Custom Whitebox Server

[adamjoeyork]

Good morning all. Having recently done my taxes I am looking at finishing the FreeNAS (TrueNAS now I guess) server build I had planned. I already have the following:

CPU: Ryzen 5 1600AF
SSD: Kingston 240GB
DISKS: WD Shucked 10TB x2
PSU: fully modular EVGA 550W

I am considering the AsRock Rack X470D4U motherboard for its dual gigabit nics and confirmed ECC support, among other server features. I would start out with 2 16GB sticks of Kingston 2666Mhz ECC with 2 slots available to one day move to 64GB. There seems to be a lack of options in the server case market, at least on eBay and Amazon, but I was considering the SilverStone SST-RM400. Having at least 4 drive bays would be nice, I'd like this to be somewhat upgradeable. Can anyone speak from experience on the performance of virtualized FreeNAS? Curious to know what the community thinks of this and any recommendations people might have.

 

[jgreco]

https://www.ixsystems.com/community...ide-to-not-completely-losing-your-data.12714/

Performance will generally be proportional to what you allocate to FreeNAS. Everyone wants to underallocate. ZFS isn't meant for that, especially on the RAM side of things.

 

[adamjoeyork]

https://www.ixsystems.com/community...ide-to-not-completely-losing-your-data.12714/

Performance will generally be proportional to what you allocate to FreeNAS. Everyone wants to underallocate. ZFS isn't meant for that, especially on the RAM side of things.

My theory was 8gb RAM for the host OS and 1GB of RAM per raw TB. So 8 + 10 +10 = 28GB. If I added any more disks than just a simple ZFS mirror I would need to buy 2 more sticks of RAM as well. For the CPU side I should hope 12 threads would be enough. Thank you much for the link!

 

[jgreco]

If you're willing to allocate that much, you'll be fine. See, the trick is that you can probably get away with allocating LESS, which tends to impact performance down the road as the pool fills and the challenges of doing all the bookkeeping for ZFS increases, but as long as you have the RAM and CPU available then you just increase the VM size. The problem is lots of people desperately want their FreeNAS VM to work on ridiculously tight resources, and are then disappointed when things don't work out well for them. I got tired of arguing with that mindset years ago. If you build the host to be able to handle the long term requirements, that will have you as a happy camper down the road. In the meantime you can take advantage of the excess resources for other VM's.

 

[adamjoeyork]

If you're willing to allocate that much, you'll be fine. See, the trick is that you can probably get away with allocating LESS, which tends to impact performance down the road as the pool fills and the challenges of doing all the bookkeeping for ZFS increases, but as long as you have the RAM and CPU available then you just increase the VM size. The problem is lots of people desperately want their FreeNAS VM to work on ridiculously tight resources, and are then disappointed when things don't work out well for them. I got tired of arguing with that mindset years ago. If you build the host to be able to handle the long term requirements, that will have you as a happy camper down the road. In the meantime you can take advantage of the excess resources for other VM's.

Wonderful write up there. Scouring around on the internet I have come to expect the magic of ZFS is not something you can implement on poor hardware. A give and take situation. Thanks very much for your time, if I might ask one more question, would you prefer a 3 drive array (Z1 I think) or a mirror with a cold spare ready to go if one of the mirrored drives goes out?

 

[jgreco]

I don't really care for either one as I strongly prefer that loss of a drive not compromise redundancy. This sets a minimum acceptable level of RAIDZ2 or three-way mirrors. Large pools are RAIDZ3 with a spare.

Of course, not all data is that valuable.

RAIDZ1 requires a minimum of three drives. To make for a comparison, bump a RAIDZ1, RAIDZ2, and mirror setup to four drives along with a cold spare:

The RAIDZ1 is by far the most efficient for space (three drives worth) but loses redundancy if a drive fails. It will be very good at large sequential file storage.

The RAIDZ2 is less efficient for space (two drives worth) but doesn't lose redundancy if a drive fails. It is also good at large file storage.

The mirrors are approximately the same efficiency for space as RAIDZ2 (in! this! specific! example!) but may lose redundancy if a drive fails. It is much better at random I/O than either RAIDZ.

Going smaller, such as your proposed situation, comparisons are harder.