How to choose, buy or assemble the server yourself?

Question

System administration

The bottom line is this: now there are three servers and all of them, in fact, are just powerful desktops. No raids. No server hardware. At the moment, they have already reached the ceiling of their upgrade resource - you can’t stick more RAM into them (but I would really like to - now it costs 16 GB and it’s not enough at peak moments), raid on the built-in ones into some absolutely random and It's scary to assemble a very cheap mother controller, with newer processors the situation is better, but even on the current ones (AMD Athlon II X4 640), the CPU usage does not exceed 40% on average.
So far, the current server machines cope with their tasks quite well (terminal servers for 30 clients each + one of them runs MS SQL and Squid for a single department). In the future (about a year), an expansion is planned, due to which the number of terminal clients on this very server with SQL and Squid will approximately double.
I would like to replace it with something more serious, at least with real server hardware. Hence the question arises: how to assemble a decent terminal and SQL server for 100 clients with at least minimal upgrade options at minimal cost (put a new CPU out of the category and plug in a couple of RAM sticks)? The price of ready-made solutions is terrifying to the core, so for the time being I settled on self-assembly of a system based on some chassis from Intel (Intel 1U R1304BTLSFAN, for example) and the subsequent selection of components for it.
PS: By the way, there is also an option to buy a ready-made server on the above platform with core i3-2130/8gb DDR3/2x2TB SATA HDD inside for $750. Naturally, with the purchase of the missing amount of RAM, at least. Will it take off? And is the use of an outdated platform justified if the system is bought with the prospect of using it for another 5 years?

Answer 1

Regarding the cost, yes, the server is not cheap, but it leads to bad luck. Believe me, tier1 servers of manufacturers are worth their price. Let's estimate, sql 30,000, CAL 6,000 (x 100 clients) 600,000, win server 30,000, cal server 1,000 (x 10 clients) 100,000, cal terminal 3000 (x 100 clients) 300,000 total more than 1 million. rub. Server 750$? OO Good
advice from @edinorog , both about the brand and the processor.
Based on experience, I’ll add a little, immediately decide on the number of processors (taking a dual-processor configuration with one processor and hoping to add a processor later is not the best option - this happens very rarely), the amount of RAM (so that all slots are not filled with small memory), the disk subsystem (SQL + 1U , you should immediately take a platform with 8 SFF; R1304BTLSFAN, by the way, 4 LFF, not enough, IMHO).
Calculate the processor from the current load (maybe E3 Xeon is enough for you, it is used in a single-processor configuration, or maybe E5 is necessary), compare performance, approximately, you can here www.cpubenchmark.net
Measure the load on the disk subsystem now interpolate, taking into account the planned load - understand what you need.
If you do not want to take a brand, then ask the company from which you will buy equipment to help you complete the equipment (intel, supermicro is better for you).
When self-assembly of server hardware, there are many subtleties. Read forums, study manufacturers' websites, configurators, etc. Look at 3nity.ru/viewforum.php?f=1&sid=293093ebf780d0bc406...

Answer 2

Well, let's start with the fact that a ready-made solution is always cooler. And not much more expensive! For example HP server. Prots definitely take xenon. Mother of two. To buy later. The acquisition of an iron raid has one drawback. If the hell comes to him ... then the data can be pulled out only with the help of the same piece of iron. So I do not recommend buying something rare for our country. Definitely consider the presence of a full-fledged remote control up to the BIOS and installing the axis from the image. The rest is no longer important. All markers are different in taste and color.

Answer 3

It is unlikely that you will outdo at least one of the engineers who have been creating servers for HP / DELL / IBM for decades in terms of knowledge. And also testing under loads and so on.
The ready-made option is the only and right choice worth the money.
And at the same time - p / o to create backups.