BUILD First FreeNAS Build, is the hardware I've chosen enough for my goals?

[Stingray88]

PCPartPicker part list / Price breakdown by merchant

CPU: Intel Core i3-4370 3.8GHz Dual-Core Processor ($142.99 @ SuperBiiz)
Motherboard: ASRock E3C226D2I Mini ITX LGA1150 Motherboard ($203.98 @ Newegg)
Memory: Crucial 16GB (2 x 8GB) DDR3-1600 Memory ($113.99 @ Newegg)
Storage: Western Digital Red 4TB 3.5" 5900RPM Internal Hard Drive ($154.00 @ B&H)
Storage: Western Digital Red 4TB 3.5" 5900RPM Internal Hard Drive ($154.00 @ B&H)
Storage: Western Digital Red 4TB 3.5" 5900RPM Internal Hard Drive ($154.00 @ B&H)
Storage: Western Digital Red 4TB 3.5" 5900RPM Internal Hard Drive ($154.00 @ B&H)
Storage: Western Digital Red 4TB 3.5" 5900RPM Internal Hard Drive ($154.00 @ B&H)
Storage: Western Digital Red 4TB 3.5" 5900RPM Internal Hard Drive ($154.00 @ B&H)
Case: Fractal Design Node 304 Mini ITX Tower Case ($85.99 @ SuperBiiz)
Power Supply: SeaSonic 520W 80+ Bronze Certified Fully-Modular ATX Power Supply ($64.99 @ SuperBiiz)
Total: $1535.94
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2015-08-19 03:40 EDT-0400

The HDDs I actually already own, I just added them so you know what I'll be using. Here is a list of my goals for this box:

* Plex/Kodi (unsure which I want to use) - I have ~7TB of media, would only need to sustain 2 streams max at any time
* Transmission - I seed 800+ torrents (~3.5TB), there is usually something seeding at any given time of the day (10-15GB/day)
* Time Machine - I've got 2 Macs that I'd like to backup hourly via Time Machine
* Mumble Server - Just for my friends, there would never be more than 5-6 people on it
* Owncloud - I currently use an FTP server and Dropbox for various purposes... and I feel like Owncloud might be able to do this all for me.

So what do you guys think? Does my hardware fit the bill for the task I'd like to use it for? I was considering a Xeon E3 instead of the i3-4370, but I thought that was more power than I really need (plus I can always upgrade later if I really need it).

A couple of other questions...

1. Should I get a CPU cooler other than stock? I don't plan on overclocking, but I know the case will be quite tight. If yes, any recommendations for a cooler thats very quiet and will fit in the Node 304?
2. Can I setup my drives as one big pool but still use them for media, torrents, Time Machine, and expect good enough read/write performance?

 

[tyronebiggums]

If you read around these forums, it's basically standard that you should use ECC ram so if the processor and motherboard don't support ECC, then you would have to go with different options. Super Micro motherboards also appear to be the defacto standard for homebuilds...

 

[Stingray88]

If you read around these forums, it's basically standard that you should use ECC ram so if the processor and motherboard don't support ECC, then you would have to go with different options. Super Micro motherboards also appear to be the defacto standard for homebuilds...

The RAM I picked is ECC RAM and the CPU and motherboard are ECC capable.

Also the only Supermicro mini ITX motherboard I could find without a soldered chip (MBD-X10SLV-Q-O) only has 4 SATA ports, and I want at least 6.

 

[Jailer]

So assuming you're going to go with a RAIDZ2 vdev for your 6 drives you're going to build your server and start it with a pool that is 75% full with plans to add more data via time machine backups? That's a really bad idea.

You would be much better off starting with 6TB drives if 6 is your limit. Otherwise add a HBA and add more drives.

 

[Nick2253]

Your hardware will probably meet your requirements, but your drive size may not.

First off, your hardware looks good. No red flags. The only yellow flag that jumps out to me is the PSU: that's many more watts than you need. A 300W PSU would likely be enough, and it would put you in a better spot in the efficiency curve (aka, lower power bill).

Secondly, you want to make sure you are using RAIDZ2 (see: http://www.zdnet.com/article/why-raid-5-stops-working-in-2009/). Based on that assumption, you'll see something like ~14TiB of free space. Now, this is where we get into problems. Like Jailer said, you've got ~10TiB of data, which puts you at ~70% full. You don't ever really want to get much above 85% full, or else you'll see big performance hits. That gives you around 2TiB left before you hit that threshold. Is that enough for you? You said you already have the drives: if that's not enough, I'd add a couple more HDDs, which may force you to rethink the Node 304/Mini ITX setup.

Third: your CPU will be more than enough for what you want to accomplish. Mumble, Transmission, and Owncloud really don't use a whole lot of CPU. Plex transcoding would be the only thing in your list that would be CPU intensive. If you don't do transcoding, the i3 is overkill. Even if you plan on transcoding one or two streams, unless they're raw bluray rips, you could probably get by with a Pentium.

On the whole, make sure you evaluate your space requirements, take a second look at PSUs, and you'll be golden. Good luck!

 

[Bidule0hm]

300 W seems too low to me, even 350 W seems too low. I'd put at least 450 W on this build, and more if you plan on adding more drives in the future. Also I recommend a 80+ Gold as the minimum because 80+ Bronze isn't that efficient, however platinum isn't worth it in general (+5 % efficiency from bronze to gold but only +2 % from gold to platinum). Take a look at the G-450 for example ;)

 

[Nick2253]

300 W seems too low to me, even 350 W seems too low. I'd put at least 450 W on this build, and more if you plan on adding more drives in the future. Also I recommend a 80+ Gold as the minimum because 80+ Bronze isn't that efficient, however platinum isn't worth it in general. Take a look at the G-450 for example ;)

I've used 300W on my build (which is quite similar) for the past 18 months or so. I guarantee, it's more than enough. My system idles under 50W, and load is just a smidge under 75W. Even peak bootup load doesn't take me over 200W.

 

[Bidule0hm]

Where the 200 W comes from?

@jgreco already has the argument about this with another member and at first I was on the low power side of things but after reading his posts I agree with him that you should always take a PSU that is oversized. I know that in the world of high reliability equipment and co the PSUs are never used at more than 50 % of their nominal power (including power peaks).

He also proved that you don't lose more than 2 or 3 $ of electricity per year by doing that (and that's a worst case).

I'll find the thread and edit to put the link as his posts were really interesting ;)

Edit: found: https://forums.freenas.org/index.php?threads/i-need-some-data-points.32829/#post-204099

 

[Nick2253]

@jgreco already has the argument about this with another member and at first I was on the low power side of things but after reading his posts I agree with him that you should always take a PSU that is oversized. I know that in the world of high reliability equipment and co the PSUs are never used at more than 50 % of their nominal power (including power peaks).

He also proved that you don't lose more than 2 or 3 $ per year of electricity by doing that.

Where the 200 W comes from?

I'll find the thread and edit to put the link as his posts were really interesting ;)

Edit: found: https://forums.freenas.org/index.php?threads/i-need-some-data-points.32829/#post-204099

200W was meant as an arbitrary line in the sand. My point was that I don't get anywhere close to 200W. The highest I've every seen my FreeNAS server pull is 114W at spin up.

50% is ridiculously conservative. I would even call 80% conservative, and that was the debate jgreco was having, not 50%. And even according to your 50% value, I've never got there. Which still makes a 300W PSU a conservatively large PSU for that load.

As far as cost difference, the real difference between a normal 80+ and a platinum at these load levels are less significant than proper sizing. Let's say we have a 50W power need, and power costs $0.12/kWH. With a 80+ PSU, we'd have a 62.5W draw from the wall, and our FreeNAS system would cost $66/yr to run 24/7. If that PSU were instead a platinum PSU operating at peak efficiency, you'd be drawing only 56W, which translates to a savings of only $7/yr.

Below 20%, efficiency drops rapidly. At our 50W number, we'd be around the 20% mark for a 300W PSU. For a 500W PSU, we'd be at the 17% mark. According to Anandtech's research on a Silver 900W PSU, at 10%, the PSU is less than 73% efficient, and at 17%, the PSU is about 82% efficient (http://www.anandtech.com/show/2624/3). Since efficiency curves under 20% are very difficult to get from manufacturers, if we allow this to apply to our situation we get:

500W PSU: 69W from the wall, $72/yr
300W PSU: 60W from the wall, $64/yr, a savings of $8/yr.

As you can see, the savings from properly sizing the PSU can outweigh the savings from a more efficient PSU, in a large part because the efficiency curve of the PSU drops so precipitously below 20%.

EDIT: One of the issues with jgreco's data is he is comparing the 360W PSU at 120V to the 550W PSU at 230V. The higher voltage will give at least a couple percentage points higher in efficiency. The chart for the 360W is also oddly flat. Another problem with comparing such data is the methodology may have been very different in collecting it, which makes any comparisons tenuous at best.

 

[Bidule0hm]

Ok, so where the 114 W come from? because one of your drive, per the datasheet, draw at least 24 W at spinup, so 6 drives will draw at least 144 W which is way over 114 W and there's the power of the rest of the system to add to that.

I didn't say we should use the 50 % mark but I said that only as an example that confirm that an (way) oversized PSU is better for reliability than a "just right" PSU.

As you can see, the savings from properly sizing the PSU can outweigh the savings from a more efficient PSU, in a large part because the efficiency curve of the PSU drops so precipitously below 20%.

Euh... yes, I didn't say the contrary.

EDIT: One of the issues with jgreco's data is he is comparing the 360W PSU at 120V to the 550W PSU at 230V. The higher voltage will give at least a couple percentage points higher in efficiency. The chart for the 360W is also oddly flat. Another problem with comparing such data is the methodology may have been very different in collecting it, which makes any comparisons tenuous at best.

Even if you include the 120/230 V efficiency difference I don't think you'll save much. And the comparison is more between a 350 and 450 or a 450 and 550 W PSU, he simply hadn't find the data for the 450 W PSU so he used the 550 W PSU data so it's a very worst case.

Oddly flat because SeaSonic PSUs are just good; for example one test I saw on the X-650 shown that the efficiency is sightly higher at a higher temp when generally is the other way around. This PSU also almost pass the platinum certification beside being only a gold PSU :)

 

[Stingray88]

So assuming you're going to go with a RAIDZ2 vdev for your 6 drives you're going to build your server and start it with a pool that is 75% full with plans to add more data via time machine backups? That's a really bad idea.

You would be much better off starting with 6TB drives if 6 is your limit. Otherwise add a HBA and add more drives.

Secondly, you want to make sure you are using RAIDZ2 (see: http://www.zdnet.com/article/why-raid-5-stops-working-in-2009/). Based on that assumption, you'll see something like ~14TiB of free space. Now, this is where we get into problems. Like Jailer said, you've got ~10TiB of data, which puts you at ~70% full. You don't ever really want to get much above 85% full, or else you'll see big performance hits. That gives you around 2TiB left before you hit that threshold. Is that enough for you? You said you already have the drives: if that's not enough, I'd add a couple more HDDs, which may force you to rethink the Node 304/Mini ITX setup.

I already own the 4TB drives, and I really do want to go with the Mini ITX setup (I live in a small, expensive apartment). I'd much rather delete some of my TV/Movie hoard than buy a whole set of 6TB drives to be honest, which is what I'll do if I start to approach 85%. A hefty portion of my media is from a time before I subscribed to Netflix, Hulu, Prime and HBO anyways... so there's a lot of content overlap with these services that I don't really need haha. Also as far as Time Machine is concerned, I plan to have it capped at 750GB as it's being used to backup a 256GB SSD and a 128GB SSD... 750GB will be more than enough for that.

Keep in mind, I was previously using a Drobo 5N in RAID5 with 5x4TB for a total of about 14.52 TB useable space... and I found that more than adequate for my needs. Also... please spare me for using one of those Drobo pieces of junk... it was a mistake I will never make again.

First off, your hardware looks good. No red flags. The only yellow flag that jumps out to me is the PSU: that's many more watts than you need. A 300W PSU would likely be enough, and it would put you in a better spot in the efficiency curve (aka, lower power bill).

Third: your CPU will be more than enough for what you want to accomplish. Mumble, Transmission, and Owncloud really don't use a whole lot of CPU. Plex transcoding would be the only thing in your list that would be CPU intensive. If you don't do transcoding, the i3 is overkill. Even if you plan on transcoding one or two streams, unless they're raw bluray rips, you could probably get by with a Pentium.

Thank you for this advice! I'll take this into consideration, I could probably stand to spend more on my PSU if I'm saving on the CPU.

I've got another question for you guys... One of my friends said that 16GB of RAM will not be enough for my drive set of 6x4TB (24TB). I was under the impression that the 1TB/1GB rule was based on your useable disk space, so in my case in a RAIDZ2 arrangement, ~14TB. Which of us is right?

 

[Nick2253]

Ok, so where the 114 W come from? because one of your drive, per the datasheet, draw at least 24 W at spinup, so 6 drives will draw at least 144 W which is way over 114 W and there's the power of the rest of the system to add to that.

That's 24W max, not 24W minimum. In practice, you see spinup load much closer to 10W, which is still twice what it draws during idle. Stagger spin-up, and you never see a huge load. And even if you spec for 24W/drive with all drives loading simultaneously, the rest of the system doesn't come close to the 160W you'd need to get to 300W.

As far as the rest of the system, it really doesn't use that much power. The CPU has a TDP of 53W, which again is an effective max draw, not a minimum. Even at boot, the CPU is not drawing that much power. DDR3 low voltage sticks are extremely energy efficient, and the mobo isn't going to draw that much power by itself. You've got no GPU or add-in cards to power. I don't get where you think this system is drawing a huge amount of power.

I didn't say we should use the 50 % mark but I said that only as an example that confirm that an (way) oversized PSU is better for reliability than a "just right" PSU.

And even at that mark, a 300W PSU is still an oversized PSU, not a "just right" PSU. A 150W PSU would be getting closer to a "just right" PSU, which is cutting it too close in my estimation.

Even if you include the 120/230 V efficiency difference I don't think you'll save much.

According to the data from Anandtech, 120/230V gives you between 2-5% depending on where you are in the efficiency curve. That's more than the difference between a silver and gold PSU at peak efficiency. I'd say that's something. Which is part of the reason that datacenters run off of 230V even in the US. Every watt counts.

And the comparison is more between a 350 and 450 or a 450 and 550 W PSU, he simply hadn't find the data for the 450 W PSU so he used the 550 W PSU data so it's a very worst case.

Even comparing a 300W and 450W would still yield around $5/yr of power savings. Whether or not that's important is up to you, but it's not nothing.

Oddly flat because SeaSonic PSUs are just good; for example one test I saw on the X-650 shown that the efficiency is sightly higher at a higher temp when generally is the other way around. This PSU also almost pass the platinum certification beside being only a gold PSU :)

It's also largely inconsistent with the other chart, which is also from a Seasonic PSU. I'm not doubting the quality of Seasonic's PSU; I use one in my main system for this very reason. But it's obvious that there's a methodological difference between the first data set and the second data set, so it's difficult to compare them without introducing some kind of bias.

If we simply look at the second chart, we see an efficiency difference between 10% and 20% that roughly matches what we see from Anandtech. In the first chart, we barely see any drop-off of efficiency below 50% load, which is highly suspect. My guess is an innocent artifact of the methodology, but that doesn't mean we can compare those data directly.

jgreco's methodology is sound, it's just based on mismatched data. With directly comparable data, you'd see power savings more in line with what I've presented here in a back-of-the-envelope fashion. Unfortunately, data like this is difficult to come by unless you collect it yourself, so comparing data between two points on the efficiency curve of one PSU will give you a more accurate comparison than comparing two different efficiency points on data collected in two different ways from two different PSUs.

 

[Bidule0hm]

That's 24W max, not 24W minimum. In practice, you see spinup load much closer to 10W, which is still twice what it draws during idle. Stagger spin-up, and you never see a huge load. And even if you spec for 24W/drive with all drives loading simultaneously, the rest of the system doesn't come close to the 160W you'd need to get to 300W.

As far as the rest of the system, it really doesn't use that much power. The CPU has a TDP of 53W, which again is an effective max draw, not a minimum. Even at boot, the CPU is not drawing that much power. DDR3 low voltage sticks are extremely energy efficient, and the mobo isn't going to draw that much power by itself. You've got no GPU or add-in cards to power. I don't get where you think this system is drawing a huge amount of power.

In fact it's the typical value, not the max, and you need to add the power of the electronics but it's only something like 1 W. There's the fans who also have a huge startup current (the spec says 2.2 A max during 1 second max at startup and 1.5 A max steady state conditions). Also when you design a system you always use the worst case scenario (at least if you design for reliability) because Murphy will get you every time otherwise.

Even comparing a 300W and 450W would still yield around $5/yr of power savings.

Well, the number are $12.3 for 5 years or $2.5 per year for a 350 and 550 W PSUs so I don't see how a 300 W and 450 W PSUs will be $5/year.

Unfortunately, data like this is difficult to come by unless you collect it yourself

Yep I agree, I'd be more than happy to do the tests myself but I've not the money to buy two PSUs just for a test.

so comparing data between two points on the efficiency curve of one PSU will give you a more accurate comparison than comparing two different efficiency points on data collected in two different ways from two different PSUs.

Excepted that it's why we want to do: to compare two different PSUs.

 

[Jailer]

I've got another question for you guys... One of my friends said that 16GB of RAM will not be enough for my drive set of 6x4TB (24TB). I was under the impression that the 1TB/1GB rule was based on your useable disk space, so in my case in a RAIDZ2 arrangement, ~14TB. Which of us is right?

16GB should be enough for your intended usage. Just be sure to purchase 2x8GB sticks. That way if it doesn't suit your needs or your needs change it leaves you room to add more in the future.

 

[Stingray88]

16GB should be enough for your intended usage. Just be sure to purchase 2x8GB sticks. That way if it doesn't suit your needs or your needs change it leaves you room to add more in the future.

Awesome. Thanks Jailer. The hardware has been ordered... look forward to putting it together soon!

 

[Nick2253]

In fact it's the typical value, not the max, and you need to add the power of the electronics but it's only something like 1 W. There's the fans who also have a huge startup current (the spec says 2.2 A max during 1 second max at startup and 1.5 A max steady state conditions). Also when you design a system you always use the worst case scenario (at least if you design for reliability) because Murphy will get you every time otherwise.

I'm genuinely curious where you get that this is the "typical value." The spec clearly says "max". I can imagine that 7200RPM drives draw something closer to that amount, and that's more or less consistent with the data I've seen. However, we're not talking about 7200RPM drives. According to Storage Review's review of the 3TB WD Green (http://www.storagereview.com/western_digital_caviar_green_3tb_review_wd30ezrsdtl), they found only 10.75W for startup, which is consistent with the data I've collected on my own system (however, my instruments are not that precise; I'm at +/- 1W or so).

Also, what fans are we talking about? There's tremendous diversity in fan power draw, so I'm curious what fan spec sheet you are referring to.

Well, the number are $12.3 for 5 years or $2.5 per year for a 350 and 550 W PSUs so I don't see how a 300 W and 450 W PSUs will be $5/year.

Excepted that it's why we want to do: to compare two different PSUs.

Exactly. But when the data is collected using different methodologies under different circumstances, you can't directly compare it in a scientifically and statistically sound way.

Just to put this to bed, let's go ahead and use real data. For this segment, I'll use data from the excellent Jonny Guru for his review of two 80 Bronze power supplies: the Silverstone 300W ST30FS and the XFX Core 450W. Same tester, same instruments, same methodology, same rating. The only thing we're looking at here that's different is wattage.

During his testing, he found the following (I used 61.5W because I didn't have to interpolate for the ST30FS):

• ST30FS - 61.5W DC @ 85.8% efficiency
• XFX - 61.5W DC @ 75.8% efficiency (I found this by interpolating between the 10% and 20% points in the table)

We run through our math again (and to compare against the jgreco article, I'm using his $0.14/kWh):

• ST300FS - 71.9W AC for $88.18/yr
• XFX - 81.1W AC for $99.46/yr

That's a savings of $11.28/yr, which is even more than I predicted.

Since we're having so much fun here :D, let's go ahead and redo the comparison at 60W between the Seasonic G-360 and the G550.

• G-360 - 60W DC @ 83.3% efficiency
• G550 - 60W DC @ 80.8% efficiency

Doing the math, again with $0.14/kWh:

• G-360 - 72W AC for $88.34/yr
• G550 - 74W AC for $90.75/yr

We see that jgreco's number are fairly accurate: we only save $2.41/yr between these two PSUs.

To be honest, this really did surprise me. We see that for both PSUs, their efficiency is about 3% lower than jgreco's number, which indicates (to me) that both these data sets were collected on 230V (Jonny Guru uses 120V), which more-or-less invalidates my earlier criticism. Also notice that Guru's numbers are not quite as flat as the chart indicates, but that may also be an artifact of the 230V. I'd be curious to know how that data in the first chart was collected to understand why it looks the way it does.

I decided to run through these numbers for a few other PSUs at different efficiency levels, which revealed that Seasonics stand alone for doing extremely well at low utilization efficiency. Nearly all other brands exhibit an extreme dropoff (as we saw with both the Silverstone and XFX PSUs, and even the Corsair PSU from Anandtech). From that perspective, the question isn't so much "What wattage?" but rather "What brand?". In particular, notice the efficiency difference between the XFX 450W and the G550W: 5% more efficient at an even lower utilization.

 

[diedrichg]

Also as far as Time Machine is concerned, I plan to have it capped at 750GB as it's being used to backup a 256GB SSD and a 128GB SSD... 750GB will be more than enough for that.

I don't know if this is the right method but this is what I use for my wife's two MacBooks' Time Machines. I did this under the assumption that each Time Machine would need its own storage space.

I setup one dataset per Time Machine. I took her (used capacity on each machine * 3 + another 20% (for good measure)) and then set that as the quota for each dataset. You can find some posts I made last year on all my settings for AFP, permissions and users for setting up Time Machine. The docs are pretty good at getting you though the setup but you may find my posts easier to follow.

Here is a ZFS calculator created by @Biduleohm to give you a fairly accurate idea of how much space you will have
https://jsfiddle.net/Biduleohm/hfqdpbLm/10/embedded/result/

 

[Bidule0hm]

I used the datasheet (not the "flyer" version, the complete datasheet that you can find here, page 16) and it's for the Seagate NAS drives as it's what you've put in your signature, not the WD greens drives.

Also, did you measure the power at the wall or on the DC side?

I used the spec for the PWM fans (which is a pain in the a** to find) because every fan is different of course, it's the max current but, again, designing for the worst case is better than for the average case.

I decided to run through these numbers for a few other PSUs at different efficiency levels, which revealed that Seasonics stand alone for doing extremely well at low utilization efficiency. Nearly all other brands exhibit an extreme dropoff (as we saw with both the Silverstone and XFX PSUs, and even the Corsair PSU from Anandtech). From that perspective, the question isn't so much "What wattage?" but rather "What brand?". In particular, notice the efficiency difference between the XFX 450W and the G550W: 5% more efficient at an even lower utilization.

That's why (and not only) I recommend SeaSonic every time ;)

 

[Nick2253]

I used the datasheet (not the "flyer" version, the complete datasheet that you can find here, page 16) and it's for the Seagate NAS drives as it's what you've put in your signature, not the WD greens drives.

The fact that they don't indicate watts or 5V current is a dead give away that the 2A specified there is not actually "typical" current. In spinup, there's obviously some kind of power draw, and the 5V rail also gets a huge current spike. I'd think that that's simply a by-product of them trying to fit everything in to one chart with one heading. The flyer is technically more accurate since it actually labels it properly as "max".

If you look at the spec sheet for the WD Green drives or the WD Red drives, they show a 1.75A max current at spinup, but in practice it's much lower, just like the Seagate NAS drives (or, for that matter, any modern ~5400-5900RPM drive). My guess is that, since Seagate didn't change the HDD controller from their Barracuda drives compared to the NAS drives, and that controller is capable of delivering 2A of spinup current, that's why the NAS drives are specced as 2A max. Though that's honestly just speculation.

Also, did you measure the power at the wall or on the DC side?

For what? The total system use is AC side.

I used the spec for the PWM fans (which is a pain in the a** to find) because every fan is different of course, it's the max current but, again, designing for the worst case is better than for the average case.

What PWM fans? A particular manufacturer? A particular model? A particular size? 2.2A just sounds excessive to me.

 

[Bidule0hm]

The flyer is technically more accurate since it actually labels it properly as "max".

I tend to trust more a datasheet than a flyer. You can't say that "max" is the proper label just because you want it to be that way. Anyway even if it's the max value you always want to use the worst case scenario when you design a system for reliability so in the end we don't care.

For what? The total system use is AC side.

You've said:

The highest I've every seen my FreeNAS server pull is 114W at spin up.

and I wanted to know where you've taken the measure. As I suspected you measured on the AC side but the PSU has pretty big capacitors inside and they smooth the power peaks you can have on the DC side (let alone the averaging of the meter you've used if it's a killawatt meter or similar, they're also not TRMS, only RMS) so the power peak is in fact much larger than what you read on the meter.

What PWM fans? A particular manufacturer? A particular model? A particular size? 2.2A just sounds excessive to me.

It's not a fan in particular, it's the specs you use when you design a fan (like you use the ATX specs when you design a PSU for example). I searched it (I downloaded it a while ago as it's hard to find) and found it here: http://www.formfactors.org/developer\specs\4_Wire_PWM_Spec.pdf (page 9)