lighttpd?

[adaliszk]

I recently did a couple of benchmarks to choose which server solution is efficient to run a high-speed api, and I found out that lighttpd actually can faster serve static content than nginx. The cgi version is not faster but I think the fpm might be (currently configuring). Is there option to add lighttpd as well to the images?

[petk]

OpenLiteSpeed does miracles... Kind of also why lighttpd and Caddy wasn't added here just yet. Benchmarking web servers is very complex task and depends on many many factors.

[adaliszk]

It does? From my initial benchmarking tests with multiple threads a simple hello world is kind of the same speed as an apache. For now I admit that the tools which basically doesn't slow the code really nice with litepseed, but it doesn't seems to have that much miracle.

Currently the avarage speeds are (i7-6700,32GB RAM,High Speed SSD):

• Apache2: 58K paralel (62K peak), 18K serial (19K peak)
• Nginx: 54K paralel (58K peak), 17K serial (18K peak)
• Lighttpd: 68K paralel (72K peak), 16K serial (20K peak)
• Litepseed: 56K paralel (66K peak), 15K serial (26K peak)
• PHP: 12K paralel (18K peak), 4K serial (8K peak)

Paralel is where I start requests until I have to wait to the server or to the client, Serial where I start them after each other and wait for them to finish. Sadly it seems I cannot call more than 80K because my cpu is burning, so I will do the tests at home with my i7-7700K, I think I will boost that up to 5GHz just for this :D

This Caddy is interestig but I don't need automatic SSL, that's added with the proxy but I will do somes tests with it soon.

[gstrauss]

@adaliszk lighttpd 1.4.46 was just released today and I am curious if it is am improvement for you. As I am a lighttpd developer, I am curious what your benchmark is here. What is your lighttpd.conf and how are you running your benchmark? As @petk said:

Benchmarking web servers is very complex task and depends on many many factors.

[adaliszk]

Hey @gstrauss ! Sorry for the late response, I will upload the tester code and dockerfiles probably this weekend, I will ping you when it's up.

The method is basically to run one docker container, call a really simple <?php echo "🐘"; file with https://github.com/wg/wrk with paralel threads note the results, after rebooting the container running the tests with https://httpd.apache.org/docs/2.4/programs/ab.html because it seems not working in paralel, and a third one is using the time tool and call curl inside and outside from the container.

I want to add to this chain the static contents as well and a few type of php scripts so it would possible to see that is there difference in runtime with more code between servers. For now I only have a benchmark comparison between the group containers on cli level: https://github.com/adaliszk/benchmark-php-cli

[petk]

Hello, @adaliszk and @gstrauss there is now a lighttpd image added as well using the default lighttpd from Alpine stable repository.

It will be more fined tuned in the near future.