Measuring Network Performance: What Speed Tests Actually Tell You

A speed test reports three numbers, download, upload, and latency, that together summarize your connection quality. They're useful for spotting problems (sudden drops vs your baseline) and comparing networks (Wi-Fi vs ethernet, ISP A vs ISP B). They're less useful as absolute truth: a single test on a single device against a single server is one data point, not gospel. Understanding what each number means and how they're measured turns the test into a real diagnostic tool.

The Three Numbers

Download Speed (Mbps)

Megabits per second, how fast data flows INTO your device. Capital M = Mega (1,000,000). Mbps (lowercase b) = megabits; MBps (uppercase B) = megabytes (8× larger). Internet speeds quoted in bits, file sizes in bytes, divide bits-per-second by 8 to get bytes-per-second.

A 100 Mbps connection downloads:

• 12.5 MB/second.
• A 1 GB file in ~80 seconds.
• A 4K stream (15-25 Mbps) easily.

Upload Speed (Mbps)

Same units, opposite direction. Often much slower than download on residential connections:

• Cable: typical 100/10 (100 down, 10 up).
• DSL: typical 50/5.
• Fiber: often symmetric (1000/1000).
• 5G/LTE: typically 50-200 down, 5-50 up.

Upload matters for: video calls, cloud backup, file sharing, streaming (Twitch broadcasters), self-hosting, large email attachments.

Latency / Ping (ms)

Milliseconds for a round-trip from your device to a server and back. Lower is better.

• <10ms, fiber to local server (same metro area).
• 10-30ms, fiber/cable to nearby cloud region.
• 30-70ms, cross-country.
• 70-150ms, international (across an ocean).
• 150ms+, satellite, distant server, or congested route.

Latency affects:

• Interactivity: gaming (>50ms = noticeable lag for competitive), video calls.
• Web page load: each round-trip (DNS, TCP, TLS, then HTTP requests) costs latency. A page needing 10 round-trips on a 100ms connection = 1 second just from latency.
• API responsiveness: REST calls take ~latency minimum.

For most workflows, latency matters more than bandwidth. A 50 Mbps / 20ms connection feels faster than a 500 Mbps / 200ms connection.

How Tests Work

The general algorithm:

1. Pre-test ping: 10-20 small requests to estimate latency.
2. Download test: download a large file (or many chunks) from a test server, measure bytes-per-second. Use multiple parallel streams to saturate the link.
3. Upload test: upload a large blob, measure rate.
4. Calculate: bytes / time = throughput.

Tools differ in:

• How many parallel streams (more streams = closer to true link capacity, but harder for browsers).
• Where the server is (nearest vs random vs user-chosen).
• What protocol (HTTPS/HTTP, sometimes UDP for ping).
• Test duration (longer = more accurate, more data used).

Browser Limitations

This tool runs in a browser, which adds overhead:

• Connection limits per origin: browsers cap parallel connections (HTTP/1.1: ~6 per origin; HTTP/2: multiplexed but still flow-controlled).
• JS processing: each chunk has to be decoded by JS, adding CPU overhead.
• Performance.now() resolution: typically 1ms; sub-millisecond timing is unreliable.
• Page lifecycle: backgrounded tabs are throttled.

Real-world: browser-based tests measure 80-95% of true link speed. For relative comparisons (vs yesterday), this is fine. For absolute "is my ISP giving me what they advertise" claims, use a dedicated client like Speedtest CLI.

What Affects Your Result

Wi-Fi vs ethernet. Wi-Fi loses 20-50% of theoretical throughput due to overhead, interference, signal strength. A gigabit ISP over Wi-Fi might test at 300-500 Mbps; same connection over ethernet at 900+ Mbps. Test both to identify Wi-Fi as the bottleneck.

Distance from router. Each wall, each meter, drops Wi-Fi quality. The router 3 rooms away might give 50 Mbps when the same router 1m away gives 800.

Wi-Fi band. 2.4 GHz penetrates walls better but is slower; 5 GHz is faster but shorter range; 6 GHz (Wi-Fi 6E) is faster still but newer device only.

Other devices on your network. A family member streaming 4K uses 25 Mbps; you'll get less. Test when no one else is active for max numbers.

Time of day. ISP peak hours (7-11pm) congestion can halve speeds. Test off-peak to compare.

Server location. Testing to a far-away server gives lower numbers due to TCP/latency interactions. Most tests pick nearest automatically.

VPN. Encryption + remote routing reduces effective throughput by 10-50% depending on VPN quality.

Device capability. Old phones, weak Wi-Fi adapters, congested CPUs all bottleneck before the network does.

Diagnosing Common Issues

Slow web browsing despite high test speeds

The test measures bulk transfer. Web browsing involves many short connections, DNS lookups, TCP handshakes, TLS, etc., each costing a round-trip. If latency is high or DNS is slow, browsing feels slow even at high throughput.

Diagnose:

• Test DNS: dig or use 1.1.1.1 / 8.8.8.8 to see if your ISP DNS is the issue.
• Test latency: ping should be <50ms to typical servers.
• Check for TCP retransmits (advanced, use tools like Wireshark).
• Try a different DNS provider.

Inconsistent speed test results

If results vary 50%+ between tests:

• Wi-Fi signal varies → wired test, or move closer to router.
• Other device activity → check for background downloads, streaming, syncing.
• ISP congestion → test multiple times across the day.
• Bufferbloat → some routers buffer too much, hurting active-test latency.

High latency on otherwise-good connection

• Far from CDN endpoints (rural, international).
• Routing problem at ISP (traceroute reveals hop with high latency).
• VPN routing through distant server.
• Bufferbloat (active traffic causes latency spikes; test with DSLReports bufferbloat test).

Asymmetric upload/download problem

Most residential connections are asymmetric (100 down / 10 up). If your upload-heavy activity (video calls, cloud backup) is slow on otherwise-fast connection, that's expected. Upgrade to a symmetric (fiber) plan if needed.

Speed Test Tools Compared

• fast.com (Netflix), simple download test, no ads. Server is Netflix CDN.
• speedtest.net (Ookla), most comprehensive, with global server network, history, plans.
• librespeed, open source, self-hostable.
• Cloudflare speed.cloudflare.com, uses Cloudflare's network, also tests TCP / packet loss / jitter.
• DSLReports speedtest, includes bufferbloat measurement.
• Speedtest CLI (speedtest command), Ookla's CLI, more accurate than browser.
• iperf3, for testing inside a network (between two devices you control).

For absolute accuracy, run speedtest-cli or iperf3. For quick browser checks, fast.com / Cloudflare / this tool.

Throughput Math

Useful conversions:

• 1 Mbps = 0.125 MB/s.
• 100 Mbps = 12.5 MB/s = 750 MB/minute = 45 GB/hour.
• 1 Gbps = 125 MB/s.
• 4K Netflix stream: ~25 Mbps = 11.25 GB/hour.
• HD video call: 1-3 Mbps each direction.
• VR streaming: 50-100 Mbps target.

These numbers help check: a 50 Mbps connection should handle two simultaneous 4K streams (25 + 25 = 50). It might just barely; in practice, 4K HDR needs more headroom for variable bitrate spikes.

Privacy

The test downloads data from a public test endpoint to measure throughput. The endpoint's CDN logs the request (your IP, time, byte count), standard for any HTTP request. No additional analytics or fingerprinting is done by this tool. Open DevTools Network during use: you'll see the test data flow and nothing else. Some commercial speed test services log results, fingerprint your browser, and sell aggregated data to ISPs and advertisers, this tool doesn't do that.