System Design flashcards, generated for you.

The ability of a system to handle increased load (users, data, requests) by adding resources. Measured by how performance changes as scale increases.

Horizontal: Add more machines (e.g., add servers to a load-balanced cluster). Vertical: Increase capacity of existing machine (e.g., more CPU/RAM). Horizontal scales better but adds complexity; vertical has limits.

A distributed system can guarantee at most 2 of 3: Consistency (all nodes see same data), Availability (system always responds), Partition tolerance (survives network splits). Real systems must choose: CP (strong consistency) or AP (eventual consistency).

Consistency: Stricter but slower (all replicas must agree before responding). Availability: Faster but risks stale reads (serve locally, sync later). Choose based on domain: banking needs consistency; social feeds tolerate eventual consistency.

Latency: Time for single request to complete (ms). Throughput: Requests completed per unit time (req/sec). Low latency matters for user experience; high throughput matters for total capacity. Both matter, but priorities differ per use case.

Caching: Stores copies of data for faster access; data loss acceptable (data-of-record elsewhere). Replication: Maintains redundant copies for fault tolerance; data loss unacceptable. Caching improves latency; replication improves durability.

Strong: All reads see latest write immediately. Eventual: Reads eventually see latest write after propagation delay. Causal: Causally-related writes are ordered; concurrent writes may diverge. Strength decreases; scalability increases in that order.

Pro: Reduces latency by serving content geographically close; offloads origin server. Con: Added cost, cache invalidation complexity, stale content risk, only effective for read-heavy static assets. Choose if geographic distribution is critical.

Sharding: Partitions data across multiple databases (each shard holds different rows/keyspace). Replication: Multiple databases hold identical data. Sharding increases capacity and throughput; replication increases availability and fault tolerance. Often used together.

Normalization: Reduces redundancy, consistency easier, slower reads (joins required). Denormalization: Data duplication, faster reads, consistency harder (updates must sync replicas). Denormalize when read performance is critical and eventual consistency acceptable.