Possibly ideal mini-ITX NAS platform

Status
Not open for further replies.

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
http://www.u-nas.com/xcart/product.php?productid=17617&cat=249

For non-rackmount use, this is what I've been wondering "why hasn't someone designed this."

I'm thinking that this, plus the new AsRock C2750D4I, would make a spectacular box for backup purposes (where max speed isn't essential). I haven't seen these bits or integrated them together, however. But it looks like it might be hard to go wrong. Dual Intel ethernets, eight drive bays, 32GB ECC RAM, and a CPU that isn't total suckage.
 

cyberjock

Inactive Account
Joined
Mar 25, 2012
Messages
19,526
Yeah, the Marvell SATA chip could be a problem though...not sure if they are compatible with FreeBSD.

I'm waiting to see how these do, and what their price is. If the price difference is less than $50 between it and a system with a g2020 CPU I'd rather go with the g2020 system. At least you can drop in a more powerful CPU if you suddenly need more power. The thread I just read was someone that appears to be disappointed with their low powered AMD llano CPU. I prefer to try to plan ahead than be disappointed later.
 

JimPhreak

Contributor
Joined
Sep 28, 2013
Messages
132
I'll let you know in a week or 2 as I have one on it's way and will be pairing it with that ASRock board an an i3 4130.

EDIT: Nvmd that's not the board I'm using. Mistook it for this board.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
Yeah, I'm not sure if I'd really want to put a ton of heat into a toaster sized box. On the positive side, the fans look like a winner. But back on the negative side, it looks like the airflow is basically straight through the drive bays and then out the fans. I see no obvious air path for the mainboard. It is essentially in a five sided box with lots of airflow moving two or three inches above, which implies relatively dead air zone, which implies slow cooker.

What did you pick for a CPU cooler, and what CPU/TDP? It looks to me like the unit basically limits the height of the system board and components to be maybe a shade higher than the top of the I/O shield. That implies that you'd probably need to use a low profile 1U passive cooler, or maybe a 2U passive that's been modified to fit, and then maybe figure a way to jimmy a small circulation fan in there too. I reeeeallllly want a 3D printer that works because it'd be so cool to be able to build custom ductwork for applications like this.
 

vegaman

Explorer
Joined
Sep 25, 2013
Messages
58
http://www.u-nas.com/xcart/product.php?productid=17617&cat=249

For non-rackmount use, this is what I've been wondering "why hasn't someone designed this."

I'm thinking that this, plus the new AsRock C2750D4I, would make a spectacular box for backup purposes (where max speed isn't essential). I haven't seen these bits or integrated them together, however. But it looks like it might be hard to go wrong. Dual Intel ethernets, eight drive bays, 32GB ECC RAM, and a CPU that isn't total suckage.
Exactly what I've been thinking.
And with an HBA added in the expansion slot you could even hook up a second (or third with an expander) enclosure if you need more disks.

As far as heatsinks, I read around and it seems like you get about 40mm. So that Noctua should fit but it might be tight with the fan added.

Sent from my GT-I9300 using Tapatalk 4
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
This is the HSF I'll be using with my Haswell i3 4130 (54W) CPU.

Hope that works out for you, but basically the 1U active fans with almost no heatsink are completely dependent on the fan to avoid catastrophe. With 4 wire PWM, maybe not as bad a risk if you can monitor for fan failure (IPMI), but I think I'd still be skeptical beyond maybe 40W TDP based on past failures. They run hot, especially in the center, which conveniently cooks the fan mechanism...
 

JimPhreak

Contributor
Joined
Sep 28, 2013
Messages
132
Hope that works out for you, but basically the 1U active fans with almost no heatsink are completely dependent on the fan to avoid catastrophe. With 4 wire PWM, maybe not as bad a risk if you can monitor for fan failure (IPMI), but I think I'd still be skeptical beyond maybe 40W TDP based on past failures. They run hot, especially in the center, which conveniently cooks the fan mechanism...

I don't expect my CPU to be running anywhere near full throttle. It's basically just going to be serving up media files and receiving backups from PC's and VM servers.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
Still, expect high chance of failure before 24 months.
 

JimPhreak

Contributor
Joined
Sep 28, 2013
Messages
132
Still, expect high chance of failure before 24 months.

I appreciate your concerns. However it is what it is at this point. I'm not going to turn back now so I guess I'll just have to run the risk. I'll probably try to undervolt the CPU as much as I can.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
Just trying to encourage being proactive and thorough thru monitoring. We saw lots of 1U active fan failures but passive merely needs to be blown out now and then. Needing to make a cross-country trip to troubleshoot failures really sucks... lesson learned. Heh.
 

JimPhreak

Contributor
Joined
Sep 28, 2013
Messages
132
Just trying to encourage being proactive and thorough thru monitoring. We saw lots of 1U active fan failures but passive merely needs to be blown out now and then. Needing to make a cross-country trip to troubleshoot failures really sucks... lesson learned. Heh.

I appreciate the advice, I really do. What kind of passive cooler would you have gone with in my situation?
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
Basically along the lines of what I outlined in #4.

Ok so look at the unit you ordered. Structurally this thing has a bunch of problems. In a typical 1U chassis the first issue you have is that the fan is parallel to the chassis cover, and only a few mm away. I *think* it isn't that bad for the U-NAS case, but basically it isn't designed to be moving a lot of air, which you can see because the second major issue is that there's very little space for the air to exit the assembly. The heatsink is 23mm tall, but along the edge about half of that is obstructed. A lot of potential for dirt buildup and even when totally clean, not a lot of airflow. The third thing is that it is made from aluminum finning. Aluminum is sure a HELL of a lot easier and cheaper to manufacture than copper, especially that sort of design where the unit is built out of stamped sheet. But aluminum has about half the heat transfer capacity, so that nifty heat pipe is their way to try to compromise. On the flip side, aluminum tends to work better at lower airflow speeds. Which this sort of fan and heatsink arrangement gives you.

Compare that to something like the Akasa AK-CC7111 or Dynatron K129. Might well work as-is passively, but I'd be real tempted to throw in some active airflow. There's that nice cavity on the side by the power switch where the system board winds up.. So this requires a little three dimensional imagination. Imagine a piece of plastic perpendicular to the mainboard, placed between the RAM and the heatsink. If cut to the precise shape of the cavity, you gain a new "floor" that is about 2 centimeters higher than the metal floor. The RAM and power connector underneath, the CPU above. This is of course a bit of fiddly plasticwork. But this is your intake duct, which brings air from the rear of the chassis (where the big fans move air) to the front. Then you punch a 40mm hole through the plastic to accommodate a 40mm fan. Pick a good 4-wire PWM fan like the R124028BU, and look, it wires right into the mainboard right there too, how convenient. So the fan pulls air through the intake duct and blows it into the upper cavity. Now, depending on how well you feel that works, you could also duct it back over towards the heatsink. You end up with both cooled memory and CPU, and a fan that is less likely to die, and even if it does, the relatively massive heatsink and all its unimpeded-by-a-fan-on-top surface space is likely to be very resilient.

Of course, my way is a bunch more work and trouble. I seem to specialize in that. ;-)
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
Oh, and, don't get all freaked out or uptight about it. I'm guessing the active 1U you ordered will work, well enough, at least for a while. In the old days CPU's used to actually burn up (scorch marks) if a cooler failed. These days, it can still do damage but if you're smart and set up IPMI, you can monitor both for CPU temperature alarms and fan failure alarms. I just like to over-engineer things.
 

cyberjock

Inactive Account
Joined
Mar 25, 2012
Messages
19,526
I'm kind of expecting these to only be modest improvements over the prior Intel Atoms.

Can I take a guess of CIFS hitting 50-60MB/sec max(since its single threaded), rsync will suck on them(again, since its single threaded) but NFS works fine since its multithreaded and only if you don't have more than 3 or so users at the same time?

I'd say for most people 50-60MB/sec would be acceptable, so for many its a valid solution. For the hardcore maniacs like myself that want triple digit speeds over CIFS I'm expecting to be disappointed.
 

JimPhreak

Contributor
Joined
Sep 28, 2013
Messages
132
Check the benchmarks that just went up on STH http://www.servethehome.com/Server-...8-core-avoton-rangeley-benchmarks-fast-power/

Quite a large improvement over the previous Atom chips, and the single threaded performance isn't as bad as most expected. I wouldn't expect big triple digit speeds either, but hopefully it would saturate gigabit - which is enough for a backup NAS or most home users I think.

My current RAID10 array on a Dell Perc 6i gets 110-120MB/s read/write speeds over my gigabit network at home. I'd be interested to see if the Avoton with ZFS can do that. I'd prefer not to take a big performance hit switching from my RAID10 setup.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
Yeah, think cyberjock and I would both guess that CIFS isn't likely to be able to do that. Borrowing Patrick Kennedy's benchmark results (ServeTheHome, an awesome resource that isn't just for home!)

Intel-Atom-C2750-UnixBench-Single-Threaded-benchmark-600x304.jpg


This is the singlethreaded result. You can see that the C2750 comes in at just a bit more than half the E3-1230v3. Basically, if I wanted a performance NAS, I'd still be looking at the E3's, but the C2750 ought to be pretty decent for many other purposes.
 

JimPhreak

Contributor
Joined
Sep 28, 2013
Messages
132
Yeah, think cyberjock and I would both guess that CIFS isn't likely to be able to do that. Borrowing Patrick Kennedy's benchmark results (ServeTheHome, an awesome resource that isn't just for home!)

Intel-Atom-C2750-UnixBench-Single-Threaded-benchmark-600x304.jpg


This is the singlethreaded result. You can see that the C2750 comes in at just a bit more than half the E3-1230v3. Basically, if I wanted a performance NAS, I'd still be looking at the E3's, but the C2750 ought to be pretty decent for many other purposes.

So in other words there's no real effective and safe way to get good performance out of my U-NAS 8-bay case? Right now my RAID10 file server has a Q6600 in it. I figured that was old enough that the newer low power chips might be comparable but I guess not.
 
Status
Not open for further replies.
Top