Laravel's bootstrap cost 316KB of the 0.66MB a warm hello-world request used

Three numbers describe one PHP request, and the dev.to post keeps them apart. memory_get_usage() is what the script has allocated through PHP's allocator at that moment. memory_get_usage(true) is what the allocator has taken from the operating system, whether the script is using it or not ^[10]. The third is the process footprint, read out of VmRSS in /proc/self/status inside the same controller method ^[12].

The PHP manual, quoted in the post, calls memory_get_usage(true) "the value that memory_limit is enforced against" and adds that the amount the operating system has given the process is a different and typically much larger number ^[11]. Which counter you read matters for capacity work. The allocator underneath is the Zend Memory Manager in Zend/zend_alloc.c, and it asks the OS for memory in fixed-size chunks ^[16], so the post reports peak and peak_real separately ^[15]. The 316KB belongs to the first of the three counters ^[2].

Both headline figures convert cleanly. MB in the post means 1,048,576 bytes, the unit memory_limit counts in ^[9], so 0.66MB is about 692,000 bytes and 316KB is 323,584 bytes. The framework is around 47 percent of the warm hello-world peak ^[1]. Set the same 323,584 bytes against the low end of the quoted range, 20MB or 20,971,520 bytes, and it comes to about 1.5 percent ^[2].

Data volume moves the number much further. 17MB for 10,000 Eloquent models works out at roughly 1.8KB per hydrated row ^[3][3], and the row is small: a SKU, a name, a 120-character description, a price, a quantity, a flag and two timestamps ^[13]. The items table holds 100,000 rows ^[13], so loading all of them at the same rate lands near 170MB ^[4]. The post says the change that moved its number by a lot was how the models got loaded, after before-and-after runs on php artisan optimize, the Composer class map, deferred providers and OPcache itself ^[18].

The warm figure holds where its conditions hold. OPcache is on and has compiled everything it is going to compile, and every reading is the third request to the route ^[8]. The application is a fresh Laravel 13 install, and the 316KB covers what runs before the controller in that install ^[1][2]; an app with a dozen package providers registering at boot is measuring its own bootstrap, not this one. The harness was a single static PHP-FPM worker in the official php:8.4-fpm-alpine image on an Apple silicon Mac, with MySQL 8.4.11 in a second container and requests driven by cgi-fcgi ^[7].

The author says years of working with Laravel and Lumen showed how differently the two behave on memory ^[19], and went in expecting the ratio to fall the other way ^[17].