The JIT in relation to PHP extensions

A few days ago I posted about Playing with the PHP JIT and included some simple benchmarking with the react-php-redis server project, which involves a lot of parsing but is ultimately still bound by I/O even when running async.

I got some questions on Twitter that are around some misconcetions of what the JIT really an do for PHP applications and what it cannot do.

So to show what the JIT is good for, I wanted to have truly CPU bound problem that was realistic from my POV.

Inside Tideways we use a datatype called HDRHistogram (high dynamic rrange histogram), a statistical datatype to calculate exact percentiles in monitoring data. For each minute and server we might have a histogram and when rendering a chart, we merge and aggregate this data in large numbers.

At the moment we use a PHP Extension interfacing with a C library to use this datatype.

I have ported the necessary code to PHP to test this with the JIT, without the JIT and against the PHP extension.

<?php

function simulate_hdr() {
    $hdr = hdr_init(1, 1000, 1);
    for ($i = 1; $i < 1000; $i++) {
        for ($j = 0; $j < 1000; $j++) {
            hdr_record_value($hdr, $i);
        }
    }
    hdr_value_at_percentile($hdr, 95);
}


for ($i = 0; $i < 5; $i++) {
    $time = microtime(true);
    simulate_hdr();

    echo number_format(microtime(true) - $time, 4) . "\n";
}

Again take the numbers with a grain of salt, these are just here to show the approximate relationships:

Runs PHP nojit PHP jit C/PHP Ext
1 0.5916 0.3671 0.0775
2 0.6322 0.4038 0.0775
3 0.6025 0.3866 0.0799
4 0.6010 0.3892 0.0829
5 0.6137 0.3947 0.0828
Average 0.6082 0.3883 0,0801
% 100,00% 63,84% 13,17%

As you can see, the JIT code runs at roughly 2/3 (63,84%) of the original non-jitted code and gets into the region of twice as fast that the RFC claims for PHPs internal benchmark. The improvement is much better than with the react-php-redis server example from a few days ago, where the improvement was only in the 5-20% region.

But compared to implementing this code directly in C as a PHP extension, even the jitted code is still 5 times slower.

Yes, with the JIT there is a massive improvement of this CPU bound problem, but it doesn’t mean we can now re-implement all PHP extensions in pure PHP and rely on the JIT to make them perform.

What the JIT does improve:

  • Make the parts of CPU bound problems that are written in PHP (!) faster.

What the JIT does not improve:

  • It does not improve performance of already fast internal functions written in C, for example hashing, encryption functions.
  • It does not improve performance (by a lot) for I/O bound problems.

To close the gap between JIT and C, we could look at PHP 7.4 including the FFI extension. It allows interfacing with C code more easily from PHP. Anthony Ferrara is building his “php-compiler” project on top FFI that would allow compiling a subset of PHP code directly to an FFI C extension.

Sign up for my newsletter to get notified when I post new content on this blog and with the occasional exclusive content only for subscribers.

By clicking on the Subscribe button I am giving my consent for Benjamin Eberlei to hold my name and email address for the purposes of contacting me with a newsletter on the topics of this blog. You can unsubscribe with one click at any time and withdraw your consent. No spam. I will never share your e-mail address. Privacy Policy