HighPoint affirmed the SSD7120 'works in FreeBSD' ... when asked if it'd work as an HBA

[TrumanHW]

They replied initially saying it'd take some time to get back to me -- then, sent a document saying they had the results sooner than anticipated. What's kind of cool is they used the same motherboard as one of the candidate MB I'd use: (X299X Designare from Gigabyte) ...

Will someone look at what this document says and decode whether they're using native FreeBSD drivers or if these are dependent on HighPoint drivers? As, they showed it "running via FreeBSD in RAID-0 and RAID-1" ... obviously, not the goal.

I have
2x SSD7120 (PCIe 3.0 x16 with 4x U.2 x4 8643 connectors)...
4x PM983 U.2 drives @ 3.84TB ...

Is there a way to keep the 4 SSD sync'd to my main array (80TB RAIDz2 of 8x 10TB) to minimize failure likelihood...
Basically, I'd want to write to the NVMe array at the max speed they'd support
Then, sync the NVMe array to the spinning array as fast as the spinning array can be written to...
After which, either deleting data that I don't need to be kept on the NVMe array ... or leaving data that's used often.

I'm thinking RAIDz1 ...? to mitigate the cost-per-TB.
Hopefully I can find a few more for $350-400.
I'm assuming RAIDz2 is overkill given their
Low URE
High DWPD
High Write speeds (short rebuild time if / when required) ..?

Thanks

 

[survive]

Hi there,

If I understand correctly, you just need to turn a PCIe x16 slot into 4 U.2 8643 connectors to hook up your drives.

If so, forget trying to dumb down the High-Point & get something like the Supermicro AOC-SLG3-4E4R-O for a third of the money.

Make a pool & replicate as you need.

-WIll

 

[TrumanHW]

If I understand correctly, you just need to turn a PCIe x16 slot into 4 U.2 8643 connectors to hook up your drives.

GREAT advice. I see an ad for under $200 ...& could sell my 2x High Points (though I did grab one for $160, which is less than the cheapest deal I found on the SM) ... but can still sell it for more than the supermicro would cost.

Would mine work as an HBA though? As mentioned, I do own them already.
Is the supermicro HBA likely to be a better product, also?

 

[TrumanHW]

My only two questions are:

• I'm unclear if you were saying (since I mentioned having a GigaByte MB) the SM 4E4R will work with 3rd party MotherBoards..?
• Did you attempt reading or making sense what High Point was saying re: SSD7120 with the PDF I'd originally attached..?

Below: My research since my last message; No need to read it unless you'd like to see what others stated...


Conversation on ServeTheHome
re: SuperMicro HBAs with some interesting remarks:

Comment 1: FWIW ...
AOC-SLG3-4E4R -- unsuccessful: with any motherboard, even those with bifurcation that worked with other NVME HBAs.
AOC-SLG3-4E4T -- Successful: with X10SRW-F (motherboard is on SuperMicro list of qualified HW for the AOC.
Our testing wasn't exhaustive, but, don't have high expectations; it's by no means guaranteed to work (and you may lose money on the AOC)

Another user provided a reply from SuperMicro:
Those cards are untested nor validated by Supermicro on: MBD X10DRH-CT & isn't guaranteed to work properly. Read the below for additional info:
• AOC-SLG3-4E4T: (Manual (PDF) ... May require removal of JP5 + JP6, & set JP7 to pins 1-2 ... to check if the drives are detected correctly
• AOC-SLG3-4E4R: Manual (PDF) ... Should be OK on slot 6 .


SuperMicro's HBA product page for above HBAs: Some only work with validated SuperMicro server systems & MBs.
They ALL refer to the validated platforms list: Neither AOC-SLG3-4E4T nor AOC-SLG3-4E4R reference any hardware other than SuperMicro.




Ironically, found a reply... specifically about -- HighPoint SSD7120: With but one poignant remark:
"...the SSD7120 works in HBA mode by default" Which may ultimately have answered my question.

Though, it'd be nice if someone could decode the PDF HighPoint sent re: FreeBSD...

 

[survive]

Hi there,

Ok, so let me preface this with I've been mucking around with adding NVMe devices to some Supermicro systems so that's where my head is at right now.

I took a look at the original post\pdf and if you look at step #5 you see that card is identified as a PLX PCIe switch, which carves up the PCIe lanes from the 16x slot into four 4x connections. Then they just use gstripe or gmirror to make your array, so there's no real dumbing down of the card, it's just a PCIe switch!

The High-Point gets you around the problem you might run into with PCIe bifurcation. Outside of any other problem where the Supermicro card wouldn't work, you also need the motherboard to be able to slice the 16x slot up into the four 4x connections your drives need. The Supermicro systems I'm using do this, your Gigabyte might not. That's what I expect they mean by MUX in the second .pdf file.

That said, if you have the High Point and it works in your system right now it might not be worth messing with. Particularly because you aren't in them for much more than the Supermicro cards would cost. Looking into what the High-points actually are (PLX switches) I think you have a surprisingly elegant solution that should work well.

-Will

 

[TrumanHW]

Hi there,

Ok, so let me preface this with I've been mucking around with adding NVMe devices to some Supermicro systems so that's where my head is at right now.

I took a look at the original post\pdf and if you look at step #5 you see that card is identified as a PLX PCIe switch, which carves up the PCIe lanes from the 16x slot into four 4x connections. Then they just use gstripe or gmirror to make your array, so there's no real dumbing down of the card, it's just a PCIe switch!

The High-Point gets you around the problem you might run into with PCIe bifurcation. Outside of any other problem where the Supermicro card wouldn't work, you also need the motherboard to be able to slice the 16x slot up into the four 4x connections your drives need. The Supermicro systems I'm using do this, your Gigabyte might not. That's what I expect they mean by MUX in the second .pdf file.

That said, if you have the High Point and it works in your system right now it might not be worth messing with. Particularly because you aren't in them for much more than the Supermicro cards would cost. Looking into what the High-points actually are (PLX switches) I think you have a surprisingly elegant solution that should work well.

-Will

Thank you VERY much Will. That's a very helpful reply. I kinda thought the same thing but knew better than to assume I was sure.

Basically, the PLX divvies up the x16 lanes to the 4 x4 devices allowing for PCIe devices to connect through one PCIe slot..?
And otherwise, it's just an HBA, facilitating access to those devices for FreeBSD in this case..?

I'm actually not using them ATM (I'd used them in windows but it wasn't my goal). I think FreeNAS is my most likely scenario (as a "data layer" that I can use to minimize waiting for my other spinning storage). IF I don't make a 6-8 SSD array (and only 4-SSDs) I might try ZFS for macOS or Windows...but I'd need to find some 4-bay U.2 enclosure (that's not a rip-off) + and then use a TB3 PCIe enclosure for the SSD7120 to access from OS X via TB3.

I presume it's doubtful I could use the TB3 on the MB to copy data from a FreeNAS array to a Thunderbolt-3 attached SSD when I need to copy data quicker than 10GbE allows...

If that doesn't work, I could add a QSFP+ card to FreeNAS, direct-connect to another QSFP+ card connecting to my mac via a PCIe to TB3 enclosure ... provided I find a QSFP+ card with macOS drivers (chelsio I'm assuming).

Thank you very much, again. :)