01
HOST INPUT
Enter target host
Global Ping Test
Test website latency from 20 locations around the world
HOW IT WORKS
Global latency testing
02
GLOBAL PING
Send pings from 20 locations
03
LATENCY ANALYSIS
Measure response times
04
REPORT
Generate latency report
What Modern Ping Tests Are Really Measuring
The number `ping` returns from a terminal is one millisecond reading. What a global ping test surfaces is a constellation of fifteen to twenty different numbers, each measuring a slightly different thing, against a target that may be five physical servers behind one DNS name. The traditional ping is ICMP echo, defined in RFC 792 from 1981. Browsers can't send ICMP. Modern web ping tests measure TCP handshake time, HTTP request roundtrip, or HTTPS request roundtrip including TLS negotiation. These three numbers can differ by 3x for the same target. Calling all of them 'ping' is a vestigial habit. Each measures a different layer. The probes themselves don't agree because they're hitting different points-of-presence. A target served by Cloudflare or Akamai is technically dozens of servers worldwide, and BGP routing decisions can send a probe in São Paulo to a PoP in Miami when one in São Paulo would be obvious. Sections below: what ping numbers actually mean by protocol, why distance is a poor predictor of latency, and what to read from the variance across probes.
ICMP, TCP, HTTP, HTTPS Ping All Measure Different Things
ICMP echo is the original ping. Send a small packet, get a small packet back, measure round-trip time. Defined in RFC 792 (1981), no ports, runs at the network layer. Many routers, firewalls, and load balancers answer ICMP without ever touching the actual server. AWS Route 53 deliberately drops ICMP at the edge. Cloudflare answers from its anycast layer regardless of what the origin is doing. The ICMP RTT is the network's optimistic floor: how fast the path can carry a packet to a device that's willing to answer. Browser-based latency tools usually mean TCP ping. Open a TCP socket, observe the SYN/SYN-ACK timing, close. One round-trip on the network. Closer to what an actual user request will pay, because TCP has to land on a real server (or its load balancer) to complete the handshake, not just any device that bounces ICMP. TCP ping numbers are typically 10-20% higher than ICMP for the same target. HTTP ping adds a full GET request after the handshake. Two round-trips minimum: TCP handshake (1 RTT), HTTP request and response (1 RTT). HTTPS ping inserts TLS negotiation between them. TLS 1.3 needs 1 RTT for handshake (or 0 RTT for resumed sessions), TLS 1.2 needs 2. A cold HTTPS request to a TLS 1.3 server is 3 RTTs minimum: TCP handshake, TLS handshake, request/response. For a target with 50ms TCP ping, expect 150ms HTTPS ping on a cold connection. With QUIC/HTTP3 the math changes: 1 RTT for cold, 0 RTT for resumed, because QUIC bundles transport and TLS handshakes. For real user experience, HTTPS ping is the first-request metric and HTTP/2 or HTTP/3 ping covers warm subsequent requests. The honest network-quality number is TCP ping. ICMP ping is a network-engineering metric that doesn't correlate well with user experience anymore.
Why Distance Is a Lousy Predictor
Light in vacuum travels 300,000 km/s. Light in fiber, due to refractive index, travels around 200,000 km/s, or two-thirds the vacuum speed. New York to London is 5,571 km of great-circle distance. The theoretical floor for a one-way photon trip is 27.5ms. Round-trip floor: 55ms. Real NYC-London latency in 2026 averages 70-80ms. The 40% overhead comes from: - Routing through major Internet Exchange Points (LINX in London, NYIIX in NYC) instead of point-to-point fiber - TCP queuing delays in routers, especially during peak hours - Submarine cable physical paths that detour around shorter great-circle distances - BGP path selection that picks 'shortest AS path', not 'shortest geographic path' Some routes outperform expectation. The Marea submarine cable runs direct from Virginia Beach to Bilbao, Spain, avoiding the more common UK detour. Latency on Marea-routed traffic is closer to the speed-of-light floor than on traditional Atlantic routes. Some routes underperform massively. Submarine cable cuts in the Red Sea (early 2024, multiple cables severed near Yemen) forced Asia-Europe traffic onto longer alternative paths through Egypt and around Africa. RTT for India-Germany traffic spiked from ~140ms to 280ms+ until repairs completed. The Baltic Sea cuts (Finland-Estonia, December 2024) had similar regional impact. Mobile and satellite latency don't follow distance at all. Starlink averages 30ms regardless of where the user is, because the path is user → satellite → ground station → backbone, and the ground stations are well-positioned. Cellular 4G adds 30-50ms of last-mile overhead independent of distance to the target. 5G NR low-latency mode can shave that to 10-15ms but the rollout is uneven. Distance gives you the floor. Routing gives you the actual number. Two probes in the same city can get 30ms different latencies to the same target if their ISPs route through different upstream providers. Same destination, different highways, different traffic.
Reading the Variance Across 20 Probes
The interesting data isn't any single ping number. It's the spread. Tight clustering across regions (variance under 30ms within a continent) indicates good CDN coverage. The target serves traffic from multiple PoPs, each probe hits a nearby one, results converge. Wide variance within one region indicates routing issues. If ten European probes return latencies between 20ms and 180ms to the same target, BGP is picking different paths from different ISPs, and some of those paths are scenic. Asymmetric regions reveal infrastructure choices. A US-hosted target showing 50ms from US probes and 250ms from Asian probes hasn't deployed in Asia. A target with similar latencies everywhere is on Cloudflare or AWS CloudFront's anycast. Failed probes carry signal. If two probes in the same region time out while others succeed, the target's firewall might be blocking that probe's IP range, or there's a BGP partition. ICMP and TCP probes that succeed where HTTPS probes fail usually mean the WAF is blocking automated requests but the network path is fine. Jitter (standard deviation across multiple pings to the same probe) indicates connection stability. WiFi interference is the most common source of jitter for end-users. Server-side, queuing in overloaded load balancers shows up as jitter. 1-2ms of jitter is normal for fiber; 5-10ms suggests congestion; over 20ms is a connection problem. Packet loss above 0.5% on TCP probes is suspicious. TCP ping should rarely lose packets because the handshake gets retried. Loss usually indicates rate-limiting (the target sees probe traffic as scanning), middlebox interference (a stateful firewall mid-path), or genuinely overloaded infrastructure. Like reading a stethoscope chart, the absolute heart rate matters less than the rhythm. A consistent 80ms across 20 probes is a healthier diagnosis than a 40ms average with five probes timing out. The WHOX ping test runs probes from 20 locations across all major regions, splits results by TCP/HTTP/HTTPS, and surfaces the variance and jitter patterns alongside the headline averages. The patterns are usually more useful than the numbers.
FREQUENTLY ASKED QUESTIONS
What is a good ping latency?
For web browsing, under 100ms is good, under 50ms is excellent. For gaming, under 50ms is ideal, under 20ms is excellent. High latency (>200ms) noticeably affects interactive applications.
Why is latency different from different locations?
Latency depends on physical distance, network routing, and server infrastructure. A server in New York will have lower latency for US users and higher for Asia-Pacific users. CDNs help reduce this by caching content closer to users.
What does packet loss mean?
Packet loss occurs when data packets fail to reach their destination. Even 1-2% loss can degrade performance. High packet loss usually indicates network congestion, poor routing, or hardware issues.
All our tests
Run the full WHOX toolkit from one place: leak checks, fingerprint probes, network diagnostics, reputation lookups, and privacy failure tests.