Skip to content
pingtest.si
SLSlovenščina Sign in

Methodology

Everything the site shows follows the rules below. The test statistics are computed in your browser; the index and dossiers on the server.

1. Latency in the browser

Round-trip time (RTT) is how long a message takes to travel from your browser to our nearest edge node and back, timed with the browser’s high-resolution clock. Early replies that include connection set-up are discarded, and we show the median of the rest because single delays do not move it. If a network restricts the default measurement path, the test switches to a fallback on the same node automatically.

2. Why not ICMP ping

Browsers cannot send ICMP, so the test runs at the application layer. Consequences: (a) values are usually a fraction of a millisecond above the ping command; (b) the browser resends lost data, so network loss often shows up as a latency spike rather than a lost packet. Server-side measurements to games and services use real ICMP or the TCP handshake time, which equals one round trip.

3. RFC 3550 jitter

We use the smoothed estimator from RFC 3550 section 6.4.1, the one VoIP and video applications use:

D(i−1, i) = RTT(i) − RTT(i−1)
J(i) = J(i−1) + ( |D(i−1, i)| − J(i−1) ) / 16

RFC 3550 defines it on one-way transit times; browser and server clocks are not synchronised, so we apply it to consecutive round trips. The 1/16 factor smooths single spikes. Server-side measurements (games, DNS) report the mean absolute difference of consecutive replies.

4. Packet loss

Loss = (sent − received in time) / sent × 100 %. A reply that arrives too late to be useful for a call or a game counts as lost, which is how real-time applications treat it too.

5. Bufferbloat

The loaded test saturates your connection in both directions, much like heavy streaming or a large download, and keeps measuring latency while it does. Only measurements taken while the line is genuinely busy count, so the result shows how much queueing in your router or at your provider adds to your delay.

Δ = median(loaded RTT) − median(idle RTT)
A+: Δ < 5 ms   A: < 30   B: < 60   C: < 200   D: < 400   F: ≥ 400 ms

A grade is given only when the loaded measurement is reliable. The throughput shown (Mbps) is indicative, not a contractual speed test.

6. Edge network

When the page loads, your browser measures the responsiveness of every edge node and the nearest one is selected automatically; a manual choice in the header is remembered. Server-side tools (traceroute, DNS, port checks) run from our measurement network, so they show the path from our vantage point rather than from your home.

7. Games and video calls

Our network measures Valve (official Steam relay points), Riot, Blizzard and AWS, as well as Zoom, Microsoft Teams and Google Meet. Results are briefly shared between visitors, so a node cannot become a traffic amplifier. "Estimate for you" adds your round trip to the edge node and the node's round trip to the service: an upper bound when your route does not pass through the node. HTTPS services can also be timed directly from your browser.

8. ISP index

  • Contributions: only when you tick "Contribute" before a test. The browser sends the median latency, jitter, loss, bufferbloat grade and node. The server resolves your IP address to a network (ASN) in memory and discards the address.
  • Providers: Telekom Slovenije (AS5603), Telemach Slovenija (AS3212), A1 Slovenija (AS21283, AS8591, AS207746), T-2 (AS34779) and ARNES (AS2107, reference only). The prefix list is refreshed from RIPEstat. Other networks are not stored.
  • Medians: every contribution immediately updates hourly and daily summaries, from which medians and 90th percentiles are computed to within about ±5 %.
  • Ranking: a provider is ranked once it has at least 30 contributions on the same node in the selected window.

Gaming score (0-100):

score = 100 − 0.5 · max(0, median − 10 ms)
            − 0.25 · bufferbloat Δ
            − 2 · median jitter
            − 10 · mean loss (%)

Limits: only some users contribute and many test over Wi-Fi, so the index is not a representative sample. The server rejects implausible values and limits contributions per address, but it cannot prove a browser reported truthfully. That is why medians and counts are published.

9. AKOS complaint prep

Following ZPS guidance (which summarises AKOS), proving a shortfall takes at least five measurements on five consecutive days, at least one hour apart, at different times of day, over a wired connection. The tool checks this in Slovenian time (CET/CEST), issues a verification code and keeps the summary and a SHA-256 fingerprint of the runs. The runs themselves stay in your browser.

This record is preliminary technical evidence for your operator. For the official procedure, use AKOS Test Net at akostest.net.

10. Data and privacy

  • IP addresses are not written down. The application handles your address in memory only: to resolve the network and for rate limiting (at most 10 minutes). Web server logs keep a shortened address only (IPv4 /24, IPv6 /48).
  • Index contributions: raw rows without an IP address 30 days, hourly sums 90 days, daily sums indefinitely.
  • AKOS records: code, provider, period, summary and fingerprint for 400 days.
  • Share links: 180 days. Analytics: only with your consent.