I was working on a client site and was wondering about similar impact that my optimizations would have on a cloud instance with even traffic (read: zero traffic). I tested various things on a demo site that I set up for a client. I thought that the results were interesting, so I thought that I’d share.
- An 8GB IO Zoom instance with CloudLinux 7.3, MariaDB 10.1, Apache 2.4 proxied by Nginx, PHP 7.1, no TLS.
- A clean-but-tiny 2GB IO Zoom Ubuntu 16.04 instance running Redis in a Docker container.
Note: These servers literally get NO traffic (they are my personal development servers), so for the most part, all things should be equal.
- Page 1 – Shows the site with NO caching/optimization. Bare WordPress.
- Page 2 – I added Redis caching and shaved about 25% off of the load time.
- Page 3 – I ran Smush Pro (in Super Smush mode) and installed/optimized Hummingbird. Not much of a difference in load time. That suggests that media size/number of page requests isn’t the bottleneck – it’s server side processing and database calls that are the problem. That said, we already knew that WordPress is very chatty with the database.
- Page 4 – Just for grins, I installed Comet Cache Pro* (I typically only install it on servers that do not allow Redis, which is all shared providers), but I thought that it would be an interesting pair-up. It made more of a difference than I expected – even more than Smush Pro and Hummingbird (though I won’t complain about the extra 5% increase in speed that the latter two gave me).
Keep in mind, these are just tests on one web site using bare servers and a MacBook Pro – obviously, page size is important on mobile.
*There are many caching plugins, but I purchased a license for CC Pro because I give preference to unlimited site license with lifetime updates. It is the same reason that I purchased an iThemes Security Pro Gold license over WordFence. Despite requiring annual renewal for updates, I still purchased a WPMU Dev membership because so far it has proven useful.
Summary: Redis is still king, and shared providers suck.