This is a long overdue post.
Back in October I took advantage of DataRobotic’s evaluation program, obtaining a DroboElite for a 30-day period for testing.
I finished up my testing and data gathering in November. My plans included comparing the iSCSI performance of the iSCSI DroboElite with server-based iSCSI target solutions such as Nexenta.
Life, work (my day job) and the sometimes unavoidable complexities of building a home server pushed my final writeup out. And out… more family and more work.
Here it is, mid February, and Data Robotics has made new announcements of business-class offerings, and I started to feel pretty bad about blowing off my final words on the DroboElite. I know a few people over at Drobo, and they went out of their way to make sure that I could get access to the DroboElite, and I didn’t live up to my end of the deal in the time frame that I suggested.
I took extensive performance data, driven by the benchmark “fio” written by Jens Axboe. It’s a very flexible benchmarking tool, I suggest you check it out. I will not present the data here, however, as there were serious outliers in every dataset that I took, that I later attributed to the compiled version of “fio” I was running on my test platform (Mac mini).
I tested with 5 new Samsung Spinpoint F3 HD103SJ 1TB drives, trying both single and dual parity modes.
Single host performance was excellent, with both random and sequential read and write performance being very good, both for larger block sizes (bandwidth) and smaller block (IOPs). I was able to get a very high percentage of the maximum expected bandwidth from the larger block testing, anywhere from 60-90MB/s. Mixed read-write performance was significantly less optimal, specifically 4k and 8k mixed random read-write.
Multiple host performance mirrored the results of the single-host performance. With 2 hosts doing large-block sequential IO, I was able to get nearly 100MB/sec combined between the two hosts. Mixed read-write performance was less than optimal, similar to the single host testing.
The above performance patterns, while not sufficient for “enterprise” levels of IO from several hosts, is certainly viable for SMB and professional offices. More importantly, like it’s smaller SOHO brethren, the DroboElite is easy to set up, and a Drobo customer isn’t going to need a consultant to get it going, and to maintain it.
Conversion to double parity, in my case going from 4+1 RAID 5 up to 4+2 double parity RAID, was painless. A couple of clicks in the GUI, and off you go. Conversion back to single parity was just as easy. Compare this to ZFS, where you cannot convert a RAID-Z directly, nor can you expand it. You can replace drives (increasing in size along the way) or add other RAID-Z groups to a zpool, but you cannot made the RAID-Z group “wider” or add additional parity.
Speaking of ZFS, I struggled for weeks (several 10′s of hours all in all) coming up with a ZFS solution that I was happy with. If I were a business professional, those hours would have been lost to the business. I’m still not completely satisfied with my ZFS file server. If I owned this Drobo, I would have spent my time playing with other things, rather than rebuilding file servers over and over in order to come up with a satisfactory ZFS solution for my needs.
Where does the DroboElite fit today given the new announcements? Looking at the specs, it seems that the DroboElite is either morphing into, or being replaced by the B800i. I have my suspicions as to what the actual details are, but will withhold speculation until I learn more.
Disclosure: I’ve worked with a couple of people who are now at Data Robotics. They made it possible for me to do this evaluation. But they haven’t asked me to color my responses in any way.