Reliable, Scalable, and Maintainable Applications

Thinking About Data Systems

Modern applications are data-intensive, not compute-intensive. The bottleneck is usually the amount, complexity, and speed of data — not raw CPU power. Databases, caches, search indexes, stream processing, and batch processing are all "data systems" with blurring boundaries.

Reliability

Reliability means the system continues to work correctly even when things go wrong. Faults (hardware, software, human) are inevitable; the goal is fault-tolerance, not fault-prevention. Hardware faults are increasingly handled by software-level redundancy. Human errors cause most outages — design systems that minimize opportunities for mistakes.

Scalability

Scalability is about describing how a system copes with increased load. First, describe load with load parameters (e.g., Twitter's fan-out problem: 12k writes/sec vs 300k reads/sec shifted the design from pull to push). Then describe performance — throughput for batch systems, response time for online systems. Use percentiles (p50, p95, p99) rather than averages, because tail latencies matter.

Maintainability

Three design principles for long-term health:

• Operability: Make it easy for ops teams to keep the system running smoothly.
• Simplicity: Manage complexity through good abstractions. Accidental complexity is the real enemy.
• Evolvability: Make it easy to adapt the system to changing requirements.