GCP Compute-Optimized Virtual Machines Benchmark (C2)

[mouakter14]

Now for the fun part, the data! We ran a multitude of tests so you could see the difference these new types of BPC machines can make. For each test, we tested the homepage for one minute with 15 simultaneous users. Below are the details of our test environments.

New C2 machine: Intel Cascade Lake Scalable processor, powered by GCP and running in an isolated container.
Machine N1: Intel Skylake processor powered by GCP and running in an isolated container.
OS: Ubuntu 16.04.6 LTS
Container Stacking: Nginx 1.15.12, MariaDB 10.4.6
PHP version: 7.3
Page Cache: We ran multiple tests, one to HIT the Kinsta cache and one to CLEAR the Kinsta cache.
Testing was conducted by Thoriq Firdaus , a WordPress contributor and web developer here at Kinsta. He has contributed WordPress Core and Translation Editor for WordPress Indonesia and develops a benin phone number data free MailChimp subscription plugin, WP Chimp .

The following data is measured in requests per second. The more you request, the better.

WordPress 5.2.3 (BYPASS Kinsta Cache)
In our first round of testing, we are intentionally avoiding Kinsta’s page cache . The reason is that we want to test CPU performance using the following stack: Nginx, PHP-FPM, and MySQL . Every request not cached on the site is handled by a PHP thread .

We are using WordPress 5.2.3 and the Twenty Nineteen theme . Our site was populated with dummy data from WP Test . It contains images of all sizes (small, medium, large) as well as an embedded tweet.

We're not too concerned about the dummy data as long as it matches on both machines. Measuring the performance difference on the GCP machines is really what we're interested in. Would having more dummy data impact performance? Yes, but we've also tested some fat client sites below so you can see the impact as well.

We tested the site for one minute with 15 simultaneous users.