Add decoding benchmark plus benchmark for GZIP-compressed CSV files

[clue]

Refs clue/reactphp-ndjson#16, clue/reactphp-zlib#24 and others
Resolves #11

[clue]

For the reference, here's what the output looks like on my machine:

$ php examples/91-benchmark-count.php < IRAhandle_tweets_1.csv 
243891 records in 4.812s => 50683 records/s

$ php examples/92-benchmark-count-gzip.php < IRAhandle_tweets_1.csv.gz 
243891 records in 4.789s => 50929 records/s

Interestingly, this means that decompressing and parsing a large file was slightly faster than parsing an uncompressed CSV file on my system (I/O overhead vs CPU usage).

It's worth noting that decoding CSV is quite fast – but decoding NDJSON is somewhat faster than decoding CSV in my benchmarks (clue/reactphp-ndjson#16):

me@me-in:~/workspace/clue-reactphp-ndjson$ php examples/91-benchmark-count.php < IRAhandle_tweets_1.ndjson 
243891 records in 1.021s => 238835 records/s

[loilo]

Comment incited by this tweet.

Running the above benchmark against this CSV file (careful when on restricted traffic volume, 90MB file), I get roughly 25% improvement on PHP 8.0, with another ≈4% on top when enabling JIT.

The running machine is a MacBook Pro, late 2019, with a 2.3GHz 8-Core Intel Core i9 CPU.

PHP 7.4

php -v

PHP 7.4.14 (cli) (built: Jan  8 2021 13:20:04) ( NTS )
Copyright (c) The PHP Group
Zend Engine v3.4.0, Copyright (c) Zend Technologies
    with Zend OPcache v7.4.14, Copyright (c), by Zend Technologies

php examples/91-benchmark-count.php < IRAhandle_tweets_1.csv

243891 records in 2.685s => 90838 records/s

Records per second roughly settle between 86k and 92k.

PHP 8.0

php -v

PHP 8.0.1 (cli) (built: Jan  8 2021 12:43:54) ( NTS )
Copyright (c) The PHP Group
Zend Engine v4.0.1, Copyright (c) Zend Technologies
    with Zend OPcache v8.0.1, Copyright (c), by Zend Technologies

Without JIT

php examples/91-benchmark-count.php < IRAhandle_tweets_1.csv

243891 records in 2.146s => 113654 records/s

Records per second roughly settle between 105k and 120k (a plus of about 25%).

With JIT

php -d opcache.enable_cli=1 -d opcache.jit_buffer_size=100M examples/91-benchmark-count.php < IRAhandle_tweets_1.csv

243891 records in 2.051s => 118918 records/s

This is not a lot faster than without JIT, but shows way less variance with records per second settling between 115k and 120k.

[clue]

@loilo Thank you very much for sharing your results here!

Really interesting to see how PHP 8 improved this somewhat and with JIT enabled may be able to achieve even better results. I haven't toyed around with this yet, but I wonder if different JIT settings (e.g. https://stitcher.io/blog/php-8-jit-setup) may yield even better results 🤘

By the way, if you're into benchmarking: https://github.com/clue/reactphp-ndjson includes a very similar benchmark script that can be executed on the same input data set. It was ~5 times faster than this one last time I checked. https://github.com/clue/reactphp-tsv includes a similar benchmark script that was ~10 times faster than this one. 🔥