How to choose a server for a project on 1C-Bitrix?

Question

System administration

How is it customary to choose server parameters for heavy projects?

Let's say there is a conditional website with 100 thousand users. goods updated every day with custom imports from 1C, and up to 5 thousand uniques with great activity. Naturally nginx + phpfpm.

The documentation of Bitrix itself contains very approximate figures without reference to project indicators. The Bitrix24 documentation describes it better - depending on the number of users, one or another configuration is needed.

I asked the Bitrix support a question, but they could not answer.

Answer 1

I asked the Bitrix support a question, but they could not answer.

And no one can answer.
Each site is unique, and it is extremely difficult to predict the load and required system resources.
Only by experience.
Rent a server, try to work, monitor the load, look for bottlenecks - and then it will be clear which server you need.
It is quite possible that it will not take off at all on one server, and it will have to be scaled.

Answer 2

Bitrix is ​​iron-demanding crap.
And you need to understand exactly what will be the bottleneck.
1. unequivocally the frequency per core, here you need to understand that most of these will be requests to the database, and the database can parallel requests for kernel requests, but the request itself is not spread over the cores, and if it is heavy, only the frequency will help here.
Therefore, 30X2GHz will be worse than 2 with 5GHz, moreover, significantly. That is why the zeons here suck, as a rule, there is a small frequency and a large number of cores, the usual i9900, which is 5 times cheaper, will give much more speed.
therefore, the maximum clock speed, in fact, is better than the I9900k or ryzen.
if the country is not important, then hetzner and there are such tariffs.
2. ssd -100% because its IO allows you to quickly lift heavy tables from the database. (and yes, not all of them are in memory) of
course, Nvme is desirable, since, again, speed also plays here, for Bitrix with 100K goods, requests can be more than 1GB in size, it physically needs to be raised into memory.
Well, yes, with 100 K of goods, you really need IO, especially when importing and exporting, and the speed itself will not be superfluous.
But iron is only tpovoe. But the number of stones is no longer so significant.
You are unlikely to eat more than 16 GB of memory, moreover, if you lift most of the database into memory.
All the same puff 7 eats memory neatly.
I recommend
https://ru.hetzner.com/hosting/produkte_rootserver...
small business on it about 100-170 parrots in the test. (Again, as collected)
Yandex cloud - also as an option.
Give the site, I'll tell you really what kind of conf is needed.
Plus, show a screenshot on the catalog page with a filter, how many requests are not cached (debugging) on ​​a second hit.
Still, 100K products here are already starting to really play the build quality of the site, and it usually does not take into account the features of the engine with such a quantity.

Answer 3

The first thing that catches your eye is why there is only one server? No matter how well linux isolates threads and queues commands, you need to separate roles by instances. Cramming everything into one, even the most powerful, server is a dead end scenario.
You must have:

1. Input balancer
   
2. Application server (N pieces)
   
3. mysql master+slave
   
4. memcached
   
5. Redis under sessions
   
6. Sphinx for search
   

Better instead of one expensive piece of iron, take a cloud in which you deploy everything described above.