Benchmarks

Benchmarks

Comparison against FreeArc 0.67.1 using its built-in presets (arc.exe -cfg-), via tests/bench_vs_freearc.sh. FreeArc is only a performance oracle; PArc does not read or write .arc. Both use 4 compression threads (PARC_THREADS=4, Arc.exe -mt4); times are the best of two runs.

Test corpora (5–43 MB): C headers from /usr/include, Python sources, .so binaries, a mixed set and 3,000 small files. Total original size: 89.2 MB.

Size

Delta of total archive size (PArc − FreeArc; negative = PArc smaller):

level	text	binaries	mixed	3,000 small files
-m1	−14.0%	−10.2%	−9.7%	−17.8%
-m2	−3.9%	−13.5%	−9.1%	−10.2%
-m3	−2.9%	−7.4%	−5.0%	−4.8%
-m4	−4.5%	−2.8%	−2.5%	−5.4%
-m5	−1.3%	−0.9%	−0.9%	−0.5%
-m6…-m9	−1.0%	−0.2%	−0.3%	−0.3%
-mx (fast decode)	−1.0%	−0.2%	−0.4%	−0.3%
-mxx (CM)	−16.1%	−8.2%	−8.2%	−7.5%

Aggregate ratio (compressed / original; lower is better): -m6 20.98%, -mx 20.98%, -mxx 19.15%; FreeArc -mx 21.06%.

Speed

Wall-clock time on the 44 MB mixed corpus (FreeArc includes Wine startup):

level	PArc	FreeArc		level	PArc	FreeArc
-m1	0.99 s	1.68 s		-m6	6.28 s	10.28 s
-m2	1.39 s	2.44 s		-m7	6.06 s	10.32 s
-m3	2.93 s	3.97 s		-m8	5.80 s	10.29 s
-m4	4.10 s	6.21 s		-m9	5.93 s	10.60 s
-m5	4.03 s	8.60 s		-mx (fast)	5.62 s	10.48 s

-mxx on the same corpus takes ~30 s.

Decoding (verification t) on the mixed corpus: -m6 2.3 s, -mx 2.3 s, -mxx 31.6 s. Moving from CM to PPMd/LZMA made -mx about 12× faster to decode while still beating FreeArc in size; -mxx keeps the old CM ratio for when decode time does not matter.

Across the four corpora PArc is smaller in 40 of 40 level×corpus pairs up to -mx and faster in 38 of 40. The two speed exceptions are -m1 and -m3 on binaries (0.94 s vs 0.59 s and 2.86 s vs 2.61 s); both still compress 10.2% and 7.4% smaller. In -m1…-m9 the typical speed advantage is 1.4–2.2× with 0.2–13.5% smaller archives. Independent blocks are compressed and verified in parallel.

Silesia Corpus

PArc 0.0.1 vs FreeArc 0.67.1, 4 threads, 211,938,580 bytes:

level	PArc size	PArc ratio	comp.	dec.	FreeArc size	FreeArc ratio	comp.	dec.
-m1	63.974.529	30,185 %	1,56 s	0,40 s	72.044.101	33,993 %	2,07 s	0,86 s
-m3	48.582.743	22,921 %	5,51 s	1,17 s	51.883.380	24,480 %	11,95 s	4,58 s
-m4	46.744.025	22,057 %	7,08 s	1,39 s	49.028.632	23,133 %	20,19 s	4,91 s
-m5	46.579.694	21,979 %	9,37 s	3,79 s	47.647.910	22,482 %	33,92 s	11,22 s
-m6	45.753.042	21,587 %	18,75 s	12,81 s	46.081.902	21,743 %	67,64 s	12,65 s
-m9	45.743.654	21,583 %	18,41 s	12,39 s	46.067.031	21,736 %	72,03 s	14,28 s
-mx	45.670.609	21,549 %	18,41 s	12,89 s	46.067.031	21,736 %	74,94 s	12,97 s
-mxx	42.915.670	20,249 %	203,94 s	207,62 s	—	—	—	—

7-Zip -mx=9 reference: 48.688.242 bytes (22,973 %), 58,90 s / 1,89 s. In Silesia, -m1…-m5 beat FreeArc in size, compression and decompression; -m4 also beats 7-Zip in all three.

Reproduce

ARC_EXE=/path/to/Arc.exe bash tests/bench_vs_freearc.sh some-directory

Numbers come from one machine and two runs per level; use them to compare methodology, not as an absolute guarantee.