S
S
s_pereiro2017-12-05 10:32:47
System administration
s_pereiro, 2017-12-05 10:32:47

Choosing a virtualization platform?

Good morning. The question arose of choosing a platform for virtualization and I don’t want to make a mistake in this, please help with advice.
Now there is a server (Intel Xeon E5620,32Gb RAM) running VMWare ESXI 6.5, which runs several virtual machines with different OS and tasks (squid, wsus, asterisk, zabbix, etc) and several servers without virtualization (asterisk , cucm , pfsense) . All this works, does not cause problems, except for the issue of fault tolerance. It was decided to transfer all this to a new server and build a cluster, bought a couple of servers (Intel Xeon E5-2603 v4, 128 Gb RAM) and one (Intel Xeon E5-2603 v4,16 Gb RAM) for backup.
I really like the product from VMWare, but the price of the license bites. Of the requirements, I would like that if one server fails, the downtime does not exceed a couple of minutes. I also want to minimize the time to solve various problems when working with the virtualization platform, it is desirable that everything works out of the box. Backups were made on schedule, there was the possibility of monitoring, easy transfer of virtual machines, device forwarding, GUI.
Now I installed Proxmox for the test, however, while there are doubts about the correctness of the choice (several problems arose - the host did not reboot on command, it was not possible to quickly deploy the system from the .ova archive, and other trifles) I am confused by the fact that this solution is not so popular.
I also consider the following scenario as an option - to install esxi with its free license, without the ability to assemble a cluster, and implement fault tolerance by setting up the systems themselves?
Thanks in advance.

Answer the question

In order to leave comments, you need to log in

6 answer(s)
A
athacker, 2017-12-05
@athacker

A cluster of two servers will not work. You need at least three -- one of the servers will be the shared storage for the other two. This is independent of the virtualization platform.
As far as licenses go, VMware has packages like vSphere Essential that don't cost as much as just vSphere. Essential Plus covers three servers and one vCenter, costs about 250-300 thousand rubles.
But one server under the shared storage is not enough, since it becomes a single point of failure (it dies - and ALL virtual machines die). Therefore, the most correct solution would be a converged infrastructure - when the virtualization servers also present their local disks in one of the software defined storage options. Among the options: Ceph (free, but capricious, and such skills are needed for support), VMware vSAN (it costs nothing) or ScaleIO (it also costs in order, but up to 12 TB of raw space can be used completely free. In the variant of three nodes in a cluster ScaleIO 12TB of raw space is 4TB of usable space.As the number of nodes grows, the overhead will decrease.For example, 12TB of raw space on 5 servers will already provide 5TB of usable space).
So it all depends on the budget. Everything is like everywhere else - a balance between cost and ease of deployment / operation. Invest more money - get a solution that is easier to deploy, easier to operate and have support. Invest less - you will have to use open source, spend more time on the pilot, and often deal with the bells and whistles yourself, scribbling messages on forums and Google at space speed :-)
Well, actually, the players in the virtualization market are not that big of a deal. Microsoft Hyper-V, VMware ESXi, KVM. Proxmox is not quite a virtualization system, it is an orchestrator for managing virtual infrastructure. KVM (or LXC, but you don't need it) is directly involved in virtualization anyway.

P
poisons, 2017-12-05
@poisons

I'll ask one question - is there a shared storage in any form or is it planned for implementation? If not, consider that you do not have any cluster, in case of a host crash, there will be no virtual machine restart, only raise from backup if.
And proxmox is a good thing, yes.

D
de1m, 2017-12-05
@de1m

On two servers, you can also deploy a failover cluster. I'm telling.
VMWare - they have VSA (vSphere Storage Appliance) - something like RAID10 over the network, each server has a copy of another server, so only half of the disk space can be used. But it works on two servers, there may still be a minus that you need a good network to have the right speed. We had two servers connected to each other via 10Gbit/s. I'm not sure if this is an option in the free version of VMWare ESXi.
MS HyperV - it has a free version, but here you need storage, where virtual disks will lie and, accordingly, will not work on two servers. Still it is desirable to have AD. If you have never worked with Hyper-V before, then this is a very convenient and good thing. Better than VMWare in my opinion.
Proxmox, KVM are different products, but both have a problem with storage, since they don't have it as standard. But Ceph can help here well, they wrote above that it’s somehow difficult to install it and work with it, but this is not so. We have a cluster inside the company, it has been operating since 2015 and another one from Hetzner. There is ceph-deploy to install. But it needs three servers.
If you choose from Proxmox or KVM, then it’s still better to take Proxmox, there you need to do less with your hands and it’s quite easy to add ceph support.

P
Puma Thailand, 2017-12-26
@opium

It's very funny how virtual machines that consume 128 gigabytes of memory will move to a server with 16 gigabytes in two minutes

R
rionnagel, 2017-12-05
@rionnagel

Hyper-v has the ability to replicate. If the host lies down, you can restore work within a couple of seconds / minutes without the need to create a cluster. It is extremely easy to set up. You can contrive and shove some dc / os cluster into virtual machines, if you organize the storage on them correctly and use docker already ^^)

I
Ingvar, 2017-12-06
@take

Why I stopped at Proxmox a long time ago. Firstly, good old flexible Linux: OpenVZ machines had incredible speed, just something. Now they have sunk, but there are containers. Secondly, there is a paid subscription, which guarantees the possibility of support. They never took her while they managed on their own, but with the growth of tasks and responsibilities .. who knows. Thirdly, I don’t know ... zfs, which is now out of the box - I liked it, I didn’t raise the cluster, the task was not to “raise it in 5 minutes”, but soon, apparently, my hands will reach.

Didn't find what you were looking for?

Ask your question

Ask a Question

731 491 924 answers to any question