Deploying systems in multiple physical locations to protect against site-level failures such as natural disasters or power outages. Data is replicated synchronously or asynchronously between regions, and traffic is routed away from the affected location using DNS or global load balancers.
- Monitoring and Alerting: Continuous observation of system metrics such as CPU usage, memory, disk I/O, and network latency. Proactive monitoring detects anomalies before they cause outages, and automated alerts notify teams to respond quickly. This is critical for reducing MTTR and maintaining uptime targets.
Code Example
Python
uptime = 365 * 24 * 60 * 60
downtime = 5 * 60
availability = (uptime / (uptime + downtime)) * 100
print('Five nines availability: {}%'.format(availability))This code calculates the availability percentage for a system that has 5 minutes of downtime per year, yielding 99.999%.
Quick Revision
- High availability measures continuous operation via uptime percentage; five nines is 99.999%.
- Key metrics are MTBF and MTTR; reducing MTTR is often more practical.
- Eliminating single points of failure through redundancy is core to HA design.
- SLAs define contractual availability targets with penalties for non-compliance.