Drobo B1200i: The heavy-duty array even your mum could use

Re: How will it perform over time in the noise department ?

@Peter R.1 it is noisy. "Banished to the server room" noisy. I had thoguht that was one of the things I put in the article...am I going nuts, or did I cut that to get the word count down? Off to reread...

Re: Windows file copy screen grab

That particular screenie was - I think - pulled from a system doing a "bulk transfer" from a LUN on my Hyper-X array (which goes zoom!) to a LUN on the Drobo. If I read the notes correctly that I took down for that screenie, it was taken as part of a larger test in which - at the time - I was moving 25 VMs from one LUN to the other, and wanted to see how this would affect the stability of the file transfer.

Basically, it tanked the throughput - from 100MB/sec to 35MB/sec – and the throughput became "unstable," fluctuating between 30MB/sec and 50MB/sec. Windows file copy couldn't figure out what to do about judging transfer time on that and flopped around like a fish out of water.

Ultimately, however, it averaged 32Mb/sec over the course of the transfer – measurement taken with a stopwatch and some basic maths – despite the serious hammering the VM nodes were putting the unit through. (I think that 3 of the VMs in the transfer group were undergoing Windows Updates - .net and MSRT – while another one was performing an integrity check on a financials database. I turned Windows' caching off for the test.

Re: Trevor.. these are nice boxes, but..

To be honest; it worries me...but less so than it probably should, were I a "proper" storage wonk. The truth is that I don't trust "dual controller" systems as the be-all and end-all of high availability either. Dual PSU, dual controller…but that still leaves a mobo, CPU, RAM and other widgets inside those boxes that are single points of failure.

The lack of real-time block-level replication is the thing that ultimately makes me twitchy.

Even if I were to go out and buy an expensive storage array from EMC, Netapp or otherwise, I wouldn't sleep at night unless everything was backed up. Ultimately, that means block-level replication between two storage devices bonded in a cluster using MPIO against at least two switches into two separate controllers on each relevant host.

And then I will still take backups with something like a Unitrends box. (HA is not backup!)

So what does the lack of a second controller in this circumstance really mean? If it fails you are down. That's a bummer. It is swappable; you can always keep a spare on hand. You could also keep a spare Drobo on hand, and just swap the disks, but that's getting into "byzantine" territory.

So my take on this is "do no use the Drobo for anything that requires HA storage." Also: "make sure you buy a second controller card and keep it on the shelf." For the majority of my workloads, I don't need HA storage. A half-hour's downtime – even at the height of the year – isn't the end of the world. I also keep backups of everything, so even if the whole thing were to vanish in a puff of smoke – a possibility for any storage array, no matter the vendor – I can recover.

I don't have a problem rolling this out to client sites to prove "scratch space" LUNs for creatives. I also have one going into service as the "volume shadow copy" LUN provider for a pair of physical servers on a client site. (That one has no SSDs; it's just a giant box of storage.)

Like any storage, it is a question of fitness for purpose. The lack of a second controller wouldn't bother me if Drobo were to add block-level replication; I'd rather have two completely separate units in lock-step than I would a single unit with various "redundant" bits.

The dual controller part would take Drobo doing some reengineering to solve. Block-level replication could be done with a software update. Until Drobo make that choice, I feel they are locking themselves out of a wider market that would otherwise give them serious consideration.