Servers for a loaded project?

Question

Nginx

Good day, dear!
A few similar questions, please help if you know:
1. Approximately how many servers do you need to withstand a load of 100,000 people a day, peak - 150,000 (large project, social network, but optimized for loads)?
At the same time: Java project
Jboss application server
PostgreSQL or MySQL database
2. What class of servers are taken for such tasks and how much does one cost on average?
3. Who knows examples of well-known projects, please share (Habr, contact, etc.)?
Thanks for the help!

Answer 1

Everything needs to grow.
If there is no project yet, take a core server for four ...
Then if the project grows and brings in money, you can grow.
And so you don’t need to take cool servers for covered projects :)

Answer 2

One powerful server will cost less. First of all, you need to monitor powerful channels and OP for the cache. You can almost always save money on the process.
I collected Core i5-750, 8 DD3-1333, 2 * 1 Tb for such a load. But the site did not receive such a load.

Answer 3

I will say one thing: 150,000 hosts a day is not 150,000 online. with the usual return, the usual lamps on the VPS will also cope. but online is different.

Answer 4

The main thing is really not in the server ... It is very important to minimize the number of calls to the database, to separate the issuance of statics from the main core (in this case, the core is (LAN port, hard drive).
Also, the basis of performance is to release the user as quickly as possible (tobish do not create CMO, do not pull AJAX every second) -
return the result as quickly as possible, otherwise drown the server with 192 cores and 512 GB of RAM.

Answer 5

Start with a VPS with more RAM. A good server will cost good money, and you definitely won’t get 100k a day right away.
Then you can rent something, and when you reach your required indicators, then buy. Otherwise, you’ll just pour money into resources that will only be needed in a year. And in a year it will be possible to take something tastier for these grandmothers :)

Answer 6

1) it is impossible to tell from these data. You did not specify peak loads and it is not clear how fast a single request is processed. At least 2: database server and application server :)
2) how do you do it. Google collects on cheap hardware, but there is a lot of iron.

Answer 7

150,000 requests, that's only 1.7 requests per second.
How long does one request work and how much does it consume resources? it won’t turn out that before the first request stops, the tenth one will start and the load will increase ...

Answer 8

You need to look not for something specific according to a too general description of the project, but simply start with some dead piece of iron: you are unlikely to get 100K hosts in the first weeks, and as the turnover grows, you will see for yourself where there are weaknesses in the project that can be fixed by increasing performance.
This is the number of queries to the database per hit, and% of queries for statics, the load on the disk subsystem, application design errors, ...
So take the average and further lick the project - (I agree with Horse) it's not about the server.

Answer 9

I would like to add. If you are afraid not to cope with the load on the VPS - take a cloud in addition.
It helps.