Bad CPU mid-build so building all over again

[winstontj]

I built a server in a Fractal Node 804 case with a supermicro x11ssh-f, e3-1245v5, 64gb and 8x 4tb hdd in mirrored vdevs --and the CPU turned out to be bad. So now I'm re-thinking my initial hardware choices.

Rather than type out a long post, I'll mention that I'm trying to build a NAS that will have space for datastores for 10-15 running VMs. I own 8x 4tb HDD, a 9211-8i hba, x11ssh-f and x11scl-f motherboards and 128gb of REGISTERED (4x 32gb) memory.

I'm basically starting over and need a motherboard + CPU + memory + slog + small file/metadata drive.

Wondering if a xeon e3-1245v5 will run that? (I think CPU will be fine but the mobo might not support the devices I want)
Will the xeon e-2200 and the x11scl-f motherboard be enough? (at least the pcie slot can bifurcate and m.2 slot is x4)
or
Should I be looking at a mATX 2011-3 or lga3647 motherboards? With a 6-channel lga3647 cpu that takes nvdimm I could run CORE and use nvdimm to take care of SLOG. In a lga 3647 mATX motherboard (6 dimms) can I run the 4x 32gb RDIMM I have and then run one 32gb NVDIMM for SLOG and one Optane 128gb pmem p100 DIMM for a metadata drive space? Can you combine pmem dimms and nvidmm in the same motherboard/chipset?

Thanks for any feedback.

Budget:
I don't really care at this point. I just want to get this thing done. It is going to cost $1,500-ish either in CPU+RAM+SLOG or in motherboard + cpu + optane + NVDIMM. (on top of what I have already bought)

 

[Ericloewe]

Does your workload really justify NVDIMMs? Because that’s a whole different level of complexity. And mixing different types? I don’t even want to imagine that!

My recommendation is to just buy a new Xeon E3 v5 and see where that leaves you in terms of performance. Work from there and if you need an SLOG, just get a suitable PCIe SSD (NAND or Optane).

 

[rvassar]

Do the x11ssh-f and x11scl-f support registered DIMM's?

 

[winstontj]

Do the x11ssh-f and x11scl-f support registered DIMM's?

No. That's what really stinks.

Does your workload really justify NVDIMMs? Because that’s a whole different level of complexity. And mixing different types? I don’t even want to imagine that!

It might sound crazy but I've worked with running low-latency stuff on fpga with battery-backed ddr2 & ddr3 (pre-nand) so I'm pretty comfortable with the idea of nvdimm. I'd feel more comfortable trying to sort out running nvdimm than optane ssd via pcie or m.2. A lot of what we ran was bsd... but it was back in the 2000's so it's been a while.

I'm regretting going with the fractal 804 case but I didn't know any better when I started. I'm worried that running the amount of VMs I need will slow the pool down to a crawl. I guess I could always use the 804 with an e3-1245v5 as a backup nas and build out an all-flash machine for the VMs.

How much will a metadata pool/device speed up a small pool of spindles?
Thanks.

 

[winstontj]

I was able to test/simulate a nvdimm slog on core easily however in scale it is difficult. In Scale I have to change/modify a bunch of things just to get a simulated nvdimm slog to show up however it does not persist through a reboot. (I assume it would not persist through an OS update either??)

Is there any chance that these features are available to non-enterprise Scale users? I have a mobo coming which will arrive shortly but I can't find anything in the forums confirming that nvdimm or pmem dimms will work in community/non-paid versions of Scale.

Can anyone suggest a decent hardware combination (mobo, cpu, amount of memory, slog) for a storage pool that will have 10-15 running esxi virtual machines on it?

Thanks.