Testing the Symfony 4.1 router

How much faster is the new Symfony router in a real world application?

We ran some tests to find out.

An improved router

A few days ago we received the fantastic news that the Symfony router had been improved significantly. Of course we wanted to see how the changes would affect our application, so we decided to test it against our own routes.

If you somehow missed the news you can read the official announcement.

If you want to know how the new router works, read Nicolas Grekas post where he dives more into detail about the new changes.

Our setup

• PHP version: 7.1.13
• Static routes: 285
• Dynamic routes: 350

The test process

The process in itself is quite simple. Since the performance of the matcher is based on the amount of configured routes that we have, we want to test against the first route, some random routes, and the last route, to get a realistic test case. This was done for both the static and the dynamic routes.

class RouteTestCommand extends BaseCommand
{
    const NR_OF_MATCHES = 50000;
    protected static $defaultName = 'route:test';
    private $router;

    public function __construct(RouterInterface $router)
    {
        parent::__construct();
        $this->router = $router;
    }
    
    protected function execute(InputInterface $input, OutputInterface $output)
    {
        $output->writeln('First static route: '.$this->matchRoutes(['/admin/statistics/index']));        
        // Add random/last static routes..

        $output->writeln('First dynamic route: '.$this->matchRoutes(['/admin/statistics/3aa63971-1f61-4b28-bd60-5bb1c9582c8a/show']));
        // Add random/last dynamic routes..
        
        $output->writeln('Not Found: '.$this->matchRoutes(['/foo/bar/baz']));
    }

    private function matchRoutes(array $routes): float
    {
        $startTime = microtime(true);
        foreach ($routes as $route) {
            for ($i = 0; $i < self::NR_OF_MATCHES; ++$i) {
                $this->router->match($route);
            }
        }
        $time = microtime(true) - $startTime;

        return number_format(1000 * $time / count($routes), 2);
    }
}

Symfony 3.4

In Symfony 3.4, the matcher has to iterate through all/most of the configured routes and try to find the route for the given url. There are some optimizations like checking group of routes that have the same static prefix, and thereafter the routes in that subset. Trying to match a url towards the end of the list will result in an increased matching time, due to the number of comparisons being done before the route is found.

That is why our tests for random route and last route take significantly more time to match than the first route. The routes that were not found will take the most amount of time to match, since all the configured routes must be iterated before drawing the conclusion that our route does not exist.

Symfony 4.1

Instead of making separate comparisons or preg_match() calls for each route, it combines all the regular expressions into a single regular expression. This means that we only have to call preg_match() once, and that is the biggest factor for faster matching.

If you’d like to read more about the regular expression optimization, check out part 2 of Nicolas Grekas medium post.

Results

The table below shows how long it took to match the given routes 50.000 times.

The Diff column displays how much faster the new Symfony router was, compared to our current one (3.4).

Routes.	3.4	4.1 (7d29a4d)	Diff
First static route	448 ms	382 ms	-17%
Random static route	1621 ms	474 ms	-242%
Last static route	1826 ms	544 ms	-234%
First dynamic route	746 ms	527 ms	-41%
Random dynamic route	1454 ms	531 ms	-174%
Last dynamic route	2039 ms	524 ms	-289%
Not Found	2112 ms	522 ms	-304%

As we can see, all of our routes were matched faster than previously. We especially see a gain in matching speed for the random, last, and when the route was not found.

We are happy with these results, since the only thing we have to do once it is released is to run composer update.

Remember that this is our results on our application. Your results will probably not be the same since the performance is highly dependent on the amount of routes that your application has and on the tree structure of the routes.

What results did you get?