How justified is the use of outdated server processors in combat systems in 2015?

Question

Iron

As a fan of the "latest & greatest", I have never considered the practical application of models of technology older than the previous one from the current generation. But the crisis has come, and management wants to reduce costs.
Now we need to expand the VDI farm. At the moment we are using Session-based (former terminal server) slave virtualization. tables., but we are slowly moving towards the need to switch to a VM-based ("traditional" virtual work table). Now the hosts are equipped with two Xeon E5-2620 v2 and keep the load in the region of 60 users each (total 120 users). In the near future, it is planned to increase the number of users to 200 with the potential to expand to 250 (i.e. +130 users relative to "now"). It is also necessary to take into account the possibility of transition from Session-based to VM-based in the next six months or a year.
At a local flea market, Xeon X5690 were found at a price two times lower than the Xeon E5-2620 v2. Comparing the performance characteristics, I allowed myself to assume that these "outdated" Xeons will be 35-50% more powerful than our "new and better" Xeon E5-2620 v2. However, the low bandwidth of RAM and the speed of QPI are very confusing. I'm not sure how much this will affect performance due to the lack of necessary knowledge.
Therefore, I ask you to suggest what pitfalls a person who decides to use these specific processors for virtualization can expect, and in general, express your opinion on the use of irrelevant, often end of life components in combat systems in our time.
PS We are using Windows Server 2012 R2 and look forward to trying out the new features of Server 2016 when it comes out. Therefore, a separate question arises regarding the support of this equipment by modern server operating systems.

Answer 1

Everything depends on your tasks. It suited us, despite the theoretically slower QPI and other things. On such hardware, we have client terminals running (there is ancient software on Visual foxpro - it can utilize no more than 1 core and no more than 2 GB of RAM). There are about 500 users, but we have not yet reached VDI - only experiments. The old software imposes restrictions, up to its complete inoperability ... PHP also runs great on these servers - the performance linearly depends only on the clock frequency and practically nothing more. Because the bandwidth of the RAM bus has practically no effect here. As with MySQL, disk I/O is orders of magnitude slower anyway. In general, specifically, in old (HP G5) servers, only the amount of RAM, and not its speed, may not suit us.
On modern servers, we run serious things like MS SQL server with 768 GB of RAM, etc.

Answer 2

Not so long ago there was an article, maybe even on Habré, how one hoster bought up cheap old hardware and made VERY cheap hosting out of it.
The issue of stability was solved by competent virtualization and redundancy, if suddenly some piece of hardware failed, it was turned off and almost thrown away. But the cost of three old pieces of iron was cheaper than one new one, and the power was comparable or even more for the same money. Well, they were able to offer customers prices below competitors, success was achieved.
In other words, as the two previous people have already told you - determine the feasibility study of the benefits of old iron in the future. If you can technically use it, if the money (which includes the salary of those who will set up / maintain it) is more profitable than new hardware, then personal preferences "I want new and shiny current" will sound unprofessional.

Answer 3

Actually, in your case, you need a feasibility study of a particular solution.
In other words, you need to calculate what will be cheaper in operation (per unit of useful work, in your case, for "one user").
Cheaper is not only the initial costs for the purchase and implementation, but also the cost of maintenance, repair (replacement), losses from downtime when equipment fails, taking into account the likelihood of such a failure (they also want to save money on screws for backups, right?) and etc.

Answer 4

Old and / or cheap components can be used. The main thing is to ensure the proper degree of redundancy. I would be rather surprised when I watched a documentary about a cool data center of some large western company. And I saw there the familiar covers of disk compartments on servers, with a characteristic brown latch. That is, even in cool offices, cheap supermicro servers can be used :-)
Moral - it is quite possible to use cheap hardware if your architecture is built correctly, and has a certain degree of redundancy. Relatively speaking, if your virtualization cluster requires N servers for your load, then buy and install N + 3 servers in this cluster. Then even if one of the cheap servers drops out, everything will be automatically rebalanced for you and you can order and buy a new cheap server in a calm mode.