❗ WARNING ❗ These benchmarks are stale and very old. I plan to integrate the generation of these benchmarks with the publishing process, so they're automatically updated with every release. That will require a lot of work, but I plan to set aside time to do so 1H 2023.
These are some very basic and unscientific benchmarks of various commands
provided by the latest release of qsv. Please see below for more information.
These benchmarks were compiled against a 1M row, 512 mb, 41 column [sample of NYC's 311 data] (https://raw.githubusercontent.com/wiki/jqnatividad/qsv/files/NYC_311_SR_2010-2020-sample-1M.7z) on a Virtualbox v6.1 Windows 11 v21H2 VM with an AMD Ryzen 7 4800H, 32GB memory and a 1 TB SSD (VM configured with Ubuntu 20.04 LTS assigned 8 CPUs and 12 GB of memory).
BENCHMARK TIME_SECS MB_PER_SEC RECS_PER_SEC
apply_op_string 0.94 48.41 1,063,829.78
apply_op_similarity 1.38 32.98 724,637.68
apply_op_soundex 1.51 30.14 662,251.65
apply_datefmt 0.76 61.41 131,578.94
apply_emptyreplace 0.34 137.29 294,117.64
apply_geocode 40.45 1.15 2,472.18
count 0.09 505.70 11,111,111.11
count_index 0.01 4551.36 100,000,000.00
dedup 3.20 14.22 312,500.00
enum 0.31 146.81 3,225,806.45
exclude 0.27 168.56 3,703,703.70
exclude_index 0.26 175.05 3,846,153.84
explode 0.68 66.93 1,470,588.23
fill 0.71 64.10 1,408,450.70
fixlengths 0.40 113.78 2,500,000.00
flatten 4.42 10.29 226,244.34
flatten_condensed 4.50 10.11 222,222.22
fmt 0.23 197.88 4,347,826.08
frequency 2.09 21.77 478,468.89
frequency_index 1.43 31.82 699,300.69
frequency_selregex 0.38 119.77 2,631,578.94
frequency_j1 2.10 21.67 476,190.47
index 0.09 505.70 11,111,111.11
join 0.89 51.13 1,123,595.50
lua 5.37 8.47 186,219.73
partition 0.36 126.42 2,777,777.77
rename 0.26 175.05 3,846,153.84
reverse 0.35 130.03 2,857,142.85
sample_10 0.17 267.72 5,882,352.94
sample_10_index 0.01 4551.36 100,000,000.00
sample_1000 0.17 267.72 5,882,352.94
sample_1000_index 0.02 2275.68 50,000,000.00
sample_100000 0.30 151.71 3,333,333.33
sample_100000_index 0.30 151.71 3,333,333.33
sample_25pct_index 0.43 105.84 2,325,581.39
scramble_index 2.80 16.25 357,142.85
search 0.13 350.10 7,692,307.69
searchset 0.55 82.75 1,818,181.81
select 0.13 350.10 7,692,307.69
select_regex 0.14 325.09 7,142,857.14
slice_one_middle 0.08 568.92 12,500,000.00
slice_one_middle_index 0.01 4551.36 100,000,000.00
sort 2.23 20.40 448,430.49
split 0.26 175.05 3,846,153.84
split_index 0.06 758.56 16,666,666.66
split_index_j1 0.34 133.86 2,941,176.47
stats 2.70 16.85 370,370.37
stats_index 2.73 16.67 366,300.36
stats_index_j1 2.72 16.73 367,647.05
stats_everything 4.32 10.53 231,481.48
stats_everything_j1 8.97 5.07 111,482.72
stats_everything_index 4.14 10.99 241,545.89
stats_everything_index_j1 8.86 5.13 112,866.81
table 2.17 20.97 460,829.49
transpose 0.53 85.87 1,886,792.45
For reference, we also benchmark the last release of xsv.
BENCHMARK TIME_SECS MB_PER_SEC RECS_PER_SEC
apply_op_string NA NA NA
apply_op_similarity NA NA NA
apply_op_soundex NA NA NA
apply_datefmt NA NA NA
apply_emptyreplace NA NA NA
apply_geocode NA NA NA
count 0.10 455.13 10,000,000.00
count_index 0.01 4551.36 100,000,000.00
dedup NA NA NA
enum NA NA NA
exclude NA NA NA
exclude_index NA NA NA
explode NA NA NA
fill NA NA NA
fixlengths 0.52 87.52 1,923,076.92
flatten 6.07 7.49 164,744.64
flatten_condensed 6.19 7.35 161,550.88
fmt 0.25 182.05 4,000,000.00
frequency 3.21 14.17 311,526.47
frequency_index 2.02 22.53 495,049.50
frequency_selregex 0.01 4551.36 100,000,000.00
frequency_j1 3.07 14.82 325,732.89
index 0.11 413.76 9,090,909.09
join 1.32 34.48 757,575.75
lua NA NA NA
partition 0.53 85.87 1,886,792.45
rename NA NA NA
reverse NA NA NA
sample_10 0.27 168.56 3,703,703.70
sample_10_index 0.09 505.70 11,111,111.11
sample_1000 0.27 168.56 3,703,703.70
sample_1000_index 0.08 568.92 12,500,000.00
sample_100000 0.37 123.00 2,702,702.70
sample_100000_index 0.39 116.70 2,564,102.56
sample_25pct_index NA NA NA
scramble_index NA NA NA
search 0.19 239.54 5,263,157.89
searchset NA NA NA
select 0.20 227.56 5,000,000.00
select_regex NA NA NA
slice_one_middle 0.11 413.76 9,090,909.09
slice_one_middle_index 0.01 4551.36 100,000,000.00
sort 2.15 21.16 465,116.27
split 0.30 151.71 3,333,333.33
split_index 0.08 568.92 12,500,000.00
split_index_j1 0.44 103.44 2,272,727.27
stats 1.45 31.38 689,655.17
stats_index 0.24 189.64 4,166,666.66
stats_everything 3.50 13.00 285,714.28
stats_everything_j1 9.91 4.59 100,908.17
stats_everything_index 2.85 15.96 350,877.19
stats_everything_index_j1 9.95 4.57 100,502.51
table 2.96 15.37 337,837.83
transpose NA NA NA
The primary purpose of these benchmarks is to provide a rough ballpark estimate of how fast each command is, to catch significant performance regressions, and to help you fine-tune qsv's performance in your environment.
The count command can be viewed as a sort of baseline of the fastest possible
command that parses every record in CSV data.
Benchmarks with the _index suffix are run with indexing enabled. The _j1 suffix are run with
parallelization disabled - i.e. --jobs 1.
Note that the qsv stats command is slower than xsv stats primarily because qsv computes
additional stats (specifically - lower_fence, q1, q2_median, q3, iqr, upper_fence; skew, modes & nullcount)
and does date type detection.