Network Latency Guide

Acceptable, ideal and poor latency thresholds by application type — VoIP, gaming, database queries, bulk transfer — for network design, troubleshooting a vague "it feels slow" complaint, or writing an SLA that actually reflects what the application needs rather than a round number someone picked.

RTT vs one-way, and why the distinction matters

Ping measures round-trip time — there and back. One-way latency, roughly half of RTT, is what actually matters for something like VoIP, where the ITU G.114 recommendation caps acceptable one-way delay at 150ms for voice quality that doesn't feel laggy. ping, traceroute, mtr and smokeping cover most of what you need to actually measure and monitor this rather than guess at it.

Tools, and what each one is actually for

• ping — basic RTT. Run ping -c 100 for something you can actually draw conclusions from, not just three pings and a guess.
• traceroute / tracert — shows latency at each hop, so you can see exactly where a delay is being introduced rather than just knowing the total is high
• mtr — traceroute and ping combined, running continuously, with per-hop packet loss statistics
• iperf3 — actual throughput and jitter between two endpoints, not just latency
• smokeping — long-term trend monitoring, the tool for catching a latency problem that's gradually getting worse before it becomes an outage

What's actually driving high latency

• Distance — the unavoidable speed-of-light floor. CDNs and edge computing exist specifically to work around this by moving content physically closer to users.
• Congestion — too much traffic for available capacity causes queuing delay. QoS, bandwidth upgrades, or traffic shaping address this directly.
• Routing inefficiency — a suboptimal BGP path adds hops that don't need to be there. Sometimes fixable with your ISP, sometimes only with direct peering.
• Buffer bloat — over-eager buffering in routers and modems adds latency specifically under load. CoDel or FQ-CoDel queue management is the actual fix, not just buying more bandwidth.
• Wi-Fi — channel contention and retransmissions add variable, unpredictable latency. For anything genuinely latency-sensitive, wired still wins, no contest.