Motherboard / Chassis selection

[tigloo]

Hi,

I'm currently trying to build a new system and I have a question regarding motherboard and chassis selection.

Background:

I'm currently running an Ubuntu-based ZFS file server consisting of very old parts. It is a trusty machine, but its hardware is a dead end. Its main tasks is hosting the family's critical data (photos, documents), music and movie share with the occasional transcoding job.

I have the U-NAS NSC-800 on my shortlist due to 8 hot swap bays and the small form factor. Both were my criteria for selecting a chassis.

My favorite mainboard right now is the Supermicro X10SDV-4C-TLN4f. My problem with this choice is the lack of SATA ports. I can recycle the M1015 from my old server but in order to fit it into the NSC-800, I would have to use a PCI extender which is probably not a good idea to use for hard disks where you want a reliable link.

On the other hand the ASRock C2750D4I would probably fulfill all my requirements but it seems like investing in an outdated platform when the Xeon D brings same or more performance at the same price point.

Does anyone have any recommendations for mainboards fitting the NSC-800 or a case that fulfills the same requirements?

I searched the forums but did not come up with any good suggestions yet.

 

[Stux]

If you're set on the UNAS I would go with the Xeon D

I think PCIe links are error corrected, so providing it works, it should work.

Leaves the built in Sata ports for SSDs

 

[tigloo]

I'm set on the U-NAS since it fulfills my requirements. I'd be willing to change if there's a different case with 8 bays and a small form factor.

It's the same with the mainboard really, I'd be set on the ASRock if it would be a Xeon D. That's of course where the problems start. :) Maybe I should try the riser and hope that it provides enough throughput for 2-3 HDDs.

 

[Stux]

if memory serves, the issue is that the UNAS comes with a 16 lane riser cable, and the Xeon D board has an 8 lane slot. If that's the case, you'd need to source a different pci extension cable

Here is a build log which should help

 

[Ericloewe]

I think PCIe links are error corrected, so providing it works, it should work.

No, but they're checksummed, so silent corruption shouldn't happen.

I'm set on the U-NAS since it fulfills my requirements. I'd be willing to change if there's a different case with 8 bays and a small form factor.

The Lian-Li PC-Q26 is fairly compact, with a more traditional layout. It's slightly larger, since it takes 10 drives, but airflow is okay, unlike what's said of the U-NAS stuff.

The only major caveat is that it only includes a backplane for two drives and the backplanes aren't appropriate for hot-swapping (they work fine, as long as you don't plug in a 3.5" HDD with the server running - SSDs shouldn't be a problem and I'd be surprised if 2.5" HDDs were problematic).

 

[Stux]

No, but they're checksummed

Thanks, I meant that :)

 

[tigloo]

Thanks for the pointer to the Lian-Li. Hotswap is not so much a necessity but easily accessible drive bays without having to open the case are.

I saw that U-NAS supplies SFF8087 cables for the case which indicates that they are using it with an HBA as well. Those should also be a lot easier to fit than 8 individual SATA cables.

 

[SFoskett]

I considered the U-NAS too but went with a full-sized tower and ATX motherboard instead. Blogged the decision here: My FreeNAS Build: Supermicro X10SL7, Intel Haswell Xeon, ECC RAM

 

[tigloo]

Thanks, that is valuable insight. It is very close to what I did with my previous build (about 10 years old now). My current server is housed in a midi tower and uses drive bays by Icybox that fit into 5.25" mounts. It's not fancy, it's not size optimized but it does the job.

 

[SFoskett]

Thanks, that is valuable insight. It is very close to what I did with my previous build (about 10 years old now). My current server is housed in a midi tower and uses drive bays by Icybox that fit into 5.25" mounts. It's not fancy, it's not size optimized but it does the job.

In that case, why not reuse the case and buy an X10-SL7 like I did? 14 SATA ports, good quality, and works with cheaper Broadwell CPU's that still perform great.

 

[VladTepes]

providing it works, it should work.

:):D Quote of the Year :D:)

 

[Stux]

:):D Quote of the Year :D:)

Yes, as opposed to non ecc ram, which works fine, until it doesn't and you lose your pool

 

[tigloo]

In that case, why not reuse the case and buy an X10-SL7 like I did? 14 SATA ports, good quality, and works with cheaper Broadwell CPU's that still perform great.

My first build about 15 years ago was a big tower. I wasn't happy with the size because it was tough to fit anywhere. So after some time I put it into a recycled midi tower (the one I am still using now), which made it better but not optimal: it can only fit 4 drive bays that are accessible from the outside and it is still somewhat big - my wife only tolerates it because it sits in a small storage room.

The goal for my next build is to maximise drive count while minimising size, which are of course contradictory goals. The two case options that seem attractive are the U-NAS and the SuperMicro 2U or 3U cases. The latter are attractive because I have a cabinet where the case could slide right in and then it would be out of sight. The downside is the noise, which again is something I want to pay attention to as well.

The more often you build, the more complicated and restricted the goal space. :)

 

[tigloo]

Yes, as opposed to non ecc ram, which works fine, until it doesn't and you lose your pool

Well, it's not as bad as that. Looking at space distribution, I guess 99% is file data and 1% is meta data (quote missing), so you are far more likely to encounter bit rot in your file data before you see it in your meta data. And even meta data is checksummed in ZFS, so you need to have really bad luck to lose the entire pool. I'm not saying that ECC is useless (I'm in fact investing in the same thing now) but it's not like omission of ECC will immediately cause you to lose your pool on the first flipped bit.

I'd be more concerned about building pools with a low number of large HDDs, such as the 6TB models sometimes posted here. If you lose one HDD over time (and that will happen sooner or later), chances are high that you will hit at least one or more read errors on another drive during resilvering and that's when you can lose your pool - unless you use at least RADZ2.

 

[Stux]

Well, it's not as bad as that. Looking at space distribution, I guess 99% is file data and 1% is meta data (quote missing), so you are far more likely to encounter bit rot in your file data before you see it in your meta data. And even meta data is checksummed in ZFS, so you need to have really bad luck to lose the entire pool. I'm not saying that ECC is useless (I'm in fact investing in the same thing now) but it's not like omission of ECC will immediately cause you to lose your pool on the first flipped bit.

I'd be more concerned about building pools with a low number of large HDDs, such as the 6TB models sometimes posted here. If you lose one HDD over time (and that will happen sooner or later), chances are high that you will hit at least one or more read errors on another drive during resilvering and that's when you can lose your pool - unless you use at least RADZ2.

Agreed :)