Server Density

twest

Adam CPU and load level. On the old servers we'd be running around 700-800% CPU during peak hours with around 6 load. Topping out the same sites on servers in Enhance on similar configs I have to stop at around 75% of sites migrated before hitting those standard CPU/load levels.

On Enhance running Enterprise Litespeed so both new/old servers were using same lsphp, both used MySQL 8.0 with similar configs, hardware on new servers is about +10% better specced.

I've learned CL w/cPanel is a lot more forgiving at handling resource limits and preventing errors (php, etc) from affecting load as badly as Enhance, so I'm sure more hours spent optimizing sites would probably drive load down a bit, but also there's been a lot of work put into optimizing during migrations already - so it's just a bit disappointing to see the load levels what they are. This is also a low time of year for traffic, sites are down across the board about -30% compared to the seasonal peaks during the holiday months, so the load levels should really be a lot lower.

Adam

twest CPU and load level. On the old servers we'd be running around 700-800% CPU during peak hours with around 6 load. Topping out the same sites on servers in Enhance on similar configs I have to stop at around 75% of sites migrated before hitting those standard CPU/load levels.

Is it PHP or MySQL generating the majority of the CPU load?

The hard resource limits are hard resource limits, there is no burstability at this point however this might be possible in future. If the limits are too low this can actually cause more load as the end user will refresh the page which then starts a new request. If you have fast storage, you probably don't need to limit IOPS and IO Bandwidth at all. The most useful limit is often nproc which I would set to between 10 and 20 for shared hosting.

8Dweb

twest At this point, we have not migrated enough customers to any single web/app server to really have good data. However, we did build out far more extensively on Enhance than our Plesk platforms (dedicated dns, mail, app servers, and more app servers).

I do see one of our servers has higher cpu and load than any of the others, but I think it is related to some odd distribution bugs we had in Sept-Oct of last year, which caused that server to be more fully loaded, and it so happens with a number of our more active website customers. I will be upgrading that VM and moving it to a far higher CPU clock node.

One thing I will say is that even if cpu/load is a bit higher, time to first byte and page load on apache (tuned) is superior to our Plesk servers. Both are running on the same hardware with same hypervisor. I have not even dabbled with OLS or LS paid yet, because my customers and resellers are ecstatic with the performance.

As we migrate more Plesk servers in this year, I will be looking to see the limit. I do tend to opt for very low customer counts on our web-servers, though, since that is one of our platform "perks". (We do not place more than 100 customers on any single web server... thus you can see why Plesk pricing was starting to really be an issue).

AdamM

Adam limiting nproc is a good solution for php and wordpress? Will it not cause problems, e.g. when a customer buys products in the store and reaches the nproc limit?

twest

Adam MySQL load looks normal/good. The load seems to mostly come from PHP processes/lsphp. Maybe I need to try some more litespeed tweaks to see if it can be any better optimized. The docs available for tuning LS are amazingly poor, it's almost like even the Litespeed techs in their forum don't know how to optimize it beyond very genetics lol

I did disable all IO limits a while back, which appears to have helped fix most of the huge load spikes that would randomly pop up temporarily. I don't think any accounts are blowing nproc, and not sure I want to test lowering that and possibly triggering 503s on client sites where they had none before. I'll hit up LS support and see if they have any ideas.

Adam

AdamM

If nproc is exceeded then the website will give a 503 error when trying to create an additional worker. However, nproc 20 allows for 20 concurrent PHP processes, for example 20 people adding something to their basket at exactly the same moment. If you can serve most of your traffic from full page cache and keep runtimes short (object caching, good database design, etc) then 20 processes would be plenty for even the busiest of websites.

Industry standard for shared hosting seems to be between 10 and 20 processes. For VPS and dedicated you would generally leave this unset.

xyzulu

I don't ever seem to have CPU issues.. (on "shared" server) only memory/mysql issues. I have been trying to optimise mysql to help this.. I could put twice the number of customers on a server (CPU wise) than I currently have, but the limits seems to be Ram and mysql.

twest

xyzulu that's bizarre because it's the opposite for me. A full server on my old setup would use around 80GB of 250GB, while on enhance it's pretty uniformly at 50GB of 250GB used per server. Although some of that could be due to the lower density per server on my Enhance setup I suppose.

Adam

twest MySQL load looks normal/good. The load seems to mostly come from PHP processes/lsphp. Maybe I need to try some more litespeed tweaks to see if it can be any better optimized. The docs available for tuning LS are amazingly poor, it's almost like even the Litespeed techs in their forum don't know how to optimize it beyond very genetics lol

lsphp is just a Litespeed compatible SAPI on top of PHP. The execution of PHP itself is the same as it would be under php-fpm and won't be any different to PHP execution on any of the legacy control panels. I don't think there is anything you can tune to make PHP use less CPU time besides for opcode caching.

If PHP is the thing consuming your CPU then maybe the difference between your Enhance setup and your cPanel setup is the default caching rules in Litespeed - are you maybe forcing caching to on with cPanel? By default it's off on Enhance. You can default it to on but in my experience this is likely to cause more support. Just make sure all of your WordPress installations have the LS Cache module installed - you can make it a default plugin through Enhance.

xyzulu

I have tried a few options, but find OLS on enhance to be best for me.. mostly.. I have a few non-wordpress sites I run on apache.

twest

xyzulu yeah I switched to enterprise litespeed and that did help load a lot vs tuned Apache. I went with enterprise so I could get persistent config to optimize LS for big servers, like more workers, etc.

charlesalmeida

twest Another heavy usage of cpu spikes I saw in the backup rsync process was consuming a lot of resources.

COLBYLICIOUS

twest is there any big difference between OLS and LiteSpeed?

xyzulu

charlesalmeida I don't see this. Do you know you can configure the number of backup processes per server as well as frequency they run?

charlesalmeida

xyzulu
If backup is large for example 30GB it consumes a high performance cpu per backup for a few minutes totally.

charlesalmeida

xyzulu
Yes, I understand, this resolves the backup issue well.

charlesalmeida

Another thing is that lsphp's resident memory is stuck in a process all the time, consuming a lot of memory. Is there a way to reduce this?

charlesalmeida

charlesalmeida
I understand that building files can be very time consuming.

twest

charlesalmeida Yeah I turned down my number of backup processes to "1" a long time ago, helped improve load spikes a lot.

Memory isn't a problem for me, on 250GB machines it's only using 50GB and the other 200GB is free to use as cache.

twest

Adam no server level caching, beyond the Enhance opcache, and no server level cache on cpanel/cl. All the sites have the same cache configs as they had before, page caching via WP Rocket or other cache plugins. Besides the base OS and panel, everything is pretty much the exactly the same for configs setup. During migrations there's been a lot of PHP bug fixing and bad plugin purging too, so technically the sites should be consuming fewer resources (both because they're more optimized now, and also traffic is currently lower than normal).

grradmin

COLBYLICIOUS
https://www.litespeedtech.com/products/litespeed-web-server/editions

twest

COLBYLICIOUS yes enormous difference. You can't customize the config on ols, meaning you're stuck with whatever defaults. That could be fine for a small server/vps, but not for the large dedis I run.