Uninterrupted System Availability for Seamless Operations

Availability refers to the ability of a system to remain operational and produce expected results. It’s often measured as a percentage, such as 99.9%, 99.99%, or 99.999% uptime in public cloud systems. This availability is crucial for ensuring that services are consistently accessible to users without significant interruptions.

The formula to calculate uptime is:

Availability=(Uptime​/Total hours)×100

Designing a system for availability involves anticipating and mitigating potential failures. During the design process, it’s important to ask “What-If” questions to identify possible failure scenarios and plan appropriate solutions:

• What if requests overload a single system? Implement a load balancer to distribute the load.
• What if there are too many write operations? Create read replicas to offload write traffic.
• What if complex queries strain the system and database? Use caching mechanisms or optimize queries and database structures.
• What if requests start queuing up? Implement connection pooling, scale app instances, optimize queries, or normalize tables and create indexes.

System design often involves trade-offs, and choosing the right approach depends on the specific requirements and constraints. Below key topics are vital for attaining system availability.

Redundancy/Replication

One strategy for enhancing availability is redundancy or replication. This includes:

1. Hot Redundancy: Suitable for critical systems intolerant to failures. For instance, adding new application servers and load balancers to handle increased traffic.
2. Warm Redundancy: Acts as a shadow mode for tolerating momentary failures, such as using database replicas that can be promoted in case of a failure(master/slave replication).
3. Cold Redundancy: Acceptable for non-critical applications where downtime is acceptable, like storing backups in cost-effective services like S3 Glacier.

Identifying the Fault

Identifying faults involves monitoring system health through mechanisms like heartbeats in distributed systems. Clusters of nodes communicate with each other, and if a node stops sending heartbeats, it’s considered faulty. Technologies like Kafka Zookeeper or load balancers use these signals to detect and respond to failures, such as choosing a new leader or avoiding sending traffic to a failed instance.

Failover mechanisms

Failover mechanisms are critical for system resilience. They involve strategies like promoting a read replica to the main instance in case of database failures, handling pod failures in Kubernetes clusters, or implementing DNS failover to redirect traffic to alternate systems if one fails.

By integrating redundancy, fault detection, and failover mechanisms into system design, we can enhance availability and ensure continuous operation even during unforeseen failures or disruptions.