Desktop Hypervisor, Remote OS using local hardware

I struggled to even think up a good title for this post, as badly as I struggle to truly find enough information about what I would like to do, and if it is actually possible, although I have an assumption it is.

Maybe after I lay out my “dream” setup, what I have in place, and what I have available hardware wise someone with more experience at this could give me a definitive answer.

Alright so hardware first:

Desktop System
-R5 2600X
-16 GB 3200mhz Corsair Vengeance LPX (rated for 3000, but passes all error and stress tests)
-Asus Crosshair VII Hero (non wifi)
-1 TB Crucial P1 NVMe
-256 GB Crucial M4
-1 TB Western Digital Blue
-RX 480 and R9 270X
-EVGA Supernova 1000 G2

Server-ish System
-H8DGi-F
-2x Opteron 2676 running stable at 3ghz all cores (16 cores, 32 threads)
-32GB Hynix ECC RDimm 1600mhz
-128GB Kingston SSD
-EVGA 650 G5

Craptop System
-HP Pavilion G6 A-Series A10-4600M
-AMD Radeon HD 7660G
-4GB DDR3
-760GB SSD

Local Network
-Ubiquiti ER-X Gigiabit router
-TP Link 5 port Gigabit Switch
-Cat 7 cable run between all connected systems.

As a note, I am currently trying to work with Proxmox VE, other suggestions are ok too.

So on to what and why, and you can feel free to tell me why it is not possible.


My ultimate goal is using the above hardware to run a baremetal hypervisor with minimal system requirements, as little a possible, that still has KVM support for that performance.

I do not want to have my desktop operating systems ON my desktop, I want to be able to load them from a remote source only as necessary, all of my local data would be the hypervisor and any software installs, or any of my production data including 3d models, other art, code, games etc.

I want to be able to have a choice of a few differ
ent distributions of Linux both LTS and Rolling distros, as well as both Windows 10 Professional, however the catch is I want these operating systems to use the hardware from my desktop in passthrough, so from everything I know just using traditional VM setups won’t work.

I’ve been at this for about 2 and a half weeks, read hundreds of write ups, watched dozens of videos, played with many different setups and I am still not even sure if what I explicitly want to do is even possible, though I have this overwhelming feeling it is (feels amirite?).

So TLDR, I want an operating system on my desktop without an operating system on my desktop, I have access to a semi powerful enterprise server and a laptop used solely as network storage.


Thoughts, suggestions, good ol’ ribbing?

1 Like

This sounds like it’ll have terrible performance, unless you’re going to have a lot of SSD storage or stacks on stacks of spindles.

It’s absolutely doable though.

To be clear, I had an openstack cluster with Ceph storage over 2x bonded 40G. (at work) We had significant latency issues, and database performance suffered due to that, so we had to outfit all the nodes with 8TB of NVMe so we could store the sensitive databases there.


I’d recommend looking at proxmox for a KVM hypervisor. You can run your VMs from there if that’s what you’re trying to do.

You can use NFS or iSCSI for remote disk access.

I highly recommend not using the laptop for storage and just expand the Opteron box if this is the route you wanna go.

2 Likes

I am happy with 70% of the performance I would have running a bare-metal operating system, its more about the convenience of choice than performance.

I can worry about the performance aspect with network and hardware upgrades in the future.

A lot of ssd storage and spindles is viable, and I am capable of pushing the internal network to 4gb/s without having to actually swap out the router.

This is where my head has been at for the last week or so, I am just not sure how to go about that entirely, any VMs I’ve used or setup always use the hardware of the device the VM is hosted on, not the remote system.

Scratch the reply I put below here, it makes no sense for me to be using slower drives unless I am using it for long term storage or backups, I’ll only look at expanding it out with SSDs

1 Like

You’re on 10G already. Assuming you’ve got 10GbE nics, you’re not going to see much improvement with any upgrades.

I would test for latency, that’s what makes your storage feel slow, when you’re loading and whatnot. (well, that and seek time, but for network, it’s just latency)

Okay, so did you want to, for example, run a VM on your Opteron server and interact with it (display/input) on the 2600x system?

Proxmox has a remote viewer for that, aside from that and VNC, there isn’t a great solution available.

You might consider a ZFS array with SSD caching, or an array with spindles for storage and one with SSD for frequently accessed data. But that’s something you’ll want to look at once you’ve completed a PoC.


I’m about to step out of the office, so my response/research capabilities will be limited.

1 Like

I want to run the VMs on my Desktop but not host them there. I’d rather if the space they physically take is temporary. Not just for convenience, but for backup purposes as well, I tend to break every single operating system I touch fairly often.

ping times on every system from every system are under 0.3ms

I have 10G cabling, but only 1G setup atm as I only have gigabit ethernet in my systems, but the router is capable of higher throughput.

Good, you’ll likely see okay performance there.

Id recommend getting some used Intel x540 (I think) nics.

I have been learning about ZFS, I feel like I get about 5% of it so far, it seems very much over my head even watching videos from Wendell and other experienced ZFS users.

It can be, but the benefit is tiered storage and flexibility.

Make some virtual disks and play around with it, that’s the best way to understand zfs.

ok had a look at some, T2? and passive or active cooled, or do you think it will even matter?

Do you mean store the VM’s on the server, but only pull an active one down to the desktop to run a session? then back to the server for storing?

Or remote sessions of a VM running on the desktop’s hardware (bare metal) but with the security of being in a VM? [edit, not sure where intended access would be from, if remote]

Because using Proxmox, headless on the desktop machine, and auto running a VM of choice with as much passed through as possible might give best performance, and storage hosted on the server (accessed via NFS/ISCSI) for “reliable” set storage?

If your server is rack mount passive, otherwise active.

1 Like

Uhhhh give me a second to correctly word this.

I want the OS stored remotely, but I want to pull it to the Desktop when I want to use that operating system on the desktop, and when I shut down the system the OS is no longer on the desktop.

The best thing I’ve been able to find is running a Proxmox VE cluster, but I just don’t understand how I can pass through my hardware on my desktop to the VM stored remotely, in this case remotely is about 25 feet away in the “server” room, bricked up garage so it has its own ventilation system separate from the house. In the summer it averages about 10-15 C ambient, this time of year I have to wear my winter coat.

1 Like

Okay, so you don’t want a thin client, that connects to a session on the remote host, with all the GPU horsepower remote, you want more like a fat client, with just the data remote.

Kinda like network booting, rather than a local OS, and shared/remote folders/drives/resources?

Exactly that.

1 Like

Well, I’ve never really done the whole pxe booting thing, but sounds slow to use on a desktop, if you have to download the OS every session. But maybe more flexible?

I thought it was tied to the network hardware ID, and only really did one OS at a time
So probably not what you are looking for.

I would go with Sarge’s suggestion of proxmox, and just use shared/remote folders as much as possible.

My system is generally 24/7, it would only load a new OS when I want to switch operating systems, slow initial load is definitely fine.

Exactly what I am looking for is flexibility. I’ll dig deeper into Proxmox, my understanding of it is quite basic. I guess its time to RTFM, all of it.

1 Like

There are other headless hypervisors, but as far as they go, the “Easiest***” one seems to be Proxmox, for actually using consumer hardware on the same machine, with only a large amount of hassle.

***not easy. Enterprise systems like Vsphere/esxi would probably be easier to manage, for those that know the system, but not on the same machine, and not headless with consumer GPU, afaik.

Again, Sarge is the expert on such things

1 Like

Thank you Trooper_Ish, I’ll do my research and wait for input from Sarge whenever he is able.

1 Like

Well so far the most relevant, easy to understand information about PXE boot without installing the OS as well is from 2012, very interesting lol.