Ping Test

Ping any host or IP from probes on every continent and see round-trip time, packet loss and jitter for each one, on a map. ICMP or TCP, IPv4 or IPv6.

Tool Networking Updated Oct 6, 2026
Learn how this works
How to Use
  1. Type a hostname such as example.com or an IP address. Private addresses (192.168.x.x, 10.x.x.x) cannot be reached from the internet, so the tool says so instead of sending them.
  2. Choose how many probes, where they are (Every continent spreads them out) and how many packets each sends (3 to 16).
  3. Leave the protocol on ICMP for a classic ping. Pick TCP and a port (443 for websites) when the host blocks ICMP: the probes then time a TCP handshake instead.
  4. Press Run live ping. Rows fill in as each probe finishes, usually within 5 seconds. Until you press it, the page shows a recorded example.
  5. Click a row to see that probe’s numbers worked out in Show Work: loss, average, jitter, the reply TTL and how far away the replying server can be.
Target
IPv4 · IPv6 · name

A live run uses 8 of the 250 free Globalping tests per hour for your IP address.

Presets
Probes on the map

Probes replying
—
Fastest average
—
Slowest average
—
Packet loss
—
Per-probe results
#ProbeResultMin / Avg / MaxLossJitterReply TTLReplied fromOutput
Live checks run on the Globalping probe network (by jsDelivr), called straight from your browser. Each probe counts as one test; the free limit is 250 tests per hour per IP address.

Worked Example

1.1.1.1 from eight probes, 5 packets each (recorded 6 October 2026). Every probe got 5 replies out of 5, so loss is (5 − 5) ÷ 5 = 0 %. The averages ranged from 0.992 ms in Los Angeles to 11.577 ms in Buffalo.

Jitter from Lagos. The five round trips were 3.20, 1.07, 3.22, 0.983 and 1.16 ms. The changes from one reply to the next are 2.13, 2.15, 2.237 and 0.177 ms. Their sum is 6.694 ms, and 6.694 ÷ 4 = 1.67 ms of jitter (inter-packet delay variation, RFC 5481). Helsinki’s five replies (1.29, 1.29, 1.35, 1.33, 1.33 ms) give only 0.02 ms.

How far away can the server be? A reply cannot travel faster than light, 299.792458 km per millisecond. Tokyo’s fastest reply took 1.693 ms, so the server is at most 1.693 ÷ 2 × 299.79 = 254 km from that probe. São Paulo’s took 0.805 ms: at most 121 km away.

The common mistake: reading 1.1.1.1 as one server that is “close to everyone”. Tokyo and São Paulo are 18,534 km apart, yet the two light-speed circles add up to only 254 + 121 = 375 km. No single machine can sit inside both, so the same address is answered by different servers in different cities. That is anycast. Likewise, the 11.6 ms from Buffalo is not the distance to “Cloudflare”, only to the nearest copy of 1.1.1.1 that Buffalo’s route reaches.

Show Work

Run a check to see how the result was produced.

Formulas

Packet loss
loss = (sent − received) ÷ sent × 100 %
5 sent, 4 replies: (5 − 4) ÷ 5 = 20 %.
Average round-trip time
avg = (RTT1 + … + RTTn) ÷ n
Only replies count; a lost packet has no RTT.
Jitter (IPDV, RFC 5481 §4.1)
J = mean of |RTTi − RTTi−1|
Each reply compared with the one before it. RFC 5481 defines it on one-way delays; on round trips it mixes both directions.
Delay variation (PDV, RFC 5481 §4.2)
PDVmax = RTTmax − RTTmin
Every reply compared with the fastest one.
Farthest possible server
d ≤ RTTmin ÷ 2 × 299.792458 km/ms
Light speed is exact by the SI definition of the metre; real fibre routes are slower and longer, so the true distance is shorter still.
Anycast proof
distance(A, B) > dA + dB
Two probes that reached the same address but are farther apart than their two light-speed limits added together reached two different machines.
Routers on the way back
hops ≈ 64 − reply TTL
Each router lowers TTL by 1 (RFC 791). 64 is IANA’s recommended starting value; a system that starts higher makes this an estimate only.

What ICMP Is and Where Ping Came From

ICMP, the Internet Control Message Protocol, was specified by Jon Postel in RFC 792 in September 1981. It carries IP’s error and test messages: Echo Request is type 8 and Echo Reply type 0; Destination Unreachable is type 3 and Time Exceeded type 11. IPv6 has its own version, ICMPv6 (RFC 4443, March 2006), where Echo Request is type 128 and Echo Reply type 129. RFC 1122 says every host must answer Echo Requests, but firewalls in front of hosts often drop them anyway.

The ping program was written by Mike Muuss in December 1983, when he ran into odd behaviour on the network at BRL, where he worked. That July, at a DARPA meeting in Norway, Dave Mills had described timing ICMP Echo packets on his “Fuzzball” systems. Muuss named the program after the sound of sonar, wrote it in an evening, and Berkeley made it a standard part of BSD Unix; it has since been ported to many systems, Windows included.

Reading the numbers. Round-trip time is mostly distance: a reply can never beat light speed, and real fibre is slower and takes longer routes. Loss means Echo Requests or Replies went missing; a little on one probe points at that path, total loss on every probe means the host is down or filters ICMP. Routers may also rate-limit the ICMP they generate (RFC 1812 §4.3.2.8), so a busy router can drop pings while forwarding real traffic normally. Jitter measures how steady the delay is, which matters for voice and games more than the average does.

TCP ping. When ICMP is blocked, the probe can time the TCP three-way handshake instead (RFC 9293 §3.5): its SYN goes to an open port and the server’s SYN,ACK comes back, one round trip. A closed port answers with a reset and a filtered one does not answer, so both show as loss.

About This Tool

This ping test sends your target to the Globalping network, a public set of probes in data centres and home connections run by jsDelivr, and shows each probe’s replies. You choose the number of probes (3 to 20), where they are (every continent, one continent, or a custom country, city, network or cloud region), 3 to 16 packets, ICMP or TCP, and IPv4 or IPv6.

Each row gives the probe’s location and network, minimum, average and maximum round trip, loss, jitter, the TTL of the replies and the address that answered; open Output to read the probe’s raw ping output. Click a row to have Show Work explain it, and to highlight that probe on the map. The tool also checks the whole run for anycast with the light-speed test above.

Privacy: the request goes from your browser to api.globalping.io, not to this site, and only when you press Run. Addresses in private, loopback, documentation and carrier-grade NAT ranges (IANA special-purpose registry) are refused before anything is sent, because no probe on the internet can reach them. Sources: RFC 792, RFC 4443, RFC 1122, RFC 1812, RFC 5481, RFC 9293 (rfc-editor.org), the IANA IP Parameters and special-purpose address registries, Mike Muuss’s “The Story of the PING Program”, and Natural Earth 1:110m land data for the map.

Related tools: Traceroute, Open Port Checker, and DNS Lookup.

Frequently Asked Questions

What is a good ping time?

It depends on distance more than anything else. Light needs about 80.7 ms just to cross the 12,093 km between Sydney and Los Angeles and come back, so a ping between those two cities can never beat that. A 20 ms round trip puts the server within 20 ÷ 2 × 299.79 = 2,998 km of the probe at the very most; 1 ms puts it within 150 km, which is why 1.1.1.1 answering Los Angeles in 0.99 ms and Sydney in 1.42 ms means there is a copy of it near each city.

What does packet loss mean in a ping test?

Loss is the share of Echo Requests that got no reply: (sent − received) ÷ sent × 100. 1 lost reply out of 5 is 20 %. A few percent on one probe usually points at that probe’s path; 100 % on every probe means the host is down or does not answer ping. Many hosts drop ICMP on purpose, so try the TCP option before deciding a site is down.

How is jitter calculated?

This tool uses inter-packet delay variation from RFC 5481: the change in round-trip time from one reply to the next, averaged. Lagos replied in 3.20, 1.07, 3.22, 0.983 and 1.16 ms; the changes are 2.13, 2.15, 2.237 and 0.177 ms, so jitter is 6.694 ÷ 4 = 1.67 ms. The spread from fastest to slowest (max − min = 2.24 ms) is a different measure, packet delay variation.

Why does a ping fail when the website still loads?

Ping uses ICMP Echo (RFC 792), and firewalls often block it while still allowing web traffic on TCP port 443. A failed ICMP ping only says the Echo Requests got no answer. Switch the protocol to TCP with port 443: if the handshake completes, the server is up and simply ignores ping.

Does this ping come from my computer?

No. Browsers cannot send ICMP packets. When you press Run, your browser asks the Globalping network (a public probe network run by jsDelivr) to ping the target from its probes, then shows their results. Nothing is sent to this site, and a plain visit sends nothing at all: the table you first see is a recorded example.

Common Use Cases

Is it down, or just me?

Ping from 10 probes on several continents. If all 10 reply, the host is up and the problem is between you and it.

Spot packet loss on one path

One probe at 20 % loss (1 of 5 lost) while the others show 0 % points at that probe’s route, not the server.

Check a CDN or anycast address

1.1.1.1 answered Tokyo in 1.69 ms and São Paulo in 0.81 ms; the cities are 18,534 km apart, so they reached different servers.

Measure jitter before a VoIP or gaming setup

Send 16 packets and compare jitter: 0.02 ms from Helsinki is steady, 1.67 ms from Lagos varies far more.

Ping a host that blocks ICMP

Use TCP on port 443 to time the handshake to a web server that drops ping packets.

CCNA and Network+ practice

See real Echo Request/Reply results, reply TTLs (64 − 58 = 6 routers) and loss figures to match the theory.

Last updated: