Network Performance
What causes packet loss — and how to isolate it
Packet loss is the single most disruptive network defect because it converts smooth interactive workloads into stuttering ones. The causes form a short list — and a systematic triage workflow finds the culprit faster than guessing.
Local Wi-Fi (most common)
Wi-Fi shares airtime with neighbors and household devices. Interference, weak signal, and channel congestion all manifest as packet loss. Symptoms: loss disappears when you plug in via Ethernet.
Fix: move closer to the AP, change channels, consolidate onto less-congested 5/6 GHz bands, add a mesh node or wired AP for the affected area.
Saturated upload
Cable plans often have modest upload bandwidth (10–35 Mbps). One device uploading a large backup can saturate the upstream and cause loss in both directions because TCP ACKs can't get back. Symptoms: loss spikes coincide with cloud sync, backups, or video uploads.
Fix: throttle the upload, schedule it for off-hours, or enable router-level QoS / SQM to prevent any single flow from monopolizing the upstream.
Bufferbloat
Excessive buffering in your modem or router during congestion creates massive latency spikes and indirect loss. Run the bufferbloat-aware speed test — loaded latency well above idle latency is the signature.
Fix: a router with fq_codel or CAKE qdiscs enabled (most modern routers support SQM).
Faulty cabling and connectors
Crushed Ethernet, corroded coax connectors, kinked patch cables. Symptoms: persistent low-level loss that doesn't change with usage; sometimes accompanied by speed-test variability.
Fix: swap suspect cables one at a time; reseat coax connectors; replace any cable older than ~10 years that's seen rough handling.
ISP / upstream issues
Maintenance, fiber cuts, peering disputes, neighborhood congestion at peak hours. Symptoms: loss appears past your router in traceroutes, often time-of-day correlated.
Fix: collect traceroutes with timestamps and open a ticket with concrete data — 'internet is slow' tickets get triaged slowly; 'consistent 5% loss at hop 4 between 7–10pm' tickets escalate fast.
The triage workflow
- Reproduce loss while plugged in via Ethernet. If it disappears, the problem is Wi-Fi.
- Run continuous ping to your gateway. Loss there = local network.
- Run continuous ping to 1.1.1.1. Loss past the gateway = ISP/upstream.
- Traceroute during a loss event. Pin down the hop where loss begins.
- Reproduce at different times of day to distinguish congestion from hardware faults.