WD Sentinel DX4000 — unacceptable power design

Update: It appears Western Digital has indirectly responded to my blog post, stating that the concerns over disk power draw on system power-up are incorrect. They state that the disks are spun up individually, with 2 second intervals between each disk. This is good to know. I would not have written my “scathing review” had Western Digital taken the time to document this behaviour somewhere, especially in their own product specification sheet. This is why NAS manufacturers need to be fully open with disclosure of how their products work, including low-level engineering details. This stuff plays a bigger role today than ever.

Western Digital has released a new SoHo NAS product, the WD Sentinel DX4000, which is driven by an Intel Atom mainboard and CPU, supports dual gigE (for failover, not trunking), and up to 4 disks in RAID-5. It’s 8x6x8 inches in size for the 4-disk version. The cost? US$1500 retail.

Remember: the product is intended for businesses or high-class end-users.

Now take a look at its AC power connectors: two circular female connectors with a male centre pin. You know, the common “wall wart” and “in-line brick” crap you’d see on consumer-grade products that results in cable clutter (and absolutely impossible to properly deal with when placed in a rack)? No, surely that couldn’t be, especially for a SoHo device of those physical dimensions…

I thought for a moment it might be a proprietary connector that required a proprietary cable that went directly to a 115V AC plug, so I downloaded the official User Manual.

Lo and behold, on page 7 of the manual (page 11 per PDF index), an in-line brick (transformer) is shown. You get only one (the 2nd connector is for failover/redundancy), and its power requirements are miniscule. Quoting page 9 of the manual (page 13 per PDF index):

Input voltage: 100V to 240V AC
Input frequency: 50 to 60 Hz
Output voltage: 19V DC, up to 6.32A

I’m a little surprised by those power requirements: 6.3A at 19V. The unit comes with 5400rpm drives — if you were to shove four 7200rpm drives in there, you almost certainly would be taxing that little brick of an AC adapter. Your drives would start falling off the bus and experiencing strange anomalies during high I/O. Let me explain:

The WD Digital Caviar Black WD2002FAEX draws 10.7 watts when reading/writing according to WD’s own specifications.

Let’s figure out the percentage of power use a single WD2002FAEX drive would require: (10.7W / (6.3A * 19V)) * 100 = 8.94%. 8.94% * 4 disks = 35.76%. Not so shabby — but remember, we also have no idea how much power the Intel Atom motherboard, CPU, RAM, and other things require. Let’s just assume 35.76% is acceptable.

Now consider drive spin-up, which draws significantly more power than during normal operation: almost 4x more. StorageReview’s explanation (which I’ve linked) does not take into consideration technology like SSU or PUIS, which we do know the WD2002FAEX supports. But let’s assume worst-case scenario, that the WD Sentinel DX4000 spins up all the drives simultaneously.

8.94% * 4 disks * 4x more power = 143.04%. Oh dear…

Considering that I cannot find a single white paper or technical document describing in detail the gains from using SSU/PUIS (only actual SATA-IO specifications for how to implement it as part of the ATA protocol), I cannot determine what the power savings are. However, to be fair, I do imagine that in a worst-case scenario the WD Sentinel DX4000 might spin up each drive with 5 seconds of delay between drives. That means the entire unit would have to wait 20 full seconds before becoming even remotely operational. That probably isn’t what happens in real life. :-)

I cannot believe that for a US$1500 product, Western Digital would choose this kind of configuration. Why is it so hard for companies to use internal transformers? Nobody in a SoHo or business-class environment wants in-line transformers or “wall warts”. Using an internal transformer/PSU would extend the depth of the device by 1-2 inches at most. RFI is not a concern given good shielding, and airflow wouldn’t be a problem assuming the PSU was placed along the outside edge and provided with ventilation slits (I’d be more worried about the drives overheating than the PSU!).

Consider that those in-line transformers are usually of sub-par quality, and are usually the first thing to die/go out on you.

What is it going to take for companies to this kind of cheap engineering? Even the Drobo FS uses the same configuration, although the DroboPro and DroboElite definitely got it right (not to mention costs US$250 less than the WD Sentintel DX4000).