Two Hosts, Same Average, Different Experience
You ping two servers. Both report an average of 40 ms.
The first returns 39, 40, 41, 40, 39, 40, 41 — a spread of 2 ms. The second returns 12, 78, 31, 95, 22, 64, 18 — a spread of 83 ms.
Identical average. The first is fine for a video call. The second will produce audible dropouts, stuttering, and the sense that "the connection is bad" despite a perfectly reasonable mean.
Average latency is the least informative number ping gives you. The distribution is where the diagnosis is.
Reading the Full Output
A standard ping run reports minimum, average, maximum, and often standard deviation or mean deviation. Each tells you something different.
Minimum is the closest thing to your physical floor — the best-case path with no queuing. It's set largely by distance and hop count, and it barely changes. If your minimum is 140 ms to a server on another continent, that's the speed of light in fibre plus routing, and no amount of troubleshooting will improve it.
Average is the number everyone quotes and the one that hides the most.
Maximum is the worst case in your sample. A maximum far above the average means something intermittent is happening — a congested link, a wireless retransmission, a router under load.
The spread between minimum and maximum is the single most useful derived figure. A tight spread means a stable path. A wide one means variable queuing somewhere.
The Ping Tool reports reachability and response times:
- Enter the hostname or IP.
- Read the individual response times, not just the summary.
- Note the spread between fastest and slowest.
Jitter Is What Breaks Real-Time Traffic
Jitter is the variation in latency between consecutive packets. It's what the min-max spread is really telling you about.
Why it matters more than average latency for interactive traffic:
Voice and video use a jitter buffer. The receiver holds incoming packets briefly and plays them out at a steady rate, smoothing over variation. The buffer has to be sized for the expected jitter — bigger buffer, more tolerance, more added delay.
High jitter forces a larger buffer, which adds latency. Or the buffer underruns and you get a gap in the audio. Either way, jitter costs you.
Rough tolerances for real-time media:
- Under 20 ms jitter: generally fine
- 20–40 ms: noticeable on demanding calls
- Above 40 ms: audible artefacts likely
Gaming is similar but less forgiving, because there's no buffer to hide behind — a late packet is a late input.
Bulk transfers barely care. Downloading a file, jitter is irrelevant; throughput and packet loss matter.
Packet Loss Punishes Disproportionately
Ping reports loss as a percentage, and small percentages do more damage than they sound like.
TCP interprets loss as congestion. Its congestion control responds by reducing the sending rate. Even 1% loss can substantially reduce achievable throughput on a long-distance connection, because every loss event triggers a backoff and the recovery takes a full round trip.
The effect compounds with latency: the higher the round-trip time, the longer each recovery takes, and the more throughput a given loss rate costs.
UDP doesn't retransmit, so loss shows up directly as missing audio frames or dropped game state.
Rough guidance:
- 0%: what you want
- Under 1%: tolerable for most uses, noticeable on voice
- 1–2.5%: degraded voice quality, reduced TCP throughput
- Above 5%: seriously impaired
Loss patterns matter too. Ten scattered single losses are less damaging than one burst of ten consecutive losses, which produces an audible gap rather than a recoverable blip.
Ping Under Load: Finding Bufferbloat
The single most revealing ping test isn't on an idle connection. It's during a heavy transfer.
Start a large download or upload, then run a ping in parallel. If latency climbs from 20 ms to several hundred milliseconds or more, you've found bufferbloat — oversized buffers in a router or modem filling up and adding queuing delay.
This is why a connection can pass a speed test with excellent numbers and still feel terrible. Throughput is fine; latency under load is not. It's the reason a video call degrades the moment someone in the house starts a large upload.
The pattern to look for:
Idle: 18, 19, 18, 20, 19 ms
Under load: 22, 340, 780, 1100, 950 ms
Mitigations involve queue management on the router — modern queuing disciplines specifically address this. It's not something you fix on the endpoint.
The Physical Floor
Light travels through optical fibre at roughly two-thirds of its vacuum speed, which works out to about 5 microseconds per kilometre.
For a round trip, double the distance:
London → New York: ~5,600 km → ~56 ms round trip minimum
London → Sydney: ~17,000 km → ~170 ms round trip minimum
And that's a straight-line cable path with zero processing delay. Real routes aren't straight, and every router adds a small amount.
This is why your minimum ping is what it is. If you're 8,000 km from a server, you cannot get below roughly 80 ms no matter what you buy. Anything advertised as reducing that distance-based latency is either moving the server closer or being misleading.
The practical use: knowing your theoretical floor tells you how much of your observed latency is avoidable. A 200 ms ping to a server 1,000 km away has around 190 ms of avoidable delay somewhere. A 200 ms ping to a server 15,000 km away is close to optimal.
What Ping Can't Tell You
A successful ping doesn't mean a service is working. ICMP is handled by the network stack, not your application. A server can respond to ping perfectly while the web server is down, the database is unreachable, or the application is returning errors.
A failed ping doesn't mean a host is down. ICMP is routinely blocked by firewalls and rate-limited by routers. Many hosts deliberately don't respond, and cloud security groups frequently block it by default.
Ping doesn't test the path your traffic takes. ICMP may be routed differently or treated with different priority than TCP traffic. A host that pings well can still have poor TCP performance on the same path.
Ping doesn't identify where a problem is. A high latency figure tells you something is slow between you and the destination. Traceroute is the tool for locating it.
A Diagnostic Sequence
When something feels slow:
1. Ping the destination. Note min, avg, max and loss — not just the average.
2. Ping something nearby — your default gateway, then your ISP's first hop. This separates "my local network" from "the internet."
3. Ping a known-good reference like a major public DNS resolver. If that's clean and your target isn't, the problem is specific to the target or its path.
4. Ping under load. Start a transfer and watch what happens to latency.
5. Run traceroute if latency is high and you need to know where.
6. Test the actual service, not just ICMP. A TCP connection to the relevant port tells you whether the service is listening.
Step 2 is the one that saves the most time. If your gateway pings at 80 ms with loss, nothing beyond it is worth investigating yet.
Practical Tips
Run enough packets. Four packets tells you almost nothing about jitter. Twenty or more gives you a distribution.
Compare against a baseline. Ping your key destinations when things are working and record the numbers. A latency figure only means something relative to normal.
Ping the hostname, then the IP. If the hostname is slow to resolve but the IP pings fine, your problem is DNS, not the network.
Watch for asymmetry. Ping measures round trip. A problem can exist in only one direction, which ping can't distinguish.
Don't over-interpret a single spike. One slow packet in fifty is normal. A pattern is a pattern.
FAQ
What's a good ping time? Depends entirely on distance. Under 30 ms to a nearby server is good; 150 ms to the other side of the world may be optimal. Compare against the physical floor for that distance.
What is jitter? Variation in latency between consecutive packets. It matters more than average latency for voice, video and gaming.
How much packet loss is acceptable? Ideally zero. Under 1% is tolerable for most uses; above that, voice quality and TCP throughput degrade noticeably.
Why does my ping increase when I'm downloading? Bufferbloat — oversized buffers filling up and adding queuing delay. It's a router configuration issue, not a bandwidth one.
Does a failed ping mean the server is down? No. ICMP is frequently blocked or rate-limited. Test the actual service port instead.
Why can't I get below 150 ms to a server on another continent? The speed of light in fibre. Distance sets a hard floor that no equipment can beat.
The Takeaway
Ping's average is the least useful number it produces. The spread tells you about jitter, the loss percentage punishes throughput far beyond its face value, and the minimum tells you what's physically possible on that path. Running it under load rather than idle is often what finds the actual problem.
Check host reachability and response times free with the Ping Tool at sadiqbd.com — no sign-up, instant results.
