huniq version 2
Command line utility to remove duplicates from the given input. Note that huniq does not sort the input, it just removes duplicates.
SYNOPSIS: huniq -h # Shows help
SYNOPSIS: huniq [-c|--count] [-0|--null|-d DELIM|--delim DELIM]
$ echo -e "foo\nbar\nfoo\nbaz" | huniq
foo
bar
baz
$ echo -e "foo\nbar\nfoo\nbaz" | huniq -c
1 baz
2 foo
1 bar
huniq
replaces sort | uniq
(or sort -u
with gnu sort) and huniq -c
replaces sort | uniq -c
, assuming the data is sorted just so it can be passed to uniq
. If having sorted output is desired, sort | uniq
should still be used.
The order of the output is stable when in normal mode, but it is not stable when in -c/count mode.
Installation
$ cargo install huniq
Motivation
Sorting is slow. By using hash tables/hash sets instead of sorting
the input huniq is generally faster than sort -u
or sort | uniq -c
when testing with gnu sort/gnu uniq.
Version History
Version 1 can be found here.
Changes made in version 2:
- The -d/-0 flags where added so you can specify custom delimiters
- Completely rewritten in rust.
- Version two is (depending on which benchmark you look at below) between 3.5x and 6.5x faster than version 1
Build
cargo build --release
To run the tests execute:
bash ./test.sh
Benchmark
You can use bash ./benchmark.sh
to execute the benchmarks. They will execute until you manually abort them (e.g. by pressing Ctrl-C).
The benchmarks work by repeatedly feeding the implementations with data
from /usr/share/dict/* and measuring memory usage and time needed to process
the data with the unix time
tool.
For the uniq
algorithm, the results are posted below: We can see that the
rust implementation blows pretty much anything else out of the water in terms
of performance. Use sort only if you really need a coffee break, because you
won't get it with huniq! It beats the C++ implementation by a factor
of between 6.5 (for very few duplicates) and 3.5 (around 98% duplicates).
Compared to sort -u
: huniq is around 30 times faster.
If memory efficiency is what you are looking for, use datamash which is not as fast as huniq
but uses the least memory (by a factor of around 3); failing that use sort|uniq
which is a
lot slower but uses just very slightly more memory than datamash.
repetitions implementation seconds memory/kb
1 huniq2-rust 0.26 29524
1 huniq1-c++ 1.67 26188
1 awk 1.63 321936
1 datamash 1.78 9644
1 shell 7.30 9736
2 huniq2-rust 0.84 29592
2 huniq1-c++ 3.28 26180
2 awk 3.71 322012
2 datamash 4.60 9636
2 shell 16.68 9740
5 huniq2-rust 2.02 29648
5 huniq1-c++ 6.21 26184
5 awk 7.69 322012
5 datamash 9.10 9992
5 shell 44.71 10184
10 huniq2-rust 3.40 29676
10 huniq1-c++ 12.84 26172
10 awk 16.73 321940
10 datamash 24.44 10032
10 shell 93.75 10036
50 huniq2-rust 14.68 29612
50 huniq1-c++ 55.32 26200
50 awk 74.91 321940
50 datamash 103.54 10936
50 shell 453.94 10956
100 huniq2-rust 43.65 29492
100 huniq1-c++ 154.99 26180
100 awk 239.66 321956
100 datamash 285.94 12148
100 shell 1062.07 12208
For the counting huniq -c
implementation, the speed advantage
was less pronounced: Here the rust implementation is between 25%
and 50% faster than the C++ implementation and between 5x and 10x
faster than sort | uniq -c
.
The increased memory usage of the rust implementation is much worse though:
The rust implementation needs about 2.2x more memory than the C++ implementation
and between 10x and 12x more memory than sort | uniq
.
repetitions implemetation seconds memory/kb
1 huniq2-rust 1.47 132096
1 huniq1-c++ 1.85 60196
1 awk 2.79 362940
1 datamash 2.28 9636
1 shell 7.71 11716
2 huniq2-rust 2.32 132052
2 huniq1-c++ 2.98 60156
2 awk 4.65 363016
2 datamash 5.27 9732
2 shell 16.37 11680
5 huniq2-rust 4.98 132092
5 huniq1-c++ 7.54 60128
5 awk 9.37 363016
5 datamash 11.22 9964
5 shell 44.77 11948
10 huniq2-rust 8.81 132048
10 huniq1-c++ 13.55 60196
10 awk 16.19 363032
10 datamash 25.12 9908
10 shell 90.01 11976
50 huniq2-rust 45.89 132092
50 huniq1-c++ 74.04 60104
50 awk 85.43 362956
50 datamash 141.90 10996
50 shell 454.42 12876
100 huniq2-rust 90.80 132080
100 huniq1-c++ 150.41 60196
100 awk 163.13 363008
100 datamash 322.70 12212
100 shell 933.67 14100
Future direction
Feature wise huniq is pretty much complete, but the performance and memory usage should be improved in the future.
This first of all involves a better benchmarking setup which will probably consist of an extra rust application that will use RNGs to generate test data for huniq and take parameters like the number of elements to create, the rate of duplicates (0-1) the length of strings to output and so on…
Then based on the improved benchmarking capabilities, some optimizations should be tried
like short string optimization, arena allocation, different hash functions, using
memory optimized hash tables, using an identity function for the uniq
function
(we already feed it with hashes, so a second round of hashing is not necessary).
License
Copyright © (C) 2020, Karolin Varner. All rights reserved.
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
Neither the name of the Karolin Varner nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL Softwear, BV BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.