Hardware suggestions

[SnappProx]

Hello!
I'm looking to host about 200 vms tops with a mixture of windows and VMs, ideally being about 4vcpu's and 8gb RAM each for very basic functionality. 32-60gig disk space TOPS, was wondering if people here had suggestions regarding hardware configurations for this setup?

If I had to say budget, it's really not that much. maybe 10 grand absolute MAXIMUM, ideally around half of that if it's possible.

Thank you very much!

 

[UdoB]

200 vms ... 8gb RAM each

maybe 10 grand absolute MAXIMUM,

That's 1.6 TiB Ram. A single 64 GiB Modul (w/ ECC) seems to cost more than 400 Euros and you need 25 --> 10000 Euros only for the RAM. Seems to fit

Reality check please... ;-)

 

[SnappProx]

That's 1.6 TiB Ram. A single 64 GiB Modul (w/ ECC) seems to cost more than 400 Euros and you need 25 --> 10000 Euros only for the RAM. Seems to fit

Reality check please... ;-)

I was really thinking this might be the case, in this case, could you recommend anything for that budget and what I should expect to get? Give me that reality check that I need.
Thanks

Edit: I also phrased it extremely poorly, 200 vms hosted =/= 200 vms running, it'd have MAYBE like 60 concurrent vms

 

[UdoB]

Give me that reality check that I need.

Sorry, I can't.

You need to know exactly what services and what quality-of-service you need. Then you specify your hardware requirements, in detail. Are you just "playing" or is it a serious business? When you sell services you want them to be stable, reliable and and as undestroyable as possible.

I would always construct a cluster, by no way I would put 200 VMs on a single machine - without redundancy.

It gets complicated quickly and there are several designs possible, think "storage" and decide between ZFS w/ replication, Ceph or NFS built upon HDD/SSD/NVMe. Don't forget redundant power supplies, an UPS, one or two PBS backup servers and so on. Also the network stack might cost some real money, if you want redundancy and higher speed than slow 1 GBit/s...

 

[SnappProx]

Sorry, I can't.

You need to know exactly what services and what quality-of-service you need. Then you specify your hardware requirements, in detail. Are you just "playing" or is it a serious business? When you sell services you want them to be stable, reliable and and as undestroyable as possible.

I would always construct a cluster, by no way I would put 200 VMs on a single machine - without redundancy.

It gets complicated quickly and there are several designs possible, think "storage" and decide between ZFS w/ replication, Ceph or NFS built upon HDD/SSD/NVMe. Don't forget redundant power supplies, an UPS, one or two PBS backup servers and so on. Also the network stack might cost some real money, if you want redundancy and higher speed than slow 1 GBit/s...

It's not for services that are being sold, it's basically just for VDI. We currently have a 6 node cluster with most of our stuff on and I'm having some issues with it.

I have virtualised TrueNAS on my node with the most storage, and set that up with an NFS share that the other nodes in the cluster have, all VM Disks are stored to this NFS share so I can always easily migrate between nodes and whatnot.

I was never sure if this was the correct approach as I never found any "best practices" for this kind of thing. The disks are SATA SSDs over 2.5gig NICS. I will likely be changing this to a bare metal TrueNAS install as I believe I messed up when creating the current VM by not passing through the disks directly to TrueNAS. It's either bare metal TrueNAS or I set up CEPH across the cluster.

I have large instability with the node the TrueNAS VM is hosted on and I can't tell if it's because of the node itself is having issues or the VM is dying and causing the node to go off, so I was hoping to see how much it'd cost if I wanted to try and build a new "lab" from scratch.

Is there no very vague recommendations you can give? It's purely going to be windows desktops and linux desktop machines. Licensing isn't a concern for now. I'm not going for fantastic performance, I'm just going for Acceptable.