Litespeed Enterprise - Caching Settings

elk · 28 Feb

Hi everyone

We use @cPFence and Litespeed Enterprise on our servers. While cpfence reduced footprint compared to bitninja, for owl monitoring its recommended to disable swap: https://my.cpfence.app/knowledgebase/70/What-is-the-Recommended-SWAP-Size-with-the-New-Owl-AutoMySQL-Module.html.

One of our server has 29 websites, 5vcpus and 14gb ram. Another one has 4 vpus and 16gb ram with 48 websites on it.
- mem usage will always be around 80 to 85%.
if we scale up, litespeed will just cache more.
we have one additional server with only 3 WordPress/Litespeed pages on it, also 4vcpu and 16gb ram and it would still use 50-55% ram.
we still ran on v11

do you have any experience with customising the cache settings to reduce cached content from litespeed and keep more mem free? or would you recommend to simply not install litespeed cache on every page?

thank you for sharing your opinions

Kosta · 28 Feb

elk my opinion - get dedicated server 6-8core cpu 96GB memory at least per 48 websites… lol

cPFence · 28 Feb

elk

Upgrading to v12 will give you an instant 30% RAM drop out of the box. Also, make sure you have LiteSpeed Cache plugins and Redis active on all your CMS sites—you can use cPFence bulk tools to manage them instantly.

elk · 28 Feb

Kosta well yes. prices for servers in switzerland won't allow us that specs.

gcloud zurich: 8vcpus/64gb (without 3y commit): 580 usd /month.
means with 48 sites, 12 usd per website/month only for the server.

elk · 28 Feb

cPFence yep i saw that. we planned to upgrade yesterday, but decided to still wait a moment.

Andrei · 28 Feb

elk gcloud zurich: 8vcpus/64gb (without 3y commit): 580 usd /month.

Google Cloud is expensive, and you can get almost the same performance and uptime with Upcloud or Hetzner (dedicated CPUs, not shared).

elk · 28 Feb

Andrei it is yes. also the commitment lock-in for better pricing is ...

however, both hetzner and upcloud do not offer data residency switzerland. that's the main problem we are dealing with. there is infomaniak which is not that expensive, exoscale which offer self-managing. and some good small providers, but you will not be able to manage everything on your own.

but anyway, i now upscaling hardware would be possible, but i don't think that we need 96gb ram ...

but my question was about, is it possible to configure litespeed to cache less (until we moved to v12 and then everything is okay. 8vcpus and 96gb is not needed for those low traffic pages. )

Andrei · 28 Feb

elk You're right; the RAM usage shouldn't be that high if you have low traffic websites, especially if they're not dynamic in nature (e-commerce, membership etc).

Why do you think the LS cache is the problem? The cache should use up the storage space, not impact memory usage, right?

gmakhs · 28 Feb

elk I believe is the way PHP is handled, process's remain open so they consume about 150mb or ram per process, the behavior is not the same on other panels, if you disable swap you will face issues with the memory limit on your websites since if they hit it, they will give 503, if swap is enabled then they will swap and cause high server load, i have been trying to say on the forums that this is not NORMAL behavior and not the same on other panels and definitely needs investigation, but i believe that its not priority right now if even logged

Similar issue exist with Nproc

cPFence · 28 Feb

elk

Then your best move for now is to make sure all your LiteSpeed Cache plugins are optimized and Redis is active. Adding Cloudflare can also give you a great boost as a quick and easy fix.

If that doesn’t help, I’d just wait until you upgrade to v12, you can expect a 30–40% RAM drop, depending on your setup. No need to upgrade your servers yet.

elk · 28 Feb

thanks. i did some research (correct me if i am wrong )

so the monitoring and cp is calculating based on the "free ram". the actual ram usage is around 40% or less. but there is also a lot in the RAM Cache/Buffer, which then would be freed.

cPFence · 28 Feb

elk

Not sure I fully understand your question, can you elaborate?

We observed around a 30% drop in RAM usage across all our servers, including the main control panel and secondary servers. On one cluster, it was up to 40%. The new v12 architecture is great, and the control panel feels much snappier.

Enhance now has a solid base to build on for upcoming features, exciting times ahead. Hopefully, more great updates are coming soon.

elk · 28 Feb

cPFence

first of all, litespeed is not the problem. after deeper ram usage inspection I saw that litespeed processes do not consume a lot of ram. thanks to all the inputs.

what i noticed is, that grafana/prometheus and enhance control panel shows us a memory utilization of e.g. 80%.
but when i inspect it on the server or on grafana/prometheus, i can see, that actual RAM used is far below 80%.
There is a lot of RAM used for Buffering + Cache.

edit: i think have the reason. it has to do with swap disabling. one server was rebooted yesterday, that one has enough "available". the other one was not rebooted yet and still shows swap (si/so).

twest · 28 Feb

elk you don't want linux to stop caching... You seem confused about the difference between used ram, free ram, cached ram... Linux will cache some things when there's unused ram available - this is good as it improves performance of everything. If the system needs to use more ram, linux will shift it from cache to use. There's zero reason to go messing with that system. This also doesn't have anything to do with your websites and their page cache.

cPFence · 28 Feb

elk

V11 (Docker) is known for being heavy on memory usage. Instead of spending time investigating this outdated setup, just flip the switch and upgrade to v12, it’s working great now.

elk · 28 Feb

twest yes, i really was. but cleared that. also found a good and funny page about it: https://www.linuxatemyram.com/

cPFence yes, i will go for it in two hours.

Kosta · 28 Feb

elk why don’t you buy servers and place them on collocation I just checked 20u space is 480$/m mate u can place at least 15 servers in that space with whatever specs you like…

elk · 1 Mar

cPFence did it

went very smooth, beside of three pages that had permission issues on the /public_html

ram is lower now, yes!