BUILD First NAS build

[MelaGo]

Hi all.
I've been planning to build a first solid NAS for a while now, right now I have a raspberry PI with an external HDD running some rsync :p.

I've read cyberjock's guide and hardware recommendations and have picked a pretty standard build.
Would you guys mind looking over it to make sure the parts are all good? Thx

Server will mainly be used for Plex (maximum of 2 transcodes) and Subsonic/Couchpotato/Sickbeard.

Case: Fractal Node 304
Mobo/CPU: ASRock C2750D4I
Every review I ran into said the C2750D4I is necessary if u require 2 Plex streams, as the C2550D4I won't be able to handle it. Do you guys agree?
RAM: Crucial CT2KIT102472BD1608B 16GB (2x8 GB)
PSU: Seasonic G-450
Is the 450W necessary? Or will 350 be enough?
Storage: 6x WD Red 3TB. Plan on running Raidz2, as recommended here.
Boot Device: Integral 8GB Fusion USB 3.0


Total comes up at about 1.6K around here.


Thanks!

 

[jgreco]

Is the 450W necessary? Or will 350 be enough?
Storage: 6x WD Red 3TB. Plan on running Raidz2, as recommended here.

Don't halfass a power supply. Undervolting your drives are a good way to cause them to fail, taking your data with them.

https://forums.freenas.org/index.php?threads/before-i-pull-the-trigger.27854/#post-184051

 

[MelaGo]

Don't halfass a power supply. Undervolting your drives are a good way to cause them to fail, taking your data with them.

https://forums.freenas.org/index.php?threads/before-i-pull-the-trigger.27854/#post-184051

450W it is. Thanks.

 

[marbus90]

360W is enough for any array of 6 HDDs. that's even more than overplenty for that.

6 HDDs, even the most evilish 7200rpm SATA Enterprise disks, are only using 2A at 12V during spinup. That's 12A or 144W total.

Rest o' the system is fine, just be sure to use the Intel blue/white SATA ports on the left hand side of the board.

 

[jgreco]

360W is enough for any array of 6 HDDs. that's even more than overplenty for that.

No, that's simply not a fact you can state like that. Show the numbers with proper deratings. . That's 144 watts ON TOP OF peak system current, and by the time you get off the merry go round, 360 is a lot closer to the edge than you make it sound.

6 HDDs, even the most evilish 7200rpm SATA Enterprise disks, are only using 2A at 12V during spinup. That's 12A or 144W total.

That's a total effin' crock. It takes power on TOP of spinup current to run the drive. Really. Sometimes they'll even list it, such as the 6W on the 5VDC for the Ultrastar 7K1000's shown as part of the spinup requirement. More often they don't list it, but that doesn't mean you can ignore it.

http://www.hgst.com/tech/techlib.ns...0067A757/$file/Ultrastar_A7K1000_final_DS.pdf

Let's try a saner, based-in-reality-and-safety estimate.

Base system (C2750D4I CPU/memory/mainboard) -45 watts, peak consumption.
Fans - 15-30 watts.
HBA - 10 watts.
Drives - 8 watts times 6 drives. This is idle current, NOT spinup current, and it is definitely additive ON TOP of the spinup. Which is ~144 watts, but could trend a bit higher.

That's 277 watts right there. However, you need to account for the fact that you shouldn't be peaking out a power supply, regardless of what the marketing glossies say. A 360 watt supply should never be asked to give more than 80% of its rated power, or around 290 watts. That fits, but it is close, and typically it is better to leave some room in there if there's no compelling reason to press down.

SO LET'S CHECK.

The likely idle current of that same system:

Base system - 15 watts
Fans - 10 watts
HBA - 10 watts
Drives - 5 watts times 6 drives

60 watts. Okay, so the cry and scream of the amateurs in the room will be "but ... but ... efficiency!" Yes efficiency drops on the larger supply. Let's be rational though.

A 60 watt system is operating at 16% of rated load on a 360W supply is probably around 80% efficient, while that same system is only around 13% rated load on a 450W supply, which is probably at least 75% efficient.

What that really means is that the 360W supply is consuming 75W while the 450W supply is possibly consuming as much as 80W. Let's assume, worst case, that it is actually consuming 5 watts more. At a cost of 14c/kWh, it might cost you up to $30.66 over the next five years for that larger supply. If it's actually that much more inefficient, which is questionable.

But on the other hand, you've got a supply that's definitely not going to be underpowered.

For the Avoton, it seems clear that the 360 could be a viable choice. It isn't stupidly small, but I've seen a LOT of people advising Xeon users with six drives to get the 360W, and that's just idiocy.

360W is enough for any array of 6 HDDs. that's even more than overplenty for that.

The hell you say. It sure isn't "more than overplenty". A 550 or 650 watt supply would be more than overplenty.

 

[Bidule0hm]

I agree with everything but this: "A 360 watt supply should never be asked to give more than 80% of its rated power, or around 290 watts."

I know it's a best practice to oversize a PSU (and usually by a factor even greater than 20 %...) for a steady load, but here it's a peak load and the PSU was designed to meet its specs all the way up to 100 % (for a not crappy PSU, of course...) so loading it to 100 % only for a few seconds will not do anything bad to either the PSU or the system. Especially as you don't reboot the server every hour and as the PSU is very very lightly loaded the rest of the time.

 

[jgreco]

I agree with everything but this: "A 360 watt supply should never be asked to give more than 80% of its rated power, or around 290 watts."

I know it's a best practice to oversize a PSU (and usually by a factor even greater than 20 %...) for a steady load

That's NAS suicide, that is. The spinup current will get you if you go that route, for any number of drives more than maybe two... you CANNOT look at just the steady or idle loads. The design has to be focused on minimum ("idle") and maximum loads, and multiple disk systems tend to get swamped out by spinup current.

but the PSU was designed to meet its specs all the way up to 100 % (for a not crappy PSU, of course...) so loading it to 100 % only for a few seconds will not do anything bad to either the PSU or the system. Especially as you don't reboot the server every hour and as the PSU is very very lightly loaded the rest of the time.

That's true for the power supply as it leaves the factory, yes. However, as a supply ages, the parts within begin to age, and sometimes an unexpected deterioration can adversely affect the performance potential of the supply as it ages. Even without that, it is not wise to rely on the design prowess of a PC parts manufacturer; these things are designed to be sold in a highly competitive, cutthroat marketplace.

One could argue that the fact that the supply only rarely has to provide peak power could actually lull you into a certain sort of complacency about the reliability of the supply, when in fact the ability of the supply to perform adequately might be compromised by an undetected failure. (This is actually a problem in any case and is the kind of design issue that is not solvable, but it can be mitigated to some extent by derating.)

In any case, the cautious electronic engineer will derate components in order to encourage a system to be more reliable, and derate modules or subsystems as well. The first three paragraphs on this page more or less summarize it pretty well. The specifics are complicated and difficult, but the end point is the same:

Don't intentionally design a system that could consume all the rated power. This leaves perilously little extra to deal with unexpecteds. A stalled fan. A seized drive. A shorted-out USB extension. We've seen ALL of those things, plus probably others I'm forgetting. Having a supply that doesn't break a sweat spinning up the system should be a design goal. I showed the math that should indicate a larger power supply is not an onerous opex.

If we accept that one shouldn't intentionally design a system that could consume all the rated power, then there has to be a target. For this size of system, I'd say 80%. This plays to both the ability to withstand certain kinds of adverse events while also avoiding complications arising from having a system that consumes too-little power at idle. That's actually a good reason not to put a 1000 watt (or even a 650 watt) supply in this box - you may discover that the supply designers didn't anticipate low draw scenarios and voltage stability is compromised. So, again, 80% is a well-selected number, at least for this scenario.

And if anyone gets the idea that I've stared at spec sheets and spreadsheets of this crap too long and might have been driven slightly crazy, I won't take offense. ;-) But honestly I just don't want to see people designing systems that are running on the bleeding edge. It'll end up undervolting your drives and might take a pile of them with it. Better a little too big than a little too small.

 

[Bidule0hm]

"you CANNOT look at just the steady or idle loads" I didn't say to calculate the PSU at steady load, I said that the oversizing is usually calculated at steady load. But of course here the steady load is very small and we need to use the start up current to calculate the PSU power.

Speaking of derating a well designed PSU is rated with a inlet air temp pretty high (usually 50 °C IIRC) so as we use it at less than that the derating is reversed for this parameter (especially as the life is usually quadratic to the temp).

"A stalled fan. A seized drive. A shorted-out USB extension." Yep, I totally forgot that, good point.

"Having a supply that doesn't break a sweat spinning up the system should be a design goal." I agree, but actually a well designed PSU shouldn't break a sweat even at 100 %... :)

I think we're both right, it's just the level of paranoia that varies :D and I'd follow the "Better a little too big than a little too small." rule. That's why I'd planned a 550 W PSU for my server (I've a 650 because the 550 was just unobtainium) which has only 8 drives (and also because maybe a day I'd expand it with more drives but I don't think I'll need to) :)

 

[marbus90]

[...]

Then you ought to go trough every system recommendation thread here and flame @Ericloewe: as well.

You can't compare a home system build in a Node 304 chassis plus the C2750D4I with a rackmount server with HBA, Expander, that lot.

C2750 - 4A tops
6x3TB WD Red - 1.5A @ 12V _only_, no 5V rated = 9A spinup current.
HBA - none.
Fans - not jet engines, I'll generously add 1A for that.

so we're at 14A in total during spinup. Let's add 20% for good measure and round that up, 17A.

The Seasonic 360W is rated at 30A @ 12V. And it's no china cracker, it's a proper Seasonic PSU. Even during spinup this system will not exceed 50% of the 360W PSU.

Let's dig trough worst case during spinup with 7200rpm Enterprise disks = 2A at 12V and 2A at 5V - still doesn't nearly exceed the Seasonic's ratings (which is 16A at 5V).

 

[Ericloewe]

Let's all take a step back and agree on a few points:

• 80% is a good target for maximum power
• A typical C2750 board isn't expected to draw more than 40W
• Ancillaries on a typical system with such a board won't exceed 20W
• WD's datasheet for Reds specifically states 1.75A +-10% maximum current draw. Say 2A for 24W, round that up to 30W for a conservative but simple number.
• A Seasonic G-360 can reasonably be expected to meet its specifications, which allow for 360W (assuming smaller rails are lightly loaded, which is the case in modern systems, 12V power draw is close to the total for all rails).
• In this case, 6x 30W + 60W equals 240W, or 66% of the PSU's rating
• If future expansion is desired, a beefier PSU is probably in order

 

[jgreco]

I didn't say to calculate the PSU at steady load, I said that the oversizing is usually calculated at steady load.

I would say that this is a contradiction. I'd add that the steady load is virtually meaningless for purposes of sizing a supply. It's helpful for calculating opex, though.

"Having a supply that doesn't break a sweat spinning up the system should be a design goal." I agree, but actually a well designed PSU shouldn't break a sweat even at 100 %... :)

In theory, I agree, but in practice, we're talking the world of PC's here, and hypercompetitive manufacturers who are all trying to make a buck in a rough, low-margin market. You CAN certainly design a power supply that doesn't break a sweat at 100% of its rated load, even after ten or more years, but the reality doesn't seem to match up.

Then you ought to go trough every system recommendation thread here and flame @Ericloewe: as well.

I don't have enough hours in the day.

You can't compare a home system build in a Node 304 chassis plus the C2750D4I with a rackmount server with HBA, Expander, that lot.

I didn't. However, a C2750D4I uses Marvell SATA controllers for the auxiliary SATA ports, and I believe only four are provided by a trusted SATA chip. This implies that you need a HBA for the remaining two drives, and/or "just risk it."

6x3TB WD Red - 1.5A @ 12V _only_, no 5V rated = 9A spinup current.

Then prove it. Cut the 5V power to the drive and show me it working. Otherwise, it needs 5V. The fact that a specific 5V current rating isn't listed during spinup doesn't mean that it isn't consuming some watts on the 5V.

 

[marbus90]

I didn't. However, a C2750D4I uses Marvell SATA controllers for the auxiliary SATA ports, and I believe only four are provided by a trusted SATA chip. This implies that you need a HBA for the remaining two drives, and/or "just risk it."

It has 6 times Intel SATA. Six working SATA ports in addition to Marvells. the Marvell energy usage is by the way already included in the 48W/4A allocation for the mobo.

Then prove it. Cut the 5V power to the drive and show me it working. Otherwise, it needs 5V. The fact that a specific 5V current rating isn't listed during spinup doesn't mean that it isn't consuming some watts on the 5V.

I don't deal with these cheap 5400rpm drives. I read datasheets, they say 12V. And:

Let's dig trough worst case during spinup with 7200rpm Enterprise disks = 2A at 12V and 2A at 5V - still doesn't nearly exceed the Seasonic's ratings (which is 16A at 5V).

The chassis doesn't allow for more 3.5" HDDs. The board doesn't allow for more HDDs (the Marvells are not stable). No more HDDs possible in that system. And I said SIX HDDs all along, not 12 or 24 HDDs.

 

[jgreco]

It has 6 times Intel SATA. Six working SATA ports in addition to Marvells. the Marvell energy usage is by the way already included in the 48W/4A allocation for the mobo.

Fine, then the HBA can be eliminated. My recollection was that four were Intel. It's been a while since I played with the C2750D4I. We've got one in the 1U AsRock 12 drive storage server variant here and it needs a HBA to make the last eight drives work right, but that might be a cabling issue since there's SFF8087 involved.

I don't deal with these cheap 5400rpm drives. I read datasheets, they say 12V.

Yes, but the drives almost certainly also require 5V. The fact that they specify a particular amperage for the 12V is so that those of us who build stuff professionally can properly size. Typically, that number is in ADDITION to the other current draw required by a drive. As I said, feel free to demonstrate I'm wrong by cutting the 5V lead to the drive. The 12V is rated in amps because that's such an important design consideration. The watts listed for idle/seeking/writing are likely combined 5V plus 12V, since the spindle is spun on the 12V but the electronics are running off the 5V.

The chassis doesn't allow for more 3.5" HDDs. The board doesn't allow for more HDDs (the Marvells are not stable). No more HDDs possible in that system. And I said SIX HDDs all along, not 12 or 24 HDDs.

I never said anything about more than six drives. What are you banging on about?

 

[Bidule0hm]

Yep, usually the 5 V is for the logic and the 12 V for the motor, so the 5 V amperage should be very low but jgreco is right that it's in addition to the 12 V amperage ;)

 

[depasseg]

Gentlemen - now that the always difficult power supply conversation is out of the way, what about the MB, RAM, and CPU?

 

[Bidule0hm]

MB --> Server grade
RAM --> ECC
CPU --> Xeon

Problem solved :D

 

[jgreco]

Yep, usually the 5 V is for the logic and the 12 V for the motor, so the 5 V amperage should be very low but jgreco is right that it's in addition to the 12 V amperage ;)

Yeah, it's hard to know. Fortunately, for 95% of the users on this forum, it isn't a significant factor so you can get by with adding the idle watts. It's more of a factor when you're dealing with a system with 48 or more drives.

 

[mjws00]

It would be interesting to see where a seasonic G360 starts to undervolt a drive in a manner that matters. Everything I can find shows less than 2% variance at full loads. I'd be willing to bet it can do 40A in its sleep for a spin-up time period, without significant heating or measurable wear. But I don't have one, and never will. It's too damn small for me to repurpose if I choose, not modular, and the efficiency gain at low load is moot.

The watts being thrown around "conservatively" are nearly double what is often measured. X10 with 6 reds pulls 114w peak measured here, here is a 3 drive system at 91w startup. There must be dozens of these and no moderate 6 drive systems are starting at anywhere near the 80% "safe" load of a 360. I'll take measurement over datasheet speculation and rule of thumb all day every day.

Of course jgreco's logic on "safer is better than sorry" is absolute truth, imho. Especially with PSU's. But in ancient times we often over-specced to make up for QC. Buying at the higher end negates this to an extent. There are plenty of crap 550's that are WAY worse than a G360. Of course throw a server chassis in the mix that might have 20A worth of fans howling... different ball game.

Hilarious thread, with some good data.

 

[depasseg]

Here's the monitoring stats from my 12-bay server:
Xeon E5-2637V3 | 128GB DDR4 ECC | X10SRH-CLN4F | On-board LSI 3008 | 12 x 4TB WD-Re SAS | 2x Intel S3700

 

[Bidule0hm]

Be careful with measurements because if you measure on the mains side then the value is probably underestimated because the caps smooth the peak.