Picture if you will, two second generation (FireWire/USB) Drobo hard drive enclosures from Data Robotics. Both are connected to a Belkin hub via USB, which is then plugged in to a Late 2009 model MacBook.
Drobo 2 contains four matching Seagate ST31500341AS hard drives (7200 RPM).
Drobo 1 is in the process of being upgraded to more storage, so it currently contains three Western Digital WD20EARS 2TB “green” drives (5900 RPM) and one Seagate ST31500341AS 1.5TB hard drive (7200 RPM).
My thinking was that one I started putting in the slower “green” drives, that Drobo would be the slower of the two, so I spent days moving dormant “backup” type data to Drobo 1, and kept Drobo 2 for stuff that might need more speed (like iMovie files).
Imagine my surprise when I did a disk speed benchmark using the excellent (and free) AJA System Test. According to this program, Drobo 1 (with the slower drives) has a write speed of around 24MB/s and a read speed of around 25MB/s. (It varies slightly between subsequent reruns of the test, but these numbers are good enough for this article.)
Drobo 2 (with the four matching faster drives) is probably going to be faster, right? Or at the very least, the same speed if Drobo itself is a bottleneck. So imagine my surprise when Drobo 1 cranked out about half speed – 13MB/s write and 12MB/s read!
What gives? I re-ran the test a number of times on the different virtual drives (each Drobo is configured to show five 1TB volumes, currently). I even went as far as plugging a single Drobo directly to the MacBook and testing, and doing the same with the other one. I kept getting about half speed on the second unit.
I know the Drobo supposedly slows down when it is full, so I did some searching. Drobo 1 was in the green (after upgrading three of the 1.5TB drives to 2TB), and Drobo 2 was showing yellow.
A quick search at the Drobo website shows that they claim a slowdown happens when it goes red, due to extra work being done by the unit to find room for data. Should this also affect read speed? And neither of my units were red.
As an experiement, I moved even more data off the yellow Drobo 2, until it was also green.
I re-ran the speed test and saw only a very minor increase in speed.
So a day later I decided to create this blog entry hoping someone else may find it in a Google search and give me some thoughts on what is going on here.
But as I run the test right now, I find that both Drobos are back to reading and writing around 25 MB/s! Huh? I still expect the one with the “slower” drives to be slower, but perhaps Drobo itself is a bottleneck and faster drives really don’t help much.
What happened to return the speed after a week of slowness? Perhaps Drobo is migrating data in the background and moving things around for performance?
Does anyone know how these things work?
Magic sometimes scares me. Especially when it is protecting all my data!