date-time-wars · itu-1.16.0
69 benchmarks · 10 candidates · ns/op, lower is better · 25.0.4 · 13th Gen Intel(R) Core(TM) i9-13900H · 2026-09-21
Summary
ITU against JDK java.time, one row per operation. Spans cover the inputs of that operation; the speed-up is JDK java.time time divided by ITU time, per input. The other candidates are in the sections below.
| Operation | What | Inputs | ITU | JDK java.time | Speed-up | Fastest overall |
|---|---|---|---|---|---|---|
| Date-time parsing | ||||||
| parse | strict RFC-3339 / ISO-8601 date-time | 3 | 18–28 | 542–577 | 21–30× | ITU |
| parseLenient | lenient date-time (optional fields) | 3 | 11–17 | 755–788 | 46–67× | ITU (char[] buffer) |
| isValid | strict RFC-3339 validity check, valid and invalid inputs | 7 | 2–13 | 536–2,397 | 43–669× | ITU |
| Date-time formatting | ||||||
| formatSeconds | format UTC, second resolution | 1 | 16 | 100 | 6.3× | ITU |
| formatMillis | format UTC, millisecond resolution | 1 | 19 | 115 | 6.1× | ITU |
| formatNanos | format UTC, nanosecond resolution | 1 | 33 | 115 | 3.5× | ITU |
| Duration | ||||||
| parseDuration | ISO-8601 duration | 2 | 25–52 | 134–247 | 4.8–5.4× | ITU |
| formatDuration | ISO-8601 duration | 2 | 14–27 | 10–31 | 0.7–1.2× | ITU |
Times in ns/op. “Fastest overall” counts every candidate except the reference rows, which are not date-time parsers.
Candidates
| Candidate | Library | Version | Under test |
|---|---|---|---|
| ITU | ITU - Internet Time Utility | 1.16.0 | ITU.parseDateTime, parseLenient, isValid, formatUtc*, parseDuration and Duration.normalized() |
| ITU (configurable) | ITU - Internet Time Utility | 1.16.0 | ConfigurableDateTimeParser with a token layout of yyyy-MM-ddTHH:mm:ss[.fff]+offset, the same inputs as the fixed parser |
| ITU (char[] buffer) | ITU - Internet Time Utility | 1.16.0 | ITU.parseLenient over a char[] window into a reusable MutableDateTimeBuffer - the zero-allocation path |
| ITU (hours) | ITU - Internet Time Utility | 1.16.0 | Duration.normalized(HOURS), so it decomposes a duration the way java.time.Duration.toString() does |
| JDK java.time | JDK java.time | 25.0.4 | OffsetDateTime.parse, a DateTimeFormatterBuilder layout with optional parts for lenient parsing, parse-and-catch for isValid, Duration.parse / toString |
| JDK Instant | JDK java.time | 25.0.4 | Instant.parse - the fastest built-in path for RFC-3339 input |
| Google HTTP client | Google HTTP client | 1.46.1 | DateTime.parseRfc3339ToSecondsAndNanos |
| Epoch millisreference | JDK java.time | 25.0.4 | Long.parseLong / Long.toString on the epoch count - the number an API would carry instead of a date-time string. Not a parser; a reference for the “epoch is faster” argument |
| ITU (epoch millis)reference | ITU - Internet Time Utility | 1.16.0 | ITU.parseEpochMilli on the same epoch-millis text, into the same DateTime the string parsers produce - the epoch argument measured to a temporal value, not to a long |
| ITU (epoch millis, char[] buffer)reference | ITU - Internet Time Utility | 1.16.0 | ITU.parseEpochMilli over a char[] window into a reusable MutableDateTimeBuffer - the zero-allocation path for epoch text, the like-for-like of the char[] buffer row |
Date-time parsing
parse strict RFC-3339 / ISO-8601 date-time
ITUITU (configurable)JDK java.timeJDK InstantGoogle HTTP client
| Input | ITU | ITU (configurable) | JDK java.time | JDK Instant | Google HTTP client |
|---|---|---|---|---|---|
| 2023-01-01T23:38:34.987654321+06:00 | 27.7 ±0.4fastest96 B/op-15.3% vs baseline | 79.4 ±2.12.87×248 B/opnew | 577 ±17.420.84×1,592 B/opnew | 437 ±7.415.79×1,760 B/opnew | 674 ±20.524.35×1,572 B/opnew |
| 3074-07-01T12:02:01Z | 18.3 ±0.3fastest96 B/op-8.9% vs baseline | 60.4 ±1.33.30×224 B/opnew | 542 ±9.929.55×1,480 B/opnew | 407 ±6.222.19×1,568 B/opnew | 499 ±11.727.24×1,272 B/opnew |
| 5050-01-01T12:02:01.123Z | 20.0 ±0.3fastest96 B/op-24.0% vs baseline | 63.4 ±0.93.17×224 B/opnew | 549 ±10.827.46×1,536 B/opnew | 433 ±6.921.65×1,680 B/opnew | 587 ±15.929.34×1,480 B/opnew |
parseLenient lenient date-time (optional fields)
ITUITU (char[] buffer)JDK java.timeEpoch millisITU (epoch millis)ITU (epoch millis, char[] buffer)
| Input | ITU | ITU (char[] buffer) | JDK java.time | Epoch millis | ITU (epoch millis) | ITU (epoch millis, char[] buffer) |
|---|---|---|---|---|---|---|
| 2023-01-01T23:38:34.987654321+06:00 | 17.0 ±0.31.40×80 B/op-30.9% vs baseline | 14.8 ±0.21.22×0 B/opnew | 787 ±10.664.87×2,808 B/opnew | 12.1 ±0.2fastest0 B/opnew | 18.6 ±0.31.53×56 B/opnew | 17.5 ±0.31.45×0 B/opnew |
| 3074-07-01T12:02:01Z | 11.3 ±0.21.24×56 B/op-18.8% vs baseline | 9.11 ±0.1fastest0 B/opnew | 755 ±15.682.86×2,616 B/opnew | 12.8 ±0.21.40×0 B/opnew | 19.0 ±0.22.09×56 B/opnew | 18.9 ±0.32.07×0 B/opnew |
| 5050-01-01T12:02:01.123Z | 12.8 ±0.21.27×56 B/op-37.9% vs baseline | 10.0 ±0.1fastest0 B/opnew | 788 ±14.178.50×2,784 B/opnew | 12.8 ±0.21.27×0 B/opnew | 19.1 ±0.31.91×56 B/opnew | 18.9 ±0.21.88×0 B/opnew |
isValid strict RFC-3339 validity check, valid and invalid inputs
ITUJDK java.time
| Input | ITU | JDK java.time |
|---|---|---|
| 2023-01-01T23:38:34.987654321+06:00 | 12.9 ±0.2fastest0 B/op-60.5% vs baseline | 559 ±11.343.22×1,592 B/opnew |
| 3074-07-01T12:02 | 1.59 ±0.0fastest0 B/op-99.8% vs baseline | 1,065 ±23.4669.26×1,640 B/opnew |
| 3074-07-01T12:02:01Z | 7.68 ±0.1fastest0 B/op-62.0% vs baseline | 545 ±10.170.93×1,480 B/opnew |
| 3074-07-01T12:02:01Zjunk | 7.75 ±0.1fastest0 B/op-99.2% vs baseline | 1,253 ±21.8161.73×2,176 B/opnew |
| 3074-13-01T12:02:01Z | 6.94 ±0.1fastest0 B/op-99.3% vs baseline | 2,397 ±39.3345.41×3,596 B/opnew |
| 5050-01-01T12:02:01.123Z | 10.5 ±0.1fastest0 B/op-56.8% vs baseline | 536 ±14.551.22×1,536 B/opnew |
| not-a-date | 1.60 ±0.0fastest0 B/op-99.9% vs baseline | 809 ±11.9506.43×1,040 B/opnew |
Date-time formatting
formatSeconds · formatMillis · formatNanos one operation per row
ITUJDK java.timeEpoch millis
| Operation | ITU | JDK java.time | Epoch millis |
|---|---|---|---|
| formatSeconds | 15.9 ±0.21.81×136 B/op-0.1% vs baseline | 100 ±1.411.40×256 B/opnew | 8.80 ±0.1fastest56 B/opnew |
| formatMillis | 19.0 ±0.21.65×144 B/op-0.4% vs baseline | 115 ±2.610.01×256 B/opnew | 11.5 ±0.7fastest56 B/opnew |
| formatNanos | 33.4 ±0.61.91×160 B/op-0.0% vs baseline | 115 ±1.76.59×264 B/opnew | 17.5 ±0.5fastest64 B/opnew |
Duration
parseDuration ISO-8601 duration
ITUJDK java.time
| Input | ITU | JDK java.time |
|---|---|---|
| -P180DT23H27M19.193964536S | 51.6 ±0.9fastest80 B/op+1.1% vs baseline | 247 ±22.64.79×312 B/opnew |
| PT2H30M | 24.7 ±0.4fastest80 B/op-0.1% vs baseline | 134 ±3.05.45×264 B/opnew |
formatDuration ISO-8601 duration
ITUITU (hours)JDK java.time
| Input | ITU | ITU (hours) | JDK java.time |
|---|---|---|---|
| -P180DT23H27M19.193964536S | 27.1 ±0.4fastest168 B/op-0.1% vs baseline | 32.3 ±0.51.20×160 B/opnew | 31.2 ±0.61.15×184 B/opnew |
| PT2H30M | 13.8 ±0.31.40×144 B/op+1.3% vs baseline | 12.8 ±0.21.29×144 B/opnew | 9.87 ±0.1fastest88 B/opnew |
Other
parseLenientToEpochSecond
ITU (char[] buffer)
| Input | ITU (char[] buffer) |
|---|---|
| 2023-01-01T23:38:34.987654321+06:00 | 17.6 ±0.3fastest0 B/opnew |
| 3074-07-01T12:02:01Z | 11.7 ±0.2fastest0 B/opnew |
| 5050-01-01T12:02:01.123Z | 13.3 ±0.2fastest0 B/opnew |
ns/op per operation; ± is JMH’s 99.9% confidence interval over all measurement iterations. The small figure under each value is the ratio to the fastest candidate on that input.
Environment
Date2026-09-21T18:17:22+02:00
Runitu-1.16.0
JDK25.0.4 (OpenJDK 64-Bit Server VM 25.0.4+7-1-26.04-Ubuntu)
JVM args-XX:+UnlockDiagnosticVMOptions -XX:+TieredCompilation -XX:+AlwaysPreTouch
Iterations2 fork(s), 3 × 1 s warmup, 10 × 1 s measurement (--thorough)
CPU13th Gen Intel(R) Core(TM) i9-13900H
OSUbuntu 26.04.1 LTS
Git5490c17
Harness floor0.54 ns/op - the JMH loop, state loads and Blackhole with no parsing at all; subtract it from a parse score for the parser's own cost
JMH 1.37. Each run records this in run.properties next to its results; library versions are read from the jar that ran, not from the pom.