Testing Apache vs Open Litespeed

I’ve hosted forums and web sites for years, but I’ve always managed customers by hand, and I figured there needed to be a better way. So I’m in the process of evaluating the Enhance Control Panel. So far I’m impressed, but since I’m in the process of setting everything up I decided it was prudent to do some testing and see if web server choice actually mattered.

What was tested

Web servers tested

Enhance offers support for 4 web servers:

Apache
Nginx
Open Litespeed
Litespeed (the paid version)

All of these are capable web servers, but I ruled out Nginx from the beginning because of the non-standard way it handles .htaccess files. While performance on Nginx can be amazing (I run a forum with over 13 million posts on nginx and it’s fast), my main purpose here is hosting Managed WordPress sites. These are sites for regular people who want great web hosting without needing to think much about it, and Nginx isn’t very WordPress friendly.

That leaves us with Apache, and the Litespeed variants. Litespeed should be faster than Open Litespeed (and it will certainly scale better, as Open Litespeed is limited to one process,) but the technology is so similar that testing OpenLiteSpeed should give numbers so close to LiteSpeed as to be immaterial.

The sample site

The web site I used for this test is a WordPress site, built using Elementor, with WooCommerce installed. It’s a reasonably complex site, which I thought would be representative of the worst-case load times for typical web hosting customers. The main page (the one tested) contains 56 separate elements that need to be downloaded, and the only optimizations performed were installing caching plug-ins, turning them on, and enabling Cloudflare CDN. Better performance is certainly possible with any fine-tuning.

Testing

The web site load times were measured three times, and the results were averaged. The tests involved Apache without a cache, Apache with WPFastestCache, OpenLiteSpeed with no cache, and OpenLiteSpeed with lscache. Lscache is the native WordPress caching plugin provided by the makers of LiteSpeed, and WPFastestCache is well regarded so this seemed like a comparable alternative.

Both of the cache plugins were configured to use Cloudflare CDN, and the page was loaded one time to prime the cache before the test was run.

Analyzing the results

The biggest takeaway is obvous: use a cache. Page load times are 15% longer without a cache in Apache, and 19% longer with un-cached OpenLiteSpeed. Caching is free and simple to set up, so that’s your biggest bang for your buck.

Time to First Byte (an important metric for web servers) improvement was even greater: un-cached Apache is 49% slower than with a cache, and un-cached OpenLiteSpeed is 64.9% slower than using lscache. If you don’t want to keep your potential customers waiting, keep this low. Note that the referenced article on TTFB suggests keeping this under 0.8 seconds – in all cases we were significant below that.

Summary:

Overall page load times were marginally better with OpenLiteSpeed: 1% to 4% better. I’m going to call this insignificant.
TTFB improvements are significant. 5% faster up to 14% faster.
Not tested is the load on the server. From past experience, moving from Apache to LiteSpeed reduces load significantly, and I would not be at all surprised if future testing showed that a given server could serve four times as many clients using LiteSpeed than it can with Apache.

Conclusion

OpenLiteSpeed is better than Apache, but the difference is not as great as I expected or hoped.

From the web host perspective, reduced load is a huge advantage that’s hard to overlook. And from a business owner’s perspective, I want that TTFB to be as low as possible, as the higher that goes the more likely your users are to simply close the tab and go elsewhere:

Time to First Byte (TTFB) is the time it takes for a server to send the first byte of data to a user after accessing a website. A slow TTFB can lead to increased bounce rates, lower user engagement, and decreased conversions and revenues. According to Pingdom, a page's bounce rate increases from 9% to 38% when the page load time goes from 2 seconds to 5 seconds. Studies have shown that improving page speed by even one second can decrease bounce rates by 9%.

Google Search AI Summary Tweet

So I will be using OpenLitespeed for now, and as usage goes up I will migrate to a paid LiteSpeed plan to handle the increased load.

That said, Apache is performing well here as well, and for many visitors the differences between the two web servers may simply be imperceptible.

Overall the difference was smaller than I expected.