Custom NAS ideas (using mining rig motherboards)

[velusip]

Since the hilarious crash of the cryptocurrency hardware market, you can buy multi-X1 boards and risers for cheap. I'm curious of your thoughts on using a combination of PCIE X1 to SATA adaptors and X1 10Gbit NICs on hardware like this? Anyone tried it?

For example: 18 slot X1 board, 15 risers and 15 2-port SATA adaptors, 1,2, or 3 10GBit (RJ45) NICs + power supplies could be less than $600.

 

[BlueLineSwinger]

Sounds fragile and messy to me.

 

[emmjawsX]

i was considering picking up a cheap mining board but ultimately couldn't think of a use case for it.

Used enterprise gear is still way better value for a NAS like you're considering - using a few $20-50 perc/mellanox cards can get the same basic functionality with any old 3/4 pci-e slot intel board made in the last decade for a lot less cost and complexity.

 

[extide]

Probably a better idea just to go with a SAS expander and one proper HBA. Also 10Gbit would be a bit limited over PCIe 3.0 x1, and quite significantly limited over PCIe 2.0 x1.

 

[velusip]

Probably a better idea just to go with a SAS expander and one proper HBA. Also 10Gbit would be a bit limited over PCIe 3.0 x1, and quite significantly limited over PCIe 2.0 x1.

I wouldn't expect the network to exceed both read and write speeds of striped sets of spinners, so PCIe 2.0 X1 lanes should be fine.

The idea was to use the case from an old iSCSI expansion drawer. Nothing fragile about that. The total cost (without drives) would be less than $600, and handle 40 drives or more. That's WAY cheaper than anything I've seen even at auction with similar capacity. I'm really curious about what kind of comparable systems you guys are finding in used enterprise gear, though.

 

[extide]

Well, 10Gbit is about 1.25GB/sec and you can do that with 10 or so spinners, not to mention 40.

Also dealing with a pile of 2ch sata cards would be a PITA. I'd much rather deal with a real SAS HBA and expander. You could easily build something like that for $600 with a Norco 4224, a used SAS2008 based card, SAS expander (or perhaps multiple SAS2008 cards if you want to direct attach them) and then whatever mobo/cpu you want. Preferably an HEDT platform or older Xeon so you have plenty of PCIe lanes and memory capacity. (Of course I would use ZFS and having plenty of memory is always nice). I mean unless you already have a bunch of this hardware, I wouldn't go that route. Sure you could make it work, but at what cost?

 

[Biznatch]

You're thinking about this all wrong. 600$ isn't even inexpensive for that mess, but you're right, it would be 'cheap'..... You can get killer deals on last gen server hardware that has 8-12 on board sata adapters and enough pcix lanes to connect at least 3- 8 port sata cards and 1- 2 or 4 port 10Gb card for far less, and it will be MUCH more robust than what you are proposing.

I'd say plan it out, start with the case and build from there.

 

[velusip]

You're thinking about this all wrong. 600$ isn't even inexpensive for that mess, but you're right, it would be 'cheap'..... You can get killer deals on last gen server hardware that has 8-12 on board sata adapters and enough pcix lanes to connect at least 3- 8 port sata cards and 1- 2 or 4 port 10Gb card for far less, and it will be MUCH more robust than what you are proposing.

I'd say plan it out, start with the case and build from there.

Thanks extide and Biznatch. I guess there's a couple other problems which made me start the thread. One is, I don't mind scratch building stuff -- so that's a personal problem. The other is the issue with what's available in my region. I haven't found good stuff or prices like you guys are talking about. The best I can find for cheap is modifying an old 4U JBOD unit, but that's not really a net save. It's just $400 + S/H plus the same work and expense to modify, upgrade. So I looked at the cost of lots of X1 slots and borrowing a metal break and building the equivalent of a high density drawer of bays starting with a Grainger cabinet. Hence the thread.

I mean, weird cobbled solutions may not be common or seem very robust, but the smaller offices I've worked for have some "interesting" setups due to space limitations.