First of all a disclaimer: I recently did some work for Exablox and I met them again last December at the IT Press Tour as well, which is also why I got the opportunity to spend more time with their product and their technology.
An update about the company
I met Exablox for the first time one year ago, when OneBlox (their product) was publicly launched at Storage Field Day. I was amazed by the architecture and the ease of use. I also immediately liked their cloud-based management tool called OneSystem.
The architecture was designed to scale without limits but the company’s strategy was to start with a product, OneBlox, clearly targeted towards the SME market. This means that it was packaged in a cost effective appliance and the first protocol exposed was SMB (Exablox now supports all SMB versions, including v3!). In the following months NFS was added alongside many other features and improvements. In this first year the company has won more than 70 customers, a clear sign of success… also because most of them continue to acquire more and more space/appliances.
The Genius is in the details
I know, it was the Devil but in this case it is the genius. One of the most interesting features of OneBlox is the fact that it can use off-the-shelf commodity hard drives that the end user can buy wherever!
In practice, the only thing you buy from Exablox is the chassis (and support), which comes at less than $10.000, after which you are free to buy the disks you like most. SAS or NL SAS 3,5” hard disks from any producer/manufacturer and at any speed. You can fit up to 8 of them in each single appliance, and you can concurrently mix sizes, models and speeds in any possible configuration. You can start with 1, 3 if you want full data redundancy, and from that moment you can non-disruptively upgrade the system by adding more disks or nodes. Small organizations love this aspect, just because they can go on Amazon and buy a hard disk for a few bucks and if it breaks down or becomes small it’s probably cheaper to buy a new bigger one than send it back for repairs!
The rest of the architecture
OneBlox is a 100% object-based scale-out storage system but Exablox has decided, very intelligently, not to expose object storage APIs at the beginning. Who cares about object storage in the SMB?
As I mentioned before NFS and SMB are the protocols now, which also means that you can use OneBlox as a datastore for your VMs. Performance is not on its side yet… but they are cooking up something new for 2015 and I’m sure it will be ready to be shown/presented soon. In the mean time, you know I’m a proponent of server-side caching too, you could use OneBlox in conjunction with one of those interesting products that I often mention, I’m sure you would be surprised by the performance and the $/GB of the entire solution… 😉
In any case, things are quickly changing and I know they are discussing whether or not it is time to propose API access to their customers. Backup software are starting to use it, some applications are doing the same and some T1 arrays are starting to destage/backup snapshots via S3 (like Solidfire for example, but others will soon follow). In some instances it could be a great enabler to help small organizations with particular security needs to deploy innovative private cloud services like sync&share or ROBO NAS. For example, an infrastructure based on OneBlox and product like CTERA could be very appealing even for the smallest of the local service providers that don’t/can’t store data outside country borders!
If you look around, you can easily find that some enterprise storage arrays have recently started to implement object storage APIs, think about HP 3PAR for example, so it wouldn’t be that odd for Exablox to do the same!
On the back-end side, the system makes multiple copies of any stored object leveraging a ring-based data distribution, which gives a very high level of data protection. But if you think that a multiple data copy mechanism introduces space inefficiency when compared to traditional double parity RAID (e.g. RAID 6) you are wrong (in this particular case at least). Global deduplication, which is always enabled, reduces data footprint and re-establishes a good overall efficiency to the cluster. On the other hand, deduplication is quite uncommon in object storage systems (encrypted and compressed data are less compressible and erasure coding is often chosen instead). But again, it has its advantages and it’s much better than other techniques for storing backups and VMs (for example).
And, even though we are talking about an SMB product, you can find some really nice and well implemented features like remote replication, non-disruptive firmware upgrades (with an adequate number of nodes), automated call-home support and so on. A really interesting, well implemented product at the end of the day!
What’s next?
Well, as I said, you’ll have to wait a couple of months to find out, but I bet we will see a new hardware appliance for different reasons. More slots, 2,5” drives maybe, more performance and an improved form factor (a proper 2U rackmount unit) are on the top of my list. But I’m also betting that new software will come along with it as well. Long story short, a good product that is maturing while maintaining its core value!
And the good thing is that current customers won’t be left behind. Exablox has already confirmed that different nodes will live together (it’s already happening with current installations, indeed).
Closing the circle
OneBlox is a nice product that is maturing very well. I liked it from the start and Exablox is doing a really good job with it.
This year will be very important for Exablox: new (faster) hardware and features will open up many more opportunities while a bigger sales channel (they are beginning to look more seriously at Europe too) will help to reach out to many more potential customers. Especially in the EU where SME is a much more larger market segment than in the US.
Last but not least, I wish they would finally release an object API interface. Cost-effective gateways and other application-based storage solutions (like sync&share for example) are much more accessible, cheaper and easy to deploy now than in the past… I think they could do surprisingly well by enabling this feature.
