latencies

Latencies refer to the delay between initiating an action and observing its effect. In computing and communications, latency is the time from a request to its completed response, commonly measured in milliseconds. It reflects the total elapsed time across all stages of handling the request.

Latencies can be categorized by domain: network latency, storage latency, and compute latency. Network latency includes propagation delay, transmission delay, queuing, and processing at routers. Storage latency covers the time to access data on disks or solid-state drives, while compute latency refers to time spent by processors and software to handle the task.

In networks, end-to-end latency is the time from client to destination. One-way latency measures a single direction; round-trip time is twice that in many tests. Latency is often reported with averages and percentiles to reflect variability, or jitter.

Causes include physical distance, link quality, congestion, routing inefficiencies, hardware processing, software overhead, and protocol handshakes. Jitter arises from uneven delay across packets.

Relevance: latency affects interactivity and real-time performance. Real-time systems require predictable, bounded latency; streaming and bulk-transfer systems can tolerate higher latency.

Mitigation: improving hardware, optimizing software, reducing protocol overhead, caching and edge computing, content delivery networks, and network engineering such as traffic shaping. For storage, faster media (SSDs), memory caching, and RDMA can reduce latency. Good design aims to minimize tail latency and jitter while balancing throughput.