I ask for advice in choosing a platform for virtualizing the IT infrastructure of an enterprise

Question

Virtualization

The goal of the project is to get a well-managed, predictable platform for virtualizing basic IT services and processes within a small enterprise (10 Servers, 150 PCs)

. Ideally, I would like to hear the advice of “experienced” people who have already encountered any implementations and groped for pitfalls.

The enterprise works almost around the clock, at least 50 active 1C users, at least 100 active Internet users, plus a number of specific services that are better to talk about separately, since they are not policy-making, and can be implemented "separately" from the main infrastructure .

The criticality of services can be defined as follows:
Everything, including 1C, can be interrupted no more than 1 time per day, after a failure it should be available within 5-10 minutes.

Planned hardware: Two servers on the same platform, differing in the number of processors, memory and storage systems
1st - 2 * E5-2630 / 64 / 8-channel controller with 4 * SAS + 4 * SSD
and 2nd - 1 * E5 -2630/32/built-in 4*SATA controller

Network infrastructure is of high quality. At the heart of a 1Gb server farm, with three "user" switches 2 pcs. 48*100Mb + 1 pc. 24 * 1000Mb, plus a small number of wireless clients (mostly not critical)

Today's structure, and plans for "virtualization" of the
Server:
1st - Win Domain Controller, DHCP, DNS (Required client for virtualization)
2nd - Win Backup Domain Controller, Print Server, File Server (Also mandatory, separate roles)
3rd - Win Server 1C Enterprise, Terminal Server (It is mandatory to virtualize, separate roles)
4th - Win Server 1C Enterprise, a weaker machine, for small databases and several terminal clients
5th - Win MSSQL DB Server (now running on a 4-year-old machine with 6 * SATA storage on an integrated controller
(naturally, the most questions ... it is possible to virtualize not immediately, or, alternatively, not completely, i.e. spread the bases)
6th - Linux Server 1C Enterprise ( virtualize)
7th - Win Server (Routing with the manufacturer's web services (the most unlikely client for virtualization)
8th and further - A certain number of services "scattered" anywhere, which are also just waiting for "their own" place ...

On the one hand, the choice is not great =) on the other hand, there are a lot of nuances, most of which I would like to foresee in advance.
I will respectfully listen to any recommendations. I will be happy to answer “leading” questions ...

PS I don’t specifically cite any specific pieces of iron, configurations, etc., I think it’s superfluous for the time being.
The system available now, without any complaints, has been working for several years, very soon (within a year) critical servers will have a reasonable operating life. So a change of iron is necessary in any case ...

Answer 1

Regardless of the chosen virtualization platform - external storage. Because otherwise - the fall of the host leads to a complete fall of virtual machines and recovery - not earlier than the restoration of the host. And at the host - the raid can crumble, restore - a few hours. So since _really_ such stringent requirements - then only external storage. With two controllers and other stuffing.
Further: if there are only two servers, then their config should be the same, so that if one node dies, the second one starts all the virtual machines and does not die from the load.
The platform is ... to taste. But based on real conditions, there are only two options, Hyper-V and VMWare.
In your conditions, it is quite possible to try something from Acceleration / Essentials Kits, but in any case, it is either slightly or more expensive than a similar solution on Hyper-V.

Answer 2

VMWare vSphere allows you to make fault-tolerant systems without external storage, based on the Virtual Storage Appliance. But for this thing to work, you need at least 4 ethernet ports on the node, at least 2 independent network cards, plus the node itself must be made of compatible hardware. If the piece of iron is not in the compatibility list, then ESX simply will not see it. The same applies to the second network card, be careful when buying.
It works like 2 nodes plus an external vSphere, or 3 nodes. it is impossible to raise fault tolerance on two servers without external storage. VMWare also allows you to do not only High Aviability (i.e. restarting the machine on the backup node after the death of the main one), but also fault tolerance (the machine continues to work on the backup after the death of the main one, with a downtime of a few seconds), but this one works fine piece in 10-gigabit networks. On gigabit, it slows down. Plus, a machine with fault tolerance enabled always eats memory resources, a processor on two nodes at once.
An alternative option is Hyper V, but there you need external storage without options. But compatible hardware is everything a Windows server runs on. True, as far as I remember, there is no FT. only HA.

Answer 3

A few years ago, the question arose about the virtualization of infrastructure and storage systems. I weighed everything for a long time, tested it, and here is my verdict - vSphere. More efficient and more justified, in terms of money, solutions can not be found.
VMware cluster for 2 nodes (minimum license for 3, so there is room for growth). Two Proliants 360 + P2000 shelf (expandable) with two SAS controllers. Padding - 50/50 SAS / SATA (SAS for system images, SATA for data, although you can change the proportions to suit your needs). I also recently created a separate LUN on the SSD for caches, which significantly affected the performance of the VM as a whole.
Another plus is that Varya can properly trunk interfaces (unlike HyperV), so you can combine all interfaces into one and give the total bandwidth to clients.
After setting up HA and LM, you get a flexible, productive and fault-tolerant solution that is adequate in terms of money. True, there is a nuance - to correctly calculate the ratio of cores and RAM, so as not to get performance degradation. It’s hard to deal with this - a bar per channel (2 modules per core), otherwise the channel starts to divide in a ratio of 1: 2. ESX Features.

Answer 4

IgoreHa , what budget are you targeting?

Answer 5

What's the budget?

Answer 6

Actually, the choice, as I understand it, is from vmWare ESXi / vSphere and Xen CLoud Platform?
The first is much more polished, but the second supports "live" migrations between iron servers by default. (vmWare has vMotion technology, but only in a very paid, about $ 1000 for each processor version)

Answer 7

If a person learns virtualization again, then Microsoft Hyper-V will probably be easier, since almost everything is running on MS ...
And I know that MS is from the evil one :)