SpeedFast Measurement Methodology

How SpeedFast measures download throughput, upload throughput, idle ping, loaded ping (bufferbloat), and jitter across a distributed edge network — documented openly for transparency and technical scrutiny.

Architectural Overview

SpeedFast is designed as a browser-native measurement tool running on top of Cloudflare's global edge network. Rather than relying on a centralized cluster of speed test servers or synthetic artificial probing, every test connects to an edge node physically close to the user.

Measurements execute inside a dedicated Web Worker to isolate network processing from the browser's main UI rendering thread, preventing UI rendering bottlenecks or JavaScript execution pauses from skewing throughput or latency calculations.

1. Latency & Jitter Measurement (Idle Phase)

Before any large payload transfer begins, SpeedFast establishes an initial baseline of round-trip network performance under zero-load conditions.

A sequence of consecutive lightweight HTTP requests is transmitted to the nearest edge endpoint. Each request records the precise high-resolution timestamp (using performance.now()) between packet transmission and response arrival.

Idle Ping is calculated as the minimum stable round-trip time (RTT). Jitter is computed as the mean absolute deviation between consecutive RTT samples, measuring latency stability before data transfer introduces congestion.

2. Download Throughput & Loaded Latency

To measure true sustainable download capacity, SpeedFast opens multiple parallel HTTP connections fetching binary chunks from the edge node. Multiple streams are essential to bypass TCP slow-start bottlenecks and accurately fill the connection's bandwidth-delay product (BDP).

The first 1.5–2.0 seconds of throughput measurements are trimmed as warm-up data to prevent TCP slow-start ramps from artificially deflating results. Sustained throughput is calculated using a rolling weighted average across all active streams.

Concurrently, continuous ping probes are fired during the active download stream. The difference between this download-loaded ping and the baseline idle ping quantifies downstream [Bufferbloat](/guides/what-is-bufferbloat) — revealing whether the local router's buffers cause real-time lag during heavy downloads.

3. Upload Throughput & Upstream Bufferbloat

Upload throughput is measured by streaming cryptographically generated random byte arrays via HTTP POST requests across parallel connections to the edge node.

Similar to the download phase, initial warm-up bursts are trimmed, and active ping probes measure upload-loaded latency to detect upstream bufferbloat.

Testing upload against close edge servers provides an accurate measurement of the physical uplink provisioned by your ISP, uncorrupted by international transit throttling.

4. Why We Avoid Artificial Smoothing

Many commercial speed tests apply aggressive peak-burst filtering or artificial smoothing algorithms to display higher numbers that match ISP advertising tiers.

SpeedFast reports realistic, sustained throughput and explicitly highlights loaded latency. We believe that knowing your true connection stability and bufferbloat is far more valuable than an inflated single-burst number.

Peer Review & Engineering Contact

Our methodology is continuously evaluated against active IETF standards and modern congestion control behaviors (BBR, CUBIC). For technical inquiries, methodology reviews, or research feedback, contact the SpeedFast Engineering Team at support@speedfast.app.