Benchmarks

Workloads

Specification building

Large-data binning
Small plots

Static export

vl-convert export

Not measured

Specification building

Large-data binning

Results

Binning of one million (x, y) points from a phibble table into hexagons, rectangles, or a histogram, and encoding of the specification as JSON. Only the bins are embedded.

Workload: Large-data binning, 1,000,000 rows, 2 numeric columns
Dataset: 1,000,000 rows

Large-data binning, 1,000,000 rows, 2 numeric columns: elapsed time, peak anonymous memory, peak resident set size, and PHP allocator peak per implementation and case, median of 7 runs.

hex(bins: 40)

hex(bins: 40): elapsed time and memory by implementation
Implementation	Elapsed time	Fastest – slowest run	Peak anonymous memory	Peak resident set size	PHP allocator peak
phiz61 KB specification.	1.67 s	1.67 s – 1.68 s	59.8 MiB	82.9 MiB	56.0 MiB

bin2d(bins: 50)

bin2d(bins: 50): elapsed time and memory by implementation
Implementation	Elapsed time	Fastest – slowest run	Peak anonymous memory	Peak resident set size	PHP allocator peak
phiz168 KB specification.	1.31 s	1.31 s – 1.31 s	60.4 MiB	83.4 MiB	56.0 MiB

histogram(bins: 60)

histogram(bins: 60): elapsed time and memory by implementation
Implementation	Elapsed time	Fastest – slowest run	Peak anonymous memory	Peak resident set size	PHP allocator peak
phiz6 KB specification.	804 ms	803 ms – 807 ms	58.9 MiB	81.9 MiB	56.0 MiB

Methods

Implementations: phiz: phiz v0.5.2-9-g760c476, plain PHP, reading a phibble v0.6.1 table
Machine: 2 AMD Zen 5 cores, 7.8 GB RAM, Linux
Software: PHP 8.5.12, phibble v0.6.1
Threads: 1
Repetitions: 7 timed runs, 0 warm-up runs
Statistic: Median over the timed runs
Time measure: Elapsed time, in s
Timing boundaries: Timed from building the Plot to holding the JSON specification: compilation, columnar reads, binning, and encoding. Table generation and PHP startup excluded.
Time spent in: The PHP process; specification building only, without browser rendering or vl-convert.
Cache: Warm
Memory metric: Peak anonymous memory; Peak resident set size; PHP allocator peak. Peak anonymous memory (RssAnon) of the whole PHP process, including the generated table, sampled every millisecond by peakmem (phpolars bench/csv-pipeline/peakmem.c). Also reported: peak resident set size, and PHP allocator peak (memory_get_peak_usage(true)).
Recorded: 2026-11-01, commit 760c476

Reproduction

From the repository root

composer install
php scripts/bench-large-data.php

Raw record: site/benchmarks/large-data-binning.json
Source: scripts/bench-large-data.php

Small plots

Results

Building of an ordinary plot from a bundled dataset and encoding of its specification as JSON, as for each chart a page serves.

Workload: Small plots, 234 to 574 inline rows

Small plots, 234 to 574 inline rows: elapsed time per plot, peak anonymous memory, peak resident set size, and PHP allocator peak per implementation and case, median of 7 runs.

scatter, 234 rows

scatter, 234 rows: elapsed time per plot and memory by implementation
Implementation	Elapsed time per plot	Fastest – slowest run	Peak anonymous memory	Peak resident set size	PHP allocator peak
phiz35.1 KB specification.	0.367 ms	0.365 ms – 0.368 ms	12.6 MiB	35.6 MiB	8.00 MiB

scatter with lm band, 234 rows

scatter with lm band, 234 rows: elapsed time per plot and memory by implementation
Implementation	Elapsed time per plot	Fastest – slowest run	Peak anonymous memory	Peak resident set size	PHP allocator peak
phiz52.4 KB specification.	0.881 ms	0.880 ms – 0.882 ms	12.6 MiB	35.6 MiB	8.00 MiB

scatter with loess band, 234 rows

scatter with loess band, 234 rows: elapsed time per plot and memory by implementation
Implementation	Elapsed time per plot	Fastest – slowest run	Peak anonymous memory	Peak resident set size	PHP allocator peak
phiz52.4 KB specification.	66.5 ms	66.5 ms – 66.6 ms	16.6 MiB	39.8 MiB	12.0 MiB

histogram, 272 rows

histogram, 272 rows: elapsed time per plot and memory by implementation
Implementation	Elapsed time per plot	Fastest – slowest run	Peak anonymous memory	Peak resident set size	PHP allocator peak
phiz8.8 KB specification.	0.123 ms	0.122 ms – 0.124 ms	12.4 MiB	35.2 MiB	6.00 MiB

time series, 574 rows

time series, 574 rows: elapsed time per plot and memory by implementation
Implementation	Elapsed time per plot	Fastest – slowest run	Peak anonymous memory	Peak resident set size	PHP allocator peak
phiz52.1 KB specification.	0.631 ms	0.630 ms – 0.632 ms	12.9 MiB	35.8 MiB	8.00 MiB

facets, 7 panels, 234 rows

facets, 7 panels, 234 rows: elapsed time per plot and memory by implementation
Implementation	Elapsed time per plot	Fastest – slowest run	Peak anonymous memory	Peak resident set size	PHP allocator peak
phiz35.1 KB specification.	0.410 ms	0.409 ms – 0.411 ms	12.6 MiB	35.4 MiB	8.00 MiB

Methods

Implementations: phiz: phiz v0.5.2-9-g760c476, plain PHP
Machine: 2 AMD Zen 5 cores, 7.8 GB RAM, Linux
Software: PHP 8.5.12
Threads: 1
Repetitions: 7 timed runs, 1 warm-up run
Statistic: Median over the timed runs
Time measure: Elapsed time per plot, in ms
Timing boundaries: Mean time per plot over 100 builds and encodings per run. One warmup run. Dataset loading and PHP startup excluded.
Time spent in: The PHP process; specification building only, without browser rendering or vl-convert.
Cache: Warm
Memory metric: Peak anonymous memory; Peak resident set size; PHP allocator peak. Peak anonymous memory (RssAnon) of the whole PHP process, sampled every millisecond by peakmem (phpolars bench/csv-pipeline/peakmem.c). Also reported: peak resident set size, and PHP allocator peak (memory_get_peak_usage(true)).
Recorded: 2026-11-01, commit 760c476

Reproduction

From the repository root

composer install
php scripts/bench-build.php

Raw record: site/benchmarks/small-plot-build.json
Source: scripts/bench-build.php

Static export

vl-convert export

Results

Export of a plot with svg(), png(scale: 2), or pdf(). phiz builds the specification and runs vl-convert, which renders it without a browser.

Workload: vl-convert export, a 234-point scatterplot with a least-squares band and a 20,000-point scatterplot

vl-convert export, a 234-point scatterplot with a least-squares band and a 20,000-point scatterplot: elapsed time per export, peak anonymous memory, PHP allocator peak, and peak resident set size of child processes per implementation and case, median of 7 runs.

234 points with lm band: SVG

234 points with lm band: SVG: elapsed time per export and memory by implementation
Implementation	Elapsed time per export	Fastest – slowest run	Peak anonymous memory	PHP allocator peak	Peak resident set size of child processes
phiz + vl-convert71 KB output; building the 52 KB specification in PHP took 0.609 ms.	314 ms	303 ms – 326 ms	13.1 MiB	8.00 MiB	77.9 MiB

234 points with lm band: PNG, scale 2

234 points with lm band: PNG, scale 2: elapsed time per export and memory by implementation
Implementation	Elapsed time per export	Fastest – slowest run	Peak anonymous memory	PHP allocator peak	Peak resident set size of child processes
phiz + vl-convert149 KB output; building the 52 KB specification in PHP took 0.606 ms.	334 ms	324 ms – 335 ms	13.2 MiB	8.00 MiB	80.8 MiB

234 points with lm band: PDF

234 points with lm band: PDF: elapsed time per export and memory by implementation
Implementation	Elapsed time per export	Fastest – slowest run	Peak anonymous memory	PHP allocator peak	Peak resident set size of child processes
phiz + vl-convert173 KB output; building the 52 KB specification in PHP took 0.61 ms.	315 ms	313 ms – 326 ms	13.1 MiB	8.00 MiB	79.3 MiB

20,000 points: SVG

20,000 points: SVG: elapsed time per export and memory by implementation
Implementation	Elapsed time per export	Fastest – slowest run	Peak anonymous memory	PHP allocator peak	Peak resident set size of child processes
phiz + vl-convert5498 KB output; building the 948 KB specification in PHP took 12.3 ms.	523 ms	511 ms – 532 ms	33.4 MiB	27.4 MiB	177 MiB

20,000 points: PNG, scale 2

20,000 points: PNG, scale 2: elapsed time per export and memory by implementation
Implementation	Elapsed time per export	Fastest – slowest run	Peak anonymous memory	PHP allocator peak	Peak resident set size of child processes
phiz + vl-convert834 KB output; building the 948 KB specification in PHP took 12.3 ms.	1.09 s	1.08 s – 1.10 s	28.1 MiB	22.0 MiB	198 MiB

20,000 points: PDF

20,000 points: PDF: elapsed time per export and memory by implementation
Implementation	Elapsed time per export	Fastest – slowest run	Peak anonymous memory	PHP allocator peak	Peak resident set size of child processes
phiz + vl-convert13780 KB output; building the 948 KB specification in PHP took 12.3 ms.	896 ms	877 ms – 899 ms	41.5 MiB	35.5 MiB	212 MiB

Methods

Implementations: phiz + vl-convert: phiz v0.5.2-9-g760c476, running vl-convert 1.9.0 once per export
Machine: 2 AMD Zen 5 cores, 7.8 GB RAM, Linux
Software: PHP 8.5.12, vl-convert 1.9.0
Threads: 1
Repetitions: 7 timed runs, 1 warm-up run
Statistic: Median over the timed runs
Time measure: Elapsed time per export, in ms
Timing boundaries: Timed from the export call to holding the output bytes: specification building in PHP, vl-convert startup, rendering, and output reading. One warmup export. Data generation and PHP startup excluded.
Time spent in: vl-convert, launched by phiz once per export. Specification building in the PHP process is given in each row's note.
Cache: Warm
Memory metric: Peak anonymous memory; PHP allocator peak; Peak resident set size of child processes. Peak anonymous memory (RssAnon) of the PHP process alone, sampled every millisecond by peakmem (phpolars bench/csv-pipeline/peakmem.c). Also reported: the largest vl-convert process's peak resident set size, from getrusage(RUSAGE_CHILDREN), and PHP allocator peak (memory_get_peak_usage(true)).
Recorded: 2026-11-01, commit 760c476

Reproduction

From the repository root

composer install
PHIZ_VL_CONVERT=/path/to/vl-convert php scripts/bench-export.php

Raw record: site/benchmarks/vl-convert-export.json
Source: scripts/bench-export.php

Not measured

No benchmark measures how long the browser takes to draw a chart with vega-embed.