Object storage bods Exablox: RAID is dead, baby. RAID is dead Object storage supplier Exablox has provided a file interface to its OneBlox object store, which is basically a Drobo-style BYOD store - Bring Your Own Disks. El Storage asked Sean Derrington, senior director of products at Exablox, some questions about the company. Some of the answers have been edited for brevity. El Storage …
RAID is dead. RAID will not solve customers’ requirements going forward. Object storage is the next generation technology that fundamentally enables organisations to easily grow and manage their storage infrastructure.
Am I missing some vital piece of common knowledge, or is there a complete logical disconnect between the second and third sentences?
These guys with the marketing babbling nonsense "object storage" are just beyond silly...
So they aren't using any RAID algorithms to create and manage redundancy on the drives arrays ? Really?
I doubt that!
NAS manufacturers have been using extended proprietary RAID modes that allow to mix drives of different sizes avoiding the limits of classic RAID levels for years already...
The "object storage" revolution it's just marketing babbling. At the low level you always need to write algorithms with some very complex math like RAID.
They provide no technical info to backup their claims. If they created something so much better than RAID then they should give some low level technical info to let people understand why it would be so much better than RAID. And why it would be as much or more reliable...
But I really think that this is all just marketing b*ll here...
"....So they aren't using any RAID algorithms to create and manage redundancy on the drives arrays ? Really? I doubt that!...." I suppose it depends on your definition of RAID. If they are simply storing files in a NAS-manner, and simply make an on-write copy of the file to a different disk pool in the system (software mirroring), then it is what some people call RAID1, but a high-end array designer that uses dispersed blocks across multiple disks might consider too parochial to call 'real RAID1'. It also doesn't address the performance issues of mixing disks - real arrays like identical disks as then your RAID operations don't get held up by a slow disk taking longer to acknowledge the write than a fast disk. Performance is also more predictable. I assume some form of storage pool is used to allow the to write across disks without having to worry about the different sizes of disk. The 'ooh look, straight onto the network' feature just looks like DHCP, only with no security over who on the network can then store or copy stuff - if the marketing guy beats you to it he can stuff all his massive slide sets on there before the CIO gets a chance to store his more important docs.
OneBlox would seem of interest as a cheap NAS solution for the unstructured adding of storage, but then there are already plenty of those on the market. Indeed, a Windows Storage Server with a few installation scripts could do pretty much the same, on an OS with 24x7 support on an international scale.
So why no Channel Partners listed on their site?
How much does it cost? Saying less than $10,000 isn't good enough - what does it really give me that a basic server and Windows Storage Server or some Linux freebie give me?
All very well saying no to RAID, but RAID is understood and mature.
RAID is about reducing the risk of data loss due to the physical failure of a spinning disk (in a nutshell.) Object orientated storage is a storage architecture that manages data as objects (in another nutshell.)
Bringing a box to market that implies no RAID, their words not mine, and is totally unsecure when joined to a network is obviously what every CIO and CSO love to buy.
Stick to the coffee lads.
They only support SMB (CIFS) ? Really? I think most would argue that NFS is a far more commonly deployed protocol amongst servers anyway.
I suspect these guys are not using RAID as in mirroring whole drives, they are probably mirroring (or more likely triple mirroring) objects across systems. So if a drive fails the objects are mirrored elsewhere, you don't need to wait on that drive or a hot spare or something to kick in. They may be using RAID for the operating system disk(s) on their controllers (or perhaps just a small SSD).
3PAR has a nice RAID system too which is similar, breaking the drives up into 1GB (as of a few years ago before that 256MB) so rebuilds are very fast and the design is quite scalable. You can even yank a drive out of an array and the system will go into "logging" mode writing data that would go to that drive to other locations for up to something like 7 minutes at which point the system assumes the drive is dead and rebuilds. You can gracefully evacuate drives as well for seamless maintenance.
IBM's XIV does something similar too I believe but last I checked they still limited themselves to 7.2k RPM disks and RAID 1 only (I suspect their CPUs can't keep up with the calculations for RAID 5/6 at that level, with 3PAR that is handled by their ASIC).
XIO (assuming that's what they are called now) does RAID across drive platters, which I thought was quite creative too, they can fail individual platters in drives and not have to replace the disk (system ships with enough spare capacity that you don't take a capacity hit over the 5-8 years or so of the system warranty).
It's not uncommon for a 3PAR array to have more than 50,000 RAID arrays on it with just a few hundred spindles(much more for those on systems older than current generation since the size of the chunklets is smaller).
"Vendors such as EMC and NetApp are just now supporting 4TB drives and charging a 10x premium relative to retail pricing"
While EMC and NetApp may charge a premium for drives, There is more to is than just the hard drive itself. Somehow I don't think Best Buy or NewEgg are going to send a live person out to your business at 3AM on a Saturday to fix your system or replace a failed drive when it is down and do that for 3-5 years as part of the price. I love how people say that they can just go to a retail store and buy a drive for less, you can do that, but there is more to data storage then just buying a drive off the shelf. It costs money to warehouse parts in every major city and have technicians available 24 hours a day, 365 days a year to do repairs. Retail stores don't provide these services or incur the costs associated with them, so of course they can charge less.
So you just buy some spares and swap them in when needed. No need for a warehouse as 10 drives (or more if needed) could probably be kept in the same location as the unit, they are available 24/7 and pretty sure most IT types can remove a faulty drive and plug in a new one without too much trouble.
Hard drive = commodity item.
"So you just buy some spares and swap them in when needed...." Hmmmm, so how good are the tools bundled with the OneBlox to help you identify the problem disk (or disks) and help you rebuild the storage pool(s)? What happens if you have used any old disk (as in the demo), can you just plug in any old replacement or does it need to be an identical replacement? Do you then need to keep several stacks of replacement drives? Can they support you at 2am on a Sunday morning if the bundled tools don't find and fix the fault? Does it guarantee to recover data, can you replace disks online, and does such changes interfere with replication to other OneBlox devices (do you have to take the storage pool offline to change or add disks)? Which disks have Exablox actually tested with and are willing to support, and which ones does that mean I am on my own with if not? Exablox may have answers to all those queries but they are not obvious from the demo vid.
".....Hard drive = commodity item." Yeah, I used to think that. Then I built a Linux server a while back with a (cheap) SCSI card and some off-the-shelf disks, saving myself a few hundred quid. All went fine until I updated the SCSI card driver as part of the OS upgrade, at which point half the disks went dead. Luckily, before just dumping them, I plugged them into a different system and they worked fine - it turned out my 'identical' disks had different firmware. I then spent several days trying to track down the 'good' firmware and flash it into the 'dead' disks as the flashing process was not supported with the SCSI card I had bought. Finally, having flashed the disk firmware, I found out the flash had somehow destroyed the inode tables on the disks and my data was unreadable (the disk manufacturer first said that was 'not possible', then finally admitted it had happened before). Cue a re-install from backups, luckily without data loss. After that I was much more wary of the idea of 'commodity'.
Assuming that it wasn't possible to roll back the driver upgrade, I think what I would have done would have been to replicate the setup of the server up to the point just before the OS & SCSI driver upgrade to see if that worked. Frankly, after a standard kernel update on an Ubuntu test server resulted in an unbootable machine, I would never do a critical system update without making sure I had a full backup beforehand (or at least taking critical drives offline beforehand).
I know of at least one tool (the name escapes me at the moment) which claims to be able to recover a number of different types of RAID configuration. Thankfully I've never had to try it.
Also, since Exablock (and would you trust anyone who talks about coming out of "stealth mode") are touting their system as object-based, I would expect that once the bad drive has been swapped out the system will automatically reconfigure itself so that the new drive becomes part of the array with overall capacity adjusted accordingly. If it does anything less than that, well they're just talking out their arses.
Anyhow, RAID is RAID, whether the drives are all sitting in the same box or scattered across the globe over iSCSI, the idea is the same. The whole "RAID is dead" trash talk is just a red herring to make it seem like they've come up with something fantastic when in fact it's probably no different to a number of other products on the market (or what any semi-competent admin could roll themselves).
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
