How Velocity Works & Why Results Differ From Other Speedtests
1. Ping (Latency): HTTP/2 Edge Proxy vs. ICMP / Native Sockets
Traditional speedtests like Ookla or command-line pings use native ICMP or raw TCP sockets to single ISP servers. Velocity runs entirely inside the browser sandbox and measures pure HTTP/2 Round-Trip Time (RTT) to global edge proxy nodes—reflecting the real-world latency you experience when loading modern websites, cloud services, and APIs.
2. Download Engine: Parallel Streams & TCP Slow-Start Warmup
At the start of a transfer, TCP congestion algorithms gradually ramp up their window size (Slow-Start). Velocity launches parallel Web Worker streams, discards the first 1.5 seconds of ramp-up (warmup), and calculates sustained throughput (80th percentile) so buffer fluctuations don't skew your final result.
3. Upload Engine: Adaptive Payloads & Socket-Buffer Smoothing
While many web speedtests estimate uploads in coarse chunks or over-count local OS TCP send buffers, Velocity uses adaptive 1–8 MB payloads verified per HTTP completion inside isolated Web Workers and computes the Interquartile Mean (25th–75th percentile) over a 1000ms rolling window for true wire-speed accuracy.
4. Why Do Speedtests Differ Slightly From Each Other?
Server Location & Routing: Each speedtest measures against different target servers. Ookla often selects local ISP-hosted nodes, Fast.com uses Netflix CDNs, and Velocity benchmarks against global edge cloud networks.
Browser Sandbox: Native desktop apps open raw OS sockets directly, whereas browser-based speedtests measure through the browser's real HTTP/TLS networking stack.