Website slows down or crashes on running benchmark (ab, wrk)

[noob404]

Hi,
I am new to OLS. Was planning to migrate from Apache as I have heard great things about OLS. So, I just setup LOMP (with PHP 8.1) with VSFTPD and phpMyAdmin on my Ubuntu 22.04 Server (Virtualbox) - 2GB RAM, 2 Cores, for testing.

My website uses PHP an has atleast 3-4 MySQL calls every page.

Haven't changed many settings. But, this is what I am noticing that wasn't there on Apache. I tried running two benchmarks (screenshots below):-
1. Ab (Apache Benchmark) - It just times out.
2. Wrk - Completes but there are timeouts in the results.

The irritating thing though is that, reloading the website after these benchmarks, it just keeps reloading and never actually reloads. I then have to go the OLS panel and graceful restart it, for the website to load.

​

My first doubt was it was a server overload, but below is the htop screenshot which clearly states that almost nil resources were being used during the benchmarks. I also checked the disk usage, which was also very low.

​

I believe that a little bit of customisation is required to get the results close to Apache or better, but, I don't know where to start. Any help is appreciated.

 

[noob404]

Ran some more tests and found some very disappointing results.
Here is the same site with Apache (completed 10K requests in 36s)


The same website with OLS (after tweaking some things following this - forum.openlitespeed.org/threads/openlitespeed-limitation-based-on-resource.4222 Took a whopping 410s


I am still not able to make a sense of it cause in both the instances when it's run on the OLS server, the CPU usage on a single core is less than 20%, while the other core uses less than 5%. On the other hand, Apache uses 100% of both the cores while benchmarking. I am sure that the problems would be solved if there was a way to get OLS to use more resources. But, even after searching for the past few days, I can't get answer to this.

 

[noob404]

UPDATE 2 - Tried reducing the core to 1 from 2, and yet get the same results, yet again proving that OLS refuses to use the full capacity of the machine. Would anyone please help out?

 

[Cold-Egg]

Is it a custom PHP site? It is weird because I did not see any PHP or OLS process running from your screenshot, have you checked if the site is loadable on your browser? Also, is it possible to apply a cache on the site?

 

[noob404]

Is it a custom PHP site? It is weird because I did not see any PHP or OLS process running from your screenshot, have you checked if the site is loadable on your browser? Also, is it possible to apply a cache on the site?

Hey, thanks for the reply.
Yes, it's a PHP site. Pardon me for the htop screenshot. I do see mariadb, osl and lsphp on htop when the benchmark is run.
Yes, it loads and functions properly on the browser.
The final screenshot is after I installed LSMCD if that's what you mean by cache.

 

[noob404]

I just re-installed LOMP on a fresh Ubuntu installation. Here is an htop screenshot from when the ab benchmark was running:-

​

Another interesting thing is that, OLS on aaPanel seems to be doing better than manual LOMP. Here is a screenshot of OLS ab results on aaPanel:-

​

The only major difference I could notice when compared was that aaPanel pointed to fcgi-php (I am guessing PHP-FPM) while manual OLS uses LSAPI.

UPDATE - Checked phpinfo on aaPanel, no mention of php-fpm. Damn, back to square one. Any ideas?

 

[Cold-Egg]

The cache I mean is LSCache, see https://docs.litespeedtech.com/lscache/noplugin/installation/#basic-rewrite-rules
Thanks, I can see the OLS and PHP process now, but the load on those process are extremely low, it might be a config issue somewhere. If the web admin is accessible via to public network, feel free to inbox me.

 

[noob404]

The cache I mean is LSCache, see https://docs.litespeedtech.com/lscache/noplugin/installation/#basic-rewrite-rules
Thanks, I can see the OLS and PHP process now, but the load on those process are extremely low, it might be a config issue somewhere. If the web admin is accessible via to public network, feel free to inbox me.

Figured it out. I noticed this issue didn't exists with MySQL. Turns out mariaDB has a max_connections parameter. I edited it under the file

/etc/mysql/mariadb.conf.d/50-server.cnf

and then ran the benchmark. There's still some throttling after some time, that I need to figure out. But, CPU usage did hit close to 100%.