How this packet loss test works

What is sent, where it goes, how each number is calculated, and what the test cannot see.

Updated

Why a browser test needs WebRTC

A web page cannot normally send raw UDP. Everything a page fetches travels over HTTP, which runs on TCP or on QUIC, and both resend lost data automatically. A test built on ordinary requests can measure speed and ping, but it can never observe loss, because the transport repairs every gap before the page sees it.

The one way a browser can send unreliable UDP is WebRTC, the technology behind browser video calls. A WebRTC data channel can be configured to be unordered and to never retransmit, which makes it behave like a game or voice stream: a packet the network drops stays dropped. This test uses a data channel configured exactly that way.

Where the packets go

When you press Start, the page asks this site for a set of short-lived credentials for Cloudflare's TURN relay service. The page then opens two WebRTC connections inside your own browser and forces both to use the relay, over UDP only. Each packet leaves your device, travels to the nearest Cloudflare edge, and is relayed straight back to the second connection on your device. Cloudflare runs the relay in hundreds of cities and routes you to the nearest automatically, so the path measured is essentially your own connection: device, wifi, router, line and provider, up to the edge of the internet.

Because both ends of the trip are on your device, the timings use one clock. There is no clock synchronisation to get wrong, and the delay measured for each packet is a true round trip.

What is sent

  • 50 packets per second, one every 20 ms, which matches the rhythm of a 20 ms voice stream or a 50-tick game.
  • 160 bytes of payload per packet, the size of a standard voice frame. Each carries a sequence number and padding; nothing about you.
  • 15, 30 or 60 seconds: 750, 1,500 or 3,000 packets. Roughly 300 KB each way for a 30-second run, so the test does not load the line it is measuring.
  • Credentials that expire after five minutes, so the relay cannot be reused after your run.

How each number is calculated

  • Packet loss is the percentage of packets sent that did not come back within one second. A packet arriving later than that is counted as lost, because to a call or game it would be.
  • Ping is the median round-trip time of the packets that returned. The median is used rather than the average so that one very slow packet does not distort the headline number; the average, minimum, 95th percentile and maximum are shown in the detail panel.
  • Jitter is the mean absolute difference in round-trip time between each returned packet and the previous one, in send order.
  • Spiky seconds are one-second slices where at least 5% of packets were lost, or where the median ping was more than double the run's median and at least 30 ms above it.
  • The verdicts rate the run for each use against thresholds for loss, ping and jitter drawn from ITU-T G.114 for conversational delay, common voice-equipment guidance for loss and jitter, and the ping bands competitive games publish. The worst of the three decides the rating, and the page names which one it was.

Limits

The test measures your connection to the nearest Cloudflare edge. It cannot see loss on the route from there to a particular game server or call bridge, and it cannot separate loss on the way out from loss on the way back, since each packet does both. It also depends on the browser keeping to time: if the tab is moved to the background, browsers slow their timers, so the test stops rather than report false loss. The calculation code is a small, separately tested module; the thresholds are listed on the acceptable packet loss page.

Hardware Solutions to Fix Packet Loss & Jitter

Resolve intermittent connection drops, eliminate Wi-Fi interference, and stabilize your latency.

Network Gear
High-Speed Wireless Whole-Home Wi-Fi

Tri-Band Wi-Fi 6E / Wi-Fi 7 Gaming & Mesh Router System (2.5G Multi-Gig)

High-throughput tri-band router with dedicated 6 GHz backhaul, quad-core processor, and Smart Queue Management (SQM) to eliminate bufferbloat and latency spikes.

Drastically lowers jitter, stabilizes Discord audio, and ends multiplayer lag spikes
Modem Upgrade ISP Upgrade

DOCSIS 3.1 Multi-Gigabit Cable Modem (32x8 Channel Bonding)

Ultra-fast DOCSIS 3.1 standalone cable modem engineered with Broadcom processing to eliminate ISP modem rental fees and prevent Puma chip latency jitter.

Stops uncorrectable codeword errors and packet drops on cable internet lines
Clean Power Zero Dropouts

Battery Backup UPS & Surge Protector for Router & Modem (850VA - 1000VA)

Uninterruptible power supply with automatic voltage regulation (AVR). Protects routers, optical network terminals (ONT), and modems from micro-brownouts that cause line drops.

Keeps fiber and cable internet connected during neighborhood grid fluctuations
Wired Connection Zero Packet Loss

Shielded Cat 8 S/FTP Double-Braided Gigabit Ethernet Cable (50ft / 100ft)

Heavy-duty 40 Gbps 2000 MHz shielded cable with gold-plated RJ45 connectors. Replaces congested wireless radio frequencies with pure wired throughput.

Eliminates 99% of Wi-Fi packet drops, radio interference, and ping spikes
Hardware Interface 2.5G Multi-Gig

USB-C to 2.5 Gbps Multi-Gigabit Ethernet Adapter (Aluminum Housing)

Plug-and-play aluminum network adapter providing an instant 2.5G / 1G RJ45 wired LAN port for MacBooks, ultrabooks, and desktop workstations lacking ethernet jacks.

Delivers full wired reliability for thin laptops without internal LAN ports
LAN Distribution 8-Port Metal

Metal 8-Port Gigabit Unmanaged Ethernet Network Switch (QoS Prioritized)

Fanless industrial metal desktop switch with non-blocking wire-speed forwarding and 802.1p traffic prioritization for gaming consoles, PCs, and smart TVs.

Connects multiple high-bandwidth devices over a single main router drop

Network Latency & Packet Loss Troubleshooting Guide

Identify network failure points and compare verified hardware solutions to restore ping stability:

Test Symptom Probable Root Cause Isolation Step Hardware Fix
High packet loss (>2% packet drop) Wi-Fi radio interference or congested 2.4/5GHz spectrum Ping your local gateway IP to test wireless hop stability Hardwire with shielded Cat 8 S/FTP ethernet cable
Severe jitter spikes (>25ms variation) Bufferbloat on ISP-issued modem/router during uploads Run ping test while performing a file upload or streaming Tri-band Wi-Fi 6E/7 router with SQM bufferbloat control
Random disconnects every few hours DOCSIS T3/T4 timeouts or micro-voltage line sags Inspect modem downstream power levels and event logs DOCSIS 3.1 standalone modem paired with battery backup UPS
Laptop Wi-Fi drops during video calls Throttled internal wireless chipset or metal desk occlusion Test ping stability next to router vs at normal desk USB-C to 2.5G aluminum wired ethernet adapter
Multi-device lag when gaming & streaming Unmanaged packet collisions across shared Wi-Fi airtime Check router client list for high-traffic background devices Dedicated 8-port gigabit switch with hardware 802.1p QoS

Network Pro Tip: Eliminating Bufferbloat on Busy Home Connections

If your ping spikes from 20ms to 200ms whenever someone uploads a video, backs up iCloud, or streams 4K video, your router is suffering from bufferbloat (oversized router packet queues that stall time-sensitive gaming packets). Upgrading to a modern router featuring Smart Queue Management (SQM like CAKE or FQ-CoDel) keeps latency flat regardless of household network load.

As an Amazon Associate we earn from qualifying purchases. Affiliate Disclosure

Frequently asked questions

Is the test accurate?

For loss, ping and jitter between your device and the nearest Cloudflare edge, yes: each figure comes from 750 to 3,000 individually timed packets. It does not measure paths beyond that edge.

Does the test store my results?

No. Results are calculated in your browser and not sent anywhere. The relay only handles the anonymous test packets.

Why do I have to keep the tab open?

Browsers throttle timers in background tabs, which would delay packets and fake loss. The test stops if the tab is hidden.