How Network Latency Drives Cloud Architecture Choices—from 5G to Gaming

1 What Is Latency

Latency is the time delay between an input and its corresponding output. Human reaction times are typically measured in milliseconds, with a commonly accepted limit of about 100 ms for perceptible responses. Mechanical reactions can be as fast as 50 ms, while high‑performance robots can react within a few milliseconds. Modern autonomous‑driving chips can achieve sub‑millisecond detection latencies.

2 Latency Requirements of Different Applications

Telecom systems classify applications by their latency and packet‑loss tolerances. Real‑time competitive games require ≤50 ms, real‑time interactive games ≤100 ms, voice ≤100 ms, video ≤150 ms, and most non‑real‑time internet services ≤300 ms. Adding 50 ms to these values defines a pass/fail threshold.

Machine‑oriented applications in 5G scenarios, such as industrial automation, remote driving, and ultrasonic detection, demand ultra‑low latency below 10 ms. High‑frequency trading and intra‑data‑center communication may need sub‑1 ms latency.

3 End‑to‑End Latency Composition

End‑to‑end latency consists of application‑layer processing delay and network‑layer transmission delay. Network transmission delay includes access delay, fixed‑line transmission, and data‑center internal networking.

Typical user‑side access delays: Wi‑Fi 5‑15 ms (light load) or <1 ms via Ethernet; 4G radio link 10‑20 ms; 5G radio link ~1 ms. Fixed‑line WAN to the cloud varies from 1‑100 ms depending on distance. Data‑center internal latency is usually <2 ms.

4 Deployment Architectures from a Latency Perspective

Industrial IoT : Requires ≤10 ms end‑to‑end latency, so network access (<1 ms) and data‑center proximity (<3 ms) are critical. Deployments often use edge‑cloud or hybrid cloud solutions located near factories.

Real‑time Audio/Video : Target ≤150 ms end‑to‑end latency, leaving ≤100 ms for the network. Multi‑region audio/video servers are recommended for 4G; a single region may suffice with 5G.

Real‑time Competitive Gaming : Aim for ≤50 ms (or ≤100 ms acceptable). Network layer must stay ≤50 ms, so multi‑region game servers are needed under 4G, but a single region could work with 5G.

Interactive Applications (e.g., live streaming, chat) : Acceptable end‑to‑end latency up to 150 ms, allowing network delay ≤100 ms. Single‑region deployment is often sufficient.

E‑commerce : Latency ≤300 ms is adequate; network delay ≤200 ms is acceptable. Single‑region deployment meets latency needs, though multi‑region redundancy is advised for reliability.

5 Measuring Latency

Ping : Simple ICMP ping provides round‑trip time between two points.

Real‑time Probing : Alibaba Cloud offers a free network probing service (https://zijian.aliyun.com/detect/ping) that measures latency from many points nationwide.

Continuous Monitoring : CloudMonitor’s site‑monitoring (https://cloudmonitor.console.aliyun.com) provides custom probes, alert rules, and supports email, SMS, phone, or DingTalk notifications.

6 Summary

Different applications have distinct end‑to‑end latency requirements, leading to varied deployment architectures.

End‑to‑end latency = application‑layer delay + network‑layer delay.

Application‑layer delay is hard to optimise; network‑layer delay can be reduced by architectural choices.

Network‑layer delay comprises access, fixed‑line, and data‑center internal delays.

Placing services closer to users reduces fixed‑line latency.

5G primarily reduces access latency, simplifying deployment for real‑time audio/video and gaming.

Latency measurement can be performed with ping, real‑time probing, and continuous monitoring tools.