30 Must-Know Concepts

Increasing the computing capacity (such as CPU, RAM, or SSD size) of a single server machine to handle higher computational demands.

Adding more commodity hardware servers to a shared resource pool managed under a dynamic load balancer, facilitating unlimited capacity expansion.

A binding legal agreement between a service provider and external clients detailing system reliability guarantees, uptime metrics, and financial penalties for failures.

An internal target metric set by engineering teams to preserve service quality, serve as a reliability boundary, and guide developer error budgets.

The actual quantitative metric measured in real-time to track compliance against active SLO objectives.

The capability of an architecture to remain operational and accessible under high-traffic spikes and hardware failures, typically measured in nines (e.g. 99.99%).

The probability that a distributed platform executes its target functions correctly under stated conditions without errors, corruptions, or bitrot for a specified timeframe.

A central component or node in an infrastructure whose single failure triggers a complete cascading system outage.

The absolute round-trip time elapsed (typically measured in milliseconds) for a single network packet to travel from a client agent to a server and return.

The capacity of requests or operations a system safely processes per unit of time, typically measured in Requests Per Second (RPS) or Queries Per Second (QPS).

The maximum rate of data transfer across a physical network connection interface, measured in bits per second (e.g. Gbps).

A consistency model guaranteeing that a read operation always returns the absolute most recent write transaction, regardless of which distributed node is queried.

A consistency model promising that if no new updates are made, all distributed replicas will eventually synchronize and converge to the same value.

The fundamental constraint stating that under a network partition, a distributed system can guarantee Consistency (C) or Availability (A), but not both.

An extension of CAP stating that even when no partition exists (E), a system must trade off response Latency (L) against data Consistency (C).

Distributing inbound client network traffic dynamically across multiple backend servers to prevent compute overload.

A network of cache proxy servers located at the network edge, closer to users, to serve media assets quickly.

The global decentralized register mapping human-readable hostnames into computer-readable IP addresses.

Storing high-frequency query responses in fast temporary memory stores to bypass slow persistent disk storage blocks.

Partitioning a massive table horizontally across multiple physically distinct database instances based on a sharding key.

Replicating transactions from a primary database asynchronously to auxiliary nodes, scaling out read capacity.

A hash distribution strategy mapping nodes and keys onto a circular ring, minimizing key remapping when servers scale out or in.

Asynchronous event buffers that decouple publishers and consumers, safely absorbing large peak traffic ingestion spikes.

An event-driven architectural pattern where publishers broadcast to topics, and multiple subscribers consume independently.

A structural safety switch that intercepts calls to a failing downstream service to conserve backend threads.

Throttling client request volumes within a specific time window to protect downstream resources from exhaustion.

Periodic lightweight signals transmitted between distributed cluster nodes to coordinate cluster status and detect silent node crashes.

A decentralized peer-to-peer communication framework where nodes exchange cluster status incrementally with adjacent neighbors, similar to epidemics.

A synchronization mechanism designed to coordinate mutual exclusion across shared resources in stateless server farms.

A transactional guarantee ensuring that executing an API request multiple times yields the exact same outcome as running it a single time.