Web server performance

flips

Was just searching for updated benchmarks and performance comparisons for webservers like nginx, h2o, Lighty, LiteSpeed and Caddy 2, but most hits I found was outdated or seemed to be wrong ...

Like this can't be the full/correct picture?

One thing for testing would be serving static files and php etc.
The other would be reverse proxy performance.
Anyone know of better tests?
(My use case is Lowend VPS'es serving low traffic sites. I can use either of these and be happy, I guess, just curious. I've been using them all, except h2o ...)

Ympker

On my last LE VPS I used lighthttpd :P

Also, don't forget about Open LiteSpeed

AC_Fan

I would argue popularity is more important, since performance is rarely a bottleneck, but issues and features are. Only nginx, Apache and LiteSpeed are used in the real world, by the average developer. Nginx for flexibility and performance, Apache for OOTB features and compatibility and LiteSpeed as an interesting, although commercial, intersection of the two.

InceptionHosting

Interesting, I have never seen that site before, I have to question the results somewhat, for no good reason though, just does not 'feel' like that is likely to be accurate.

I have had an idea for a while now but never got around to typing it up here, I will murder it in short form as it seems to fit.

I was thinking of hosting a friendly competition, we set up a wordpress site, a static site and an alternative (TBD) site so 3 content types, make them available as an export with zero optimisation.

We exclude Cloudflare (because @Neoon is right) and we set the specs e.g. KVM VPS 2 GB RAM, running OpenVZ 7 with 2 containers, 1 as your httpd stack and 1 as the local test source running a minimal UI (maybe), this rules out all latency. (I can build the image and provide deployment instructions for the OpenVZ 7 KVM image with the ready-made containers if needed)

Then it is purely down to optimisation, everyone participating MUST publish all changes and also what environment it is running on e.g. OVH/Hetzner/Inception/ExtraVM etc and CPU specs (basically the top bit of @Mason's yabs.

Then we can collectively tune and add our own insights referencing any tips we picked up from others until we all agree on a best-case general config for each category.

I will get some LES Optimisation king t-shirts/hoodies made to give out as prizes for MVP's etc.

Thoughts?

chocolateshirt

It is nice idea, Ant. I hope I could join the competition.

flips

@Ympker said:
On my last LE VPS I used lighthttpd :P

Also, don't forget about Open LiteSpeed

Yes, I like Lighty, but it seems the next major version might never be officially released.

I've been running Open LiteSpeed, too, but in my tests (some static files and a small PHP app), lighttpd (Lighty) and php-fpm performed better.
(WP with LSCache on LiteSpeed seems quite good, though.)

Ympker

@flips said:

@Ympker said:
On my last LE VPS I used lighthttpd :P

Also, don't forget about Open LiteSpeed

Yes, I like Lighty, but it seems the next major version might never be officially released.

I've been running Open LiteSpeed, too, but in my tests (some static files and a small PHP app), lighttpd (Lighty) and php-fpm performed better.
(WP with LSCache on LiteSpeed seems quite good, though.)

Lighty = Lighthttpd? Now everything makes sense

flips

@Ympker said: Lighty = Lighthttpd?

mezoology

Last 2 days I was playing around with Cloudlinux, installed DA, Plesk and Cyberpanel and ran some loader.io loads
I noticed (or maybe it's me have no clue about CentOS) that Debian/Ubuntu waaay outperformed both Centos and cloudlinux.
I used Apache, Apache+Nginx, Nginx, Litspeed and openlitspeed.
on Centos/Cloudlinux CPU flooded with lots of PHP/LPHP/PHP-FPM consuming 100% with maybe 25% of the load that i put on ubuntu/debian (that never passed 40% CPU with same panels and WS's, and exact same worpress site).
I might be missing some tweaks on Centos tho. Asked uncle Google, but he seems hiding some info from me.

havoc

From the testing I've done it seemed much more worthwhile to invest time into optimising the content rather than the server. i.e. something static-like will perform way better even on a slower web server by 10x

The second you're waiting for some dynamically generated page backed by a slow ass database there is only so much you can do to squeeze more % out of it.

Meaning you need a JAMstack solution for optimal speed - ideally with the dynamic bits implemented via serverless so that those scale with load.

Fritz

Never heard h20, looka promising too.

Although for simple static html, I prefer to use Python simple http server.