Zotac Zbox min-PC as VMware vSphere ESXi6 host

[aiea96701]

I have received advice from our project manager that he believes that we should remove/refresh all our Dell Servers hosting ESXi6 with Zotac Zbox mini-PCs bare version.  He believes this is where the industry is going and highly recommends it.  He is pushing this quite heartedly and our manager is actually considering.  So, I am being asked to to do "Due Diligence" to see if this is , in fact, the way to go or is this a "pie in the sky" project. 

Anyone also considering this?

[HawkieMan]

To be honest, it sounds smart and has a lot of merit, but:

1. Licensing will skyrocket

2. You will have to increase networking massively

3. Your largest pracitcally possible VM will be the size of the hosts

4. Host overhead will increase

So have those factors in mind . If you are running a docker style enviroment with micro VMs, then it can make sense, but personally I would say that the licensingalone is enough to discourage this path.

[aiea96701]

thanks for replying.  The issues with VM capacity planning was addressed and it was decided that the VMs would be stored on a SAN.  The Zbox is bare so we will be able to configure the memory to 64GB and Quad Core processor.  Don't believe there is capability to handle multiple CPUs only Cores.

1. Licensing will skyrocket.................how would licensing skyrocket if we deploy mini-PCs in lieu of rack mounted Dell PowerEdge servers?  Is it because of the limited socket arrangements in mini-PCs?

2. You will have to increase networking massively...........again, is this because of the limited NICs that can be installed on mini-PCs?

3. Your largest pracitcally possible VM will be the size of the hosts............are you speaking of the "resources" on the host?  If we deploy the mini-PCs with 64GB memory and Quad Core, that would be the limitation, correct?

4. Host overhead will increase...............Overhead with respect to the number of hosts due to the limited resources available on the mini-PCs, correct? 

So, what I hear you saying is that it will take more mini-PC hosts to equal what a server-designed hardware could provide (ie  A Dell PowerEdge couid allow for multiple internal NICs and multiple CPU sockets whereas the mini-PCs would only provide 1-2 NICs and single socket).

[HawkieMan]

Let me address your questions one by one here:

1. License will skyrocket because vmware is based on a pr socket license. So to maintain the same capasity in VMs you must add more sockets. Example: a Xeon Based server with 2.97Ghz CPU and 1 sockets (1 socket, 6 cores as example), will have same license cost as a micro system with 1 socket and 1,4Ghz core (1 socket, 4 cores). So license cost pr compute resource increase dramatically. Best sense when you look at cost, where license and memory usually are the largest cost factors, is to get as much CPU as possible pr system. Micro systems  defeats that math.

Imagine if you today have 3 hosts with 1 cpu at 2.Ghz and 4 cores, that means you get 3 x 1 socket licenses, and a total compute of 24Ghz of compute power for the cost of 3 socket licenses.

If you replace these hosts with 6 microservers with 1 cpu and 4 cores at 1.4 ghz it means you get 6x1 socket licenses, and a total compute of 33,6 Ghz. In short a double of license, but just a small increase in compute power.

2. The NICs in these micro PCs is not made for server use, and doesnt have the througput, processor or buffers to handle the same loads pr network port as a server NIC. In vmware you need both managment, load and vmotion, and usually these micro PCs have just a single NIC. So that must be checked.

3. You cannot run a VM that is larger than a single host capasity (as it must run on one of the hosts), so the largest VM you can have = the microserver capasity itself. This is not really much, and you will then propably end up with running a single VM pr server. And this defeats the purpose of virtualization itself, since you dont have spare capacity to share with other VMs.

4. Each host needs some resources for itself and its OS, and more hosts = more overhead.

In addition these "servers" cannot use ECC memory, or Buffered memory, and so on.

In the future, when we might see complete solutions made of small microservers in a clustered solution, but they will be custom made for this, and not be based on off the shelf HW for home use.

[aiea96701]

very much appreciate this information.