What are the options for building a cluster for office work?

Question

Operating Systems

For starters, context.We have some nice machines in the office with i5 processors, 500G disks and 8G RAM. There are slightly more employees and replenishment is planned. Computing power is redundant 95% of the time until it's time for builds or heavy calculations. Most of the time, employees will write code, work with office documents and office services (mail, 1C, git, redmine, etc.). Therefore, it was decided to combine existing good computers into one good cluster system on which to raise an RDP server, and all transfer employees to thin clients. At the same time, there is a wish that the resulting solution would have shared memory (in our case, 4x8G), and the disks of these machines should be combined in pairs into a 2-raid array (in total, 2 partitions of 500G each should be available for operation in the resulting system). One more nuance:

Unfortunately, we don’t have a single person in our team with experience in creating such a cluster, and therefore, on the one hand, we don’t know how realistic our Wishlist is and, on the other hand, we don’t quite know where to run.

Intermediate result:
With the help of Google, I have found Rocs so far , but it somehow smells of mothballs - all articles on the Internet are very old, they are not mentioned in Russian Wikipedia, and on the official website the latest documentation update is from 2007. Yes, and I doubt what will come of this decision to make a shared environment for office work.

I also found Kerrighed - it already looks more like what we need. but, if I understand correctly, here we are building a single OS with 16 available cores (for our case), butRAM and hard drives are not combined . All right with the disks - my eye has already fallen on GlusterFS , but why is everything so tight with the RAM?

Questions to respected experts and other no less respected scholars:

1) Are our requests for building a cluster realistic?
2) What technologies would you recommend using to build such a cluster?

3) Bonus lamer question: are there any clustering solutions that will allow four employees to work behind 4 workstations (controller and 3 nodes), so that they would not be transferred to thin clients? Or do all decisions categorically exclude this and require the transfer of system engineers out of sight into a deserted server room?

Answer 1

Build a local cloud based on, for example, Proxmox VE, raise virtual machines inside, work through thin clients / MiniX or any other analogues.
Memory clustering, in your interpretation, is impossible (at least there are no such solutions at the moment, including commercial ones) for one simple reason - DDR3.
The data transfer rate of which is from 50 to 150 Gigabits per second. You do not have and you will not buy them in the nearest store network cards (even optical / thunderbolt) with a throughput at least close to the lower threshold. Not to mention the switch, the backbone of which should resolve, a much larger amount of transmitted data.