October 5, 2012
For the last year or so, I've had two Drobos on my network: an 800i, and a 1200i. These aren't the versions you connect with FireWire, USB or even Thunderbolt. They're both iSCSI, and designed for use in a business situation. The 1200i in particular is designed to be rack-mounted all the way, and while you could keep an 800i in the same room you work in, unless you're in IT, I'm pretty sure you wouldn't want to.
QUICK primer: iSCSI is simply a way to use an Ethernet-based TCP/IP network as the connection between your storage and your storage. The server that talks to the iSCSI array uses an initiator to communicate with the RAID. In general, it's not as fast as things like Thunderbolt or Fibre Channel, but it's not slow, and it doesn't require a separate connection medium or infrastructure. If you have a solid Ethernet network, (Gigabit is REALLY recommended), you've got 90% of the work done for iSCSI.
The 800i is a bit of a hybrid, in that you use USB for the initial setup, mostly things like networking information, setting an administration password, single or dual drive protection, etc. Once that's done, you then decide what machines you want talking to, or managing the 800i via iSCSI and install the Drobo Dashboard software on them. This brings me to one of my first gripes about the Drobos, and the company's continued lack of understanding of the business computing market:
There's no way to just install the initiator on a server, and configure it to talk to the Drobo.
You have to install the entire Dashboard software on every computer you want to talk to the units, regardless of if you want that particular server to be a file server front end for the Drobo, or just manage it. It's a bit of overkill, and the reboot requirement for OS X, (not, amusingly, for Windows) makes this more annoying. Any time there's an update to the Drobo Dashboard or the initiator, you have to reboot, eventually, every Mac talking to the Drobos. For driver-like stuff, i.e. the initiator, I can almost understand that. But outside of that? Really? (Yes, you can use a third-party initiator. My experiences with those, at least on OS X, were less than stellar.)
Note for Linux admins: While the Drobo Dashboard doesn't have a Linux version, Drobo does have a procedure for using their iSCSI units with Linux, at http://www.drobo.com/downloads/how-to/HT-0077-00_iscsi_drobo_on_linux.pdf. The procedure uses Red Hat as their example, but you should be able to modify that procedure for other distros, such as Ubuntu.
The 1200i is a "pure" iSCSI system. It has three Gigabit Ethernet iSCSI ports, and one Gigabit management port. As with the Drobo 800i, you can only create HFS+/NTFS-formatted volumes via a Mac or Windows server connected to one of the iSCSI (data) ports. The initial price point on the 1200i is somewhat higher than the 800i, as you can only buy it with either six or twelve 2TB SAS drives, so the low end configuration is $11,999. But, with that you get a unit with dual, redundant power supplies, the ability to add a second iSCSI controller card for further redundancy, and a field-swappable fan unit. Drobo will sell you a spares kit that contains one each Fan unit, controller card, expander/secondary controller card, power supply and 2TB SAS drive for $3999. Clearly this is not a unit aimed at the home market. The 1200i is a solid SMB iSCSI storage unit, and for companies running ESX/Citrix farms, the 1200 is certified for use with those products.
You can get into an 800i for relatively cheaply, prices start, sans any kind of discounts, at $2999 for a drive-less unit. Drobo recommends you use server-grade drives, (and you should) but, if you have a good backup system in place, and you have dual-drive redundancy, you can be rather happy with consumer grade drives. You can put pretty much any SATA 3.5" drive in a Drobo. In my current setup, I"ve six 3TB drives and two 4TB drives. I'll slowly upgrade that to all 4TB over time. The reason this works is because Drobos support a wonderful concept called "thin provisioning". Rather than requiring you to buy all the drives you think you'll need at once, and making you deal with backup/restore/temp migration if you ever want to increase the size of the drives, Drobos allow you to start with a small number of drives, say 2-3 in the case of the 800i, and add as needed. If you want to replace a drive with a larger one, you just remove a drive, and replace it. When you do this, the Drobo automatically rebuilds itself to accommodate the extra space, without any downtime.
Note that the rebuild period can be rather long. When asking Drobo support about what seemed like an overly long rebuild time, I was told it can take 24-36 hours per terabyte. If you're rebuilding a 14TB data set...well, it took weeks. Lost no data, AWESOME, but yeah. Don't be impatient with a Drobo.
But now for the important question: How reliable/rugged is it?
I'm unimpressed by stuff that works under ideal circumstances. Everything should work in that case. Even ridiculously fragile things like Syquest and Jazz drives, (the youngin's can look those up while they get off'n mah lawn), could regularly handle not being used with aplomb. That doesn't tell me how reliable something is. It's when you get into the middle of a situation and you end up doing things that make sense at the time, but later, you realize were amazingly stupid, and you don't lose data...that's when I'm impressed.
For all of the faults i have with how you manage Drobos, the backing store Drobo uses on the 800i and 1200i is pretty darned bulletproof. For example, in one case, I had a bad drive that managed to masquerade as all kinds of other things for a while. So, I was rebooting the wrong way, i.e. power switch and power cable, randomly swapping drives, you name it until I could make the friggin' red lights go away. You could almost hear the 800i saying "help meeeeeee! helllllp meeeeeee!"
Data loss? Zero.
In another case, we managed to run out of space, and while I was dealing with that, had a drive go bad. At one point, we'd decided "the heck with it, we'll just restore from backups, there's no way this data's still here. While we're at it, we'll add a couple 4TB drives to it (the 800i) to help avoid the space issue next time." So I stick the 4TB drives in to replace the last two 2TB drives and the Dashboard gets very stern with me. "You have removed too many drives, and now I can't rebuild. Put the drives that were there back in and let me fix this mess."
So I put the drives back in, and about three weeks later, the volume was back. No data lost. For all my complaints about how Drobo does things, (and there are more than a few), in the one area they couldn't afford to fail at, namely protecting my data, they succeeded perfectly. I can forgive a lot of things for that.
That said, there's a lot of things about a Drobo that are just weird, or show me that the company, for having a really great product, still doesn't understand the business computing market. It's a good thing I can forgive a lot for bulletproof because Drobo makes you forgive a lot.
On the weird side...let's say you have a drobo full of drives and you've had a couple close calls, and you decide that you just want to blow out the volume on the set, add a couple drives, and create a new volume. This should be cake, right? Small number of steps:
- Delete volume in Dashboard. Couple clicks at most.
- Take out two smallest drives, replace with new bigger ones
- Create new volume
Well, step one worked great. Step 2 about drove me to a Taur Urgas level of insanity. I put in two 4TB drives, one gets the red light of doom. Okay, let me swap it, and see if the light follows, that would tell me i'd just bought a bum drive. Nope, light stays in the same place. Okay, let me put back in one of the 3TB drives that was in there...red light in a different place.
Repeat with some variations until it occurs to me that if the Drobo keeps a significant part of the data set information on the drives themselves, maybe I was confusing the thing too much. Okay, fine, just put one 4TB drive in. It'll complain, but the drive should be fine...red light. As a last resort, with that one 4TB drive in, I do a Drobo reset, which blows up the data set, and resets the entire Drobo back to the factory settings. Shut it down, hook up USB, (oh did I mention you have to reboot to switch between USB and iSCSI on the 800i? Or that you can't have both connected at once? Okay, there, I mentioned it. Two baby ducks nibbling you to death), and do the reset. Comes up green.
Put in the next drive, wait a few minutes...green. One at a time, re-add drives. No problems. Until I get to the last one, which evidently still had some data set information on it, because as soon as I put it in, YOU HAVE REMOVED TOO MANY DRIVES. Whaaaaa? Okay, fine, I have another drive still in the static bag. Put THAT one in. Life is good. Next time I know, if I recreate the volume, do a reset with all the drives in the unit. There is a desperate need for an option that lets me reset the disk pack without resetting the entire unit. This is also something that illustrates Drobo's lack of understanding of the business market. I've done a bit of looking around, and there's not much in the way of good technical documentation. There's a decent bit of basic stuff, but for the kinds of info someone wanting to use a Drobo in a VMware farm would expect? "Skimpy" is the descriptive here. For example, with my 1200i, I kept hitting a situation where the volumes would unmount at random after some time. A day, an hour, we were never sure. I and the Drobo support people were losing our minds. I kept sending them logs from the 1200i and the host computers, and nothing. Then, based on an offhand comment from one of the support folks, I assigned the management port to a different VLAN and voila, no more problems.
That's sort of mentioned in the documentation, but not at the level it should be. It should be in rather large letters: Don't have your management and iSCSI interfaces on the same network. Weird stuff could happen. But, when the documentation is skimpy, and aimed at the non-IT person, and you can't find a collection of 1200i knowledgebase articles by such search terms as "1200i" or "Drobo 1200i", well, it's all too easily to find yourself in strange situations.
As it turns out, you have to search for B1200i. Then you get results. Sigh. Another baby duck.
But management is the glaring weak point of the Drobos. Glaring
You can't use the logs on the silly things, because they're encrypted. That is part of what I consider the real sin of how Drobo does things. There is no way to really see what's going on with a device. You have indicator lights, some emails, an occasional popup, and that's it. The logs are encrypted. (IT people understand why I'm so...offended...by that.) The only people who can use Drobo logs are Drobo support folks. Marvelous. There can be all kinds of warnings about something about to go terminally wrong that I could trivially fix, but because I can't read logs, I have to wait for an email, or for a notice from the Dashboard application, assuming I have it running somewhere at all times. I don't, because to run it, I have to be logged in. I don't like to stay logged into servers...forever. Especially not to run a fat resource hog like the Drobo Dashboard. Even worse, at some point, someone figured out how to decrypt the logs. Drobo's answer? Better encryption.
Another baby duck. Or five.
Even the emails are just...not good, because they aren't sent by the unit, they're sent by the Dashboard app. That doesn't sound bad until you realize that if you have say, three drobos, and they all have different names, because well, you want to know which one you're talking about, and one gets low on space. All the email from that copy of the Dashboard is going to look like the same thing, regardless of the unit, because you're getting email from the Dashboard, not the unit. Is it talking about the 1200i or the 800i? You can't really know until you fire up the Dashboard and look. The emails themselves don't contain any details so they're the RAID version of "Check Engine".
Really, there's nothing about the Dashboard, when we're talking about the 800i and 1200i-class units that should be more than an iSCSI initiator/initiator configuration utility, along with a configuration utility to do the initial setup for the units. The rest of it should, and could, be handled by an embedded web server on the drive units themselves. They already only work via Ethernet anyway, what is gained by not moving the dashboard functions to the units? At that point, if you're using https, customers can more easily integrate monitoring into their own consoles.
Of course, Drobo could go one step further and implement SNMPv3 support for the 800i and the 1200i, which would allow any one of the gigantic plethora of SNMP-based network management consoles to manage the Drobos. Again, I'm not talking about the DAS models. The 800i and 1200i are iSCSI, business-class RAIDs. They are not going to be on your desktop. They are (hopefully) going to be in a proper server room where you won't be looking at them. Having the only way to monitor them be a Java App that you have to run in a login session, with encrypted logs so that when things go wrong, you have no way of telling what the unit itself is really seeing? That's completely unacceptable, and it is only, only the fact that the Drobo backing store has been, for me at least, so very bulletproof that I can even begin to recommend them for any use at all.
If it weren't for that, i'd tell everyone reading this to only use Drobos on the desktop, it's the only market the company seems to understand. Now, there are some signs of hope. I know some folks working for Drobo, and they get this issue, but it has to get fixed *now*, not some years from now. You cannot sell a rack-mounted iSCSI NAS device with only home or home office level monitoring and management features. If Drobo wants these products to be taken seriously by the kinds of people running Citrix and VMware farms, then this "Desktop management is all anyone ever needs" philosophy has to be taken out back and shot. For desktop products, sure. But when you're talking about a beast like the 1200i that handles automatic tiering between SSD and spinning disk?
No way. At that point, you have to do things the right way for the market you're targeting, and encrypted logs + Drobo Dashboard is not the right way for business computing.
I love how reliable the Drobo backing store is. I don't even mind the speed. They're not speed demons, but with Jumbo Frames and a direct Ethernet connection, I was able to get transfer speeds in the ~650Mbps range. That's not saturation by any means, but from an 800i with Best Buy hard drives? I can live with it, and my users have been living with it rather happily.
But when I think about managing the silly things, and what it would be like to manage four, five or more of them? If I'm going to do that, why not just get some Thunderbolt-based disk arrays? They don't offer remote management, but I don't expect them to in that case, and the host to disk transfer rates for Thunderbolt is really, really fast. If I'm going to have to waste a lot of time managing things, I should at least get speed. I want to like the Drobos more, but right now, it's only the fact that I've yet to lose data on them, in spite of trying really, really hard to, that makes me able to recommend them at all, and it's a recommendation full of 'buts'. If Drobo gets their management act together, the 800i and 1200i's will finally be the products they should be.
CommentsWarning for Notes users: The commenting system uses HTML.
I know this will be scary for some of you, especially Notes fans. However, open standards, rah-rah.
If you want to use less-than or greater-than signs, or other similar characters that HTML reserves,
you'll simply have to learn to do it the HTML way. Luckily, HTML is kind of popular, no matter what
your re-educators have told you, and you can easily find help on the intertubes.