Measure network jitter from ping output: PDV pseudo-range (RFC 5481), de-jitter buffer size, and the ITU-T Y.1541 and G.1010 limits it meets.
Illustrative series showing the shape each link type tends to have — not measurements of a named network.
Paste ping, tracert or a bare list. Timing fields are read directly, so byte counts and addresses are ignored. At least 2 samples; 20 or more for a usable quantile.
Ping reports round-trip time. Every ITU limit is one-way, so each sample is halved before it is compared — that assumes a roughly symmetric path.
One-way delay and delay variation are ITU-T G.1010 Tables I.1 and I.2, and the variation column is what the application must be presented with AFTER buffering — which is why it is far tighter than anything a network delivers. The network budget a de-jitter buffer is sized to absorb is the ITU-T Y.1541 Class 0 and 1 IPDV bound of 50 ms. Where G.1010 prints "N.A." no variation target exists, so only delay is tested. The last column fills in once you paste a measurement.
| Application | One-way delay | Delay variation | This link |
|---|---|---|---|
| Conversational voice | 150 preferred, 400 limit | 1 ms | Not measured |
| Videophone | 150 preferred, 400 limit | Not applicable | Not measured |
| Voice messaging | 1,000 preferred, 2,000 limit | 1 ms | Not measured |
| Streaming audio | 10,000 ms | 1 ms | Not measured |
| Interactive games | 200 ms | Not applicable | Not measured |
| Command and control | 250 ms | Not applicable | Not measured |
ITU-T Y.1541 (12/2011), Table 1. Classes 2 to 5 leave IPDV unspecified. Y.1541 §5.3.1 notes that evaluating the IPDV bound needs a sample of at least 1000 packets.
| Parameter | Nature of the objective | Class 0 | Class 1 |
|---|---|---|---|
| IPTD | Upper bound on the mean one-way transfer delay | 100 ms | 400 ms |
| IPDV | Upper bound on the 1 − 10⁻³ quantile of delay minus the minimum delay | 50 ms | 50 ms |
| IPLR | Upper bound on the packet loss probability | 1 × 10⁻³ | 1 × 10⁻³ |
An estimate from the samples you pasted, not a network audit. Conditions change by the minute, so measure again before acting on a borderline verdict.
You might also find these calculators useful
Calculate network latency: propagation, transmission, and processing delays
See what your packet loss does to throughput, calls, video and games
Find the bandwidth-delay product and the TCP window a link needs
How much internet speed you need, and how long a transfer takes
Jitter is the variation in packet delay, and it is what makes a call break up while a speed test still reads fast. Paste a ping run and this calculator reports the two delay-variation metrics the IETF actually defines, the de-jitter buffer size that variation implies, and whether the result sits inside the ITU-T limits written for voice, video and gaming. Most guidance you will find online quotes a single "30 ms" threshold with no source. The real answer has two numbers at two different points in the chain, and this page shows both.
Latency is how long a packet takes; jitter is how much that time changes from packet to packet. RFC 5481 defines two distinct measures. IPDV compares each packet with the one before it, so it captures packet-to-packet roughness — but its signed mean is usually zero, which makes it awkward to summarise. PDV compares each packet with the fastest packet in the run, so every value is positive and its high percentiles describe the worst case a receiver has to absorb. PDV is the one standards bodies set objectives against, and the one that tells you how large a de-jitter buffer must be.
PDV and IPDV
Measure during a bad call rather than after it. A PDV pseudo-range over 50 ms with normal latency points at congestion or wireless interference, not at bandwidth.
The buffer has to cover the variation without adding delay the codec cannot afford. This page reports both the buffer size and the total one-way delay it produces.
Naming the metric, the quantile and the sample size turns "my internet is laggy" into a measurement a provider can reproduce and check against Y.1541.
Ping several endpoints and compare pseudo-ranges. The steadiest link often beats the nominally closest one, and the application table shows which limits each clears.
Bandwidth and delay variation are independent. A gigabit link with a congested hop delivers packets in bursts, and a codec that needs them evenly spaced cannot recover the timing.
RFC 5481 §3.2 puts it plainly: the buffer must be big enough to absorb the variation, and too large a buffer throws away interactivity. The high PDV percentile is the number that decision is made from.
ITU-T Y.1541 states its IPDV objective as a quantile of delay minus the minimum delay, so a claim about jitter is only checkable if it names the quantile and the interval.
RFC 5481 notes the signed mean IPDV is normally zero. When it is not, the delay is trending across the run or the measurement clocks disagree — a different problem with a different fix.
There are two published answers because there are two points in the chain. Across the network, ITU-T Y.1541 Table 1 bounds IPDV at 50 ms for its Classes 0 and 1 — measured as the 99.9th percentile of delay minus the minimum delay. What reaches the codec after buffering is far tighter: ITU-T G.1010 Table I.1 gives conversational voice a delay-variation target under 1 ms. The de-jitter buffer is what turns the first into the second. The widely repeated "30 ms" figure is vendor design guidance, not a published objective, and it sits between the two without matching either.
Latency is the time a packet takes; jitter is how much that time varies. They move independently. A satellite link can have 600 ms of latency and almost no variation, which is fine for streaming and hopeless for conversation. A fibre link can have 8 ms of latency and 60 ms of variation during a congested minute, which is the opposite problem. Real-time applications have separate budgets for each, which is why this page reports both.
PDV, for almost anything you would use this page for. RFC 5481 defines IPDV against the previous packet and PDV against the fastest packet in the run. Because IPDV can be negative and its mean is usually zero, it does not summarise well. PDV is always positive, its percentiles describe the worst case a receiver must absorb, and it is the form ITU-T Y.1541 sets its objective against. IPDV is still useful as a texture measure and as a validity check: a signed mean far from zero means the delay is drifting or the clocks disagree.
Because it has not finished converging. RFC 3550 §6.4.1 defines J as an exponential average with a gain of one sixteenth, starting from zero, so after n packets it has only reached 1 − (15/16)^(n−1) of its settled value: about 44 per cent after 10 packets and 96 per cent after 50. An RTP receiver runs for the whole call and is fully converged; a pasted ping run is not. RFC 3550 §6.4.4 is explicit that the field "is only a snapshot ... and is not intended to be taken quantitatively", which is why this page shows it alongside how far it has converged rather than as the answer.
Yes, unedited, from any platform or system language. The parser reads the timing fields — "time=" on English systems, "tiempo=" on Spanish ones — and ignores everything else, so byte counts, addresses and TTL values cannot be mistaken for delays. If a paste looks like tool output but has no timing fields, the page says so instead of computing something. A plain list of numbers also works, separated by commas, spaces or newlines.
Because ping measures a round trip and every ITU limit on this page is one-way. G.1010 gives conversational voice 150 ms one-way preferred, and Y.1541 bounds mean one-way delay at 100 ms for Class 0. Comparing a round-trip figure against a one-way limit fails links that are comfortably fine — a 200 ms round trip is roughly 100 ms each way. Halving assumes the path is symmetric, which is usually close enough and occasionally not; switch the basis to one-way if you have measurements from a tool that timestamps at both ends.
A de-jitter buffer holds arriving packets briefly so they can be released at even intervals. RFC 5481 §3.2 frames the trade-off: it must be large enough to absorb the delay variation, and if it is too large the interactivity you wanted is lost anyway. §7 names the 99.9th percentile of PDV as the figure used in buffer design, which is what this page reports as the pseudo-range. Its cost shows up as added one-way delay, so the application table tests the total rather than the raw measurement.
More than most guides suggest. Estimating a percentile p needs roughly 1/(1 − p) samples, so the 95th percentile needs about 20 and the 99.9th that Y.1541 uses needs at least 1000 — Y.1541 §5.3.1 says so directly, and warns that using a lower quantile instead "will result in under-estimates of de-jitter buffer size". This page names the quantile it could support and whether that is Y.1541-comparable, rather than quietly reporting a lower one as though it met the objective.
Often, yes. Wired beats wireless, because most home variation is radio contention. Quality-of-service marking helps when the bottleneck is your own uplink and a large upload is competing with real-time traffic. Changing endpoint or region can avoid a congested path entirely. Where the variation is upstream and outside your control, a larger de-jitter buffer trades delay for smoothness, which is a mitigation rather than a fix.