Review of Data Robotics' Drobo 'storage robot'

Last week I gave a fairly harsh review of the Drobo "storage robot" product based on some initial investigation I did.  I concluded that while it's possible to do some level of hard drive mixing, it really wasn't worth it since you didn't get meaningful capacity out of it.  I based that on the fact that I loaded three 100 GB hard drives and one 1000 GB hard drive and found that the user only gained 100 GBs of capacity.  I spent some time in their offices today to get a more in-depth look at the product and spoke with their Director of Marketing Jim Schaff.

As it turns out, Data Robotics admitted this was true but that the configuration I looked at was an unrealistic and worst case scenario for them.  Schaff pointed out that someone would realistically have a storage growth and hard drive acquisition pattern of 250, 500, 750, and 1000 GBs as the cost of storage got cheaper over time so I will apologize to Data Robotics for only getting part of the story.  In this more realistic mixed drive size configuration, the Drobo would have an advantage over traditional RAID arrays because it can use multiple layers of RAID to maximize the amount of redundant storage the user gets whereas a traditional RAID isn't so flexible.  To illustrate this, the following is a screen capture of this hard drive configuration which shows that the Drobo will give the user 1.4 TB (1500 GB) of storage with the mixed drive size configuration and it will do it automatically behind the scenes for the user.

The best case a traditional RAID would only give you an effective redundant fault tolerant capacity of 1000 gigabytes and the 250 GB drive would be an extra standalone non-redundant drive.  The Drobo would not only give you a 50% advantage in redundant capacity, it also reconfigures the RAID on the fly to give you a seamless continuous redundant multi-level RAID volume.  To see how this works in action, I came up with the following illustration based on what I saw this afternoon.

In this configuration, we have two stripe + parity arrays (functionally equivalent to RAID Level 5) plus a mirrored array on top.  Note that this grey represents the amount of capacity used to facilitate fault tolerance but the parity information is actually distributed across all the drives just like RAID Level 5.  This produces a net capacity redundant capacity of 1500 GBs.  The 250 GB space left in the last drive isn't used until the smallest hard drive is replaced with a 1000 GB hard drive.  When a user runs low on space, they can remove the 250 GB hard drive and put in another 1000 GB hard drive when they get cheaper and end up with the following configuration.

With this new configuration, we still have two stripe + parity arrays with a single mirrored array which gives us 2250 GBs of redundant capacity.

So the bottom line is that Data Robotics really has created a much more flexible RAID technology that allow users to pay (with ever dropping prices) as they grow.  Other RAID solutions like the Intel embedded Matrix RAID or Software RAID can probably be updated to support this kind of functionality but Drobo is currently unique in this area.  Whether or not this would produce ideal performance characteristics or not is questionable but the ease of use is a definite plus for the Drobo.  See video demo here.

Another consumer friendly innovation is that the Drobo doesn't require the user to pick up a screw driver to put the hard drive in to a carriage bay.  That means you can just plug in a bare SATA drive right in to the unit as is.  I tried this myself and the Drobo felt mechanically solid that I didn't have any problems sliding in the hard drive.  There's a simple clip on the left side that locks in or releases the hard drive.  This requires more mechanical engineering on the Drobo but it makes life much easier for the user.

The downside to the Drobo is the moderately steep list price of $499 (reduced from original $699) with no hard drives included and the moderate performance levels.  Schaff told me that the performance was somewhere between 18 and 24 MB/sec [UPDATE 6/27/2007 - Actual performance seems to be in the 11 to 16 MB/sec range according to Tomshardware] with the write operations on the slower end of that spectrum.  In fairness, other USB devices and even expensive NAS (Network Attached Storage) units at higher costs run around the same moderate speed.  There are some more expensive Firewire 800 units on the market that will support higher speeds but they cost more money and they don't have all these ease of management features like dynamic mixed size RAID and bare-drive support.

On the other hand, do-it-yourself geeks like me can build a very high-performance internal hot-swap 5-drive RAID that's literally TEN TIMES FASTER for $130 but that requires a PC with an Intel ICH8R or ICH9R RAID controller ($120) and a dual core CPU ($120).  We can build a server that supports this configuration with gigabit throughput for about $600 including shipping and that lets you serve data on the network at 70 MB/sec or more out of a single Gigabit Ethernet adapter which can be shared by all the PCs on the network.  That does however require some skill or at least willingness to learn how to build a PC.  But this isn't the market that Data Robotics is targeting; they're after Photographers or other professionals that just want something to work out of the box without having to mess with all these settings let alone build something.  Those people lose money by the hour and they can't afford to waste time building or learning about this stuff.  Most of these users already have 2 to 4 external USB drives cluttering up their desks that aren't redundant and aren't seen as a single drive making things very difficult to manage.  This is the market that Data Robotics is targeting and I believe they've come up with a very appealing product for this market.

I asked Jim Schaff why Data Robotics didn't offer a NAS unit that could be shared by multiple PCs on the network.  Schaff explained to me that the NAS market is around 400,000 units while the DAS (Direct Attach Storage) market mostly via USB is around 11,000,000 units.  Of course this 11M market probability includes dirt cheap sub $50 USB 2.0 cases that hold single hard drives so I'm not sure if that's a completely fair comparison.  It would be interesting to see how things break down in the above $400 market.  Schaff also pointed out that the NAS market has high unit returns because customers give up if they can't figure out how to get the device working.  The other point was that the Drobo could be attached to an Apple Airport Extreme router or Linksys NAS head which has a USB 2.0 port that allows you put any USB device on the network.  That may reduce the performance further so I can't say I like this option.  Still, there's nothing that will rule out future versions of the Drobo that will support Firewire 800 DAS configuration or a Gigabit NAS version of the Drobo that performs better.

Product review summary

• Very user friendly with simple color LED indicators
• Very easy capacity management
• Multi-level RAID hidden under the hood
• Unique pay-as-you-grow seamless capacity expansion
• Bare drive support for tool-less drive installation and removal
• Very solid mechanical construction and build quality
• Looks good

Cons:

• Low performance level but comparable to other USB devices
• Moderately expensive (but still very competitive with alternative products)
• Not good for do-it-yourselfers and PC enthusiasts