I know for certain what software-defined storage is. It's the new black I’m not going to re-define software-defined. Instead I’d like to look around and try to make sense of different interpretations and architecture designs that make this claim. I have intentionally left out some solutions because they are not software-defined for me, while for others, I may have just forgotten to mention them or …
I disagree that PD is only attractive t those wanting to consolidate multiple storage platforms. I think it's the perfect scale out solution. One can pick best of breed NFS storage as building blocks and keep adding as needed. The ability to direct data between the underlying storage is also attractive.
I have no beef with software defined storage but there is an additional perspective on the decline of storage revenues.
Once storage belonged to EMC, IBM, Hitachi, HP and NetApp. Some competition but a narrow club and storage was (a) difficult and (b) expensive, where ever you looked.
These days Nimble has gathered >8,000 customers who love them, Pure has stolen 1,500 of EMC's prize house accounts and the hyper converged folk can probably answer 10% of storage needs.
There is fierce competition for the old guard, so even when they do win time and again their prices are forced down. It has come to Dell (shrinking) buying EMC (shrinking).
It's a funny old world were storage is interesting again. I think SDS and HCI have their place but I suspect most of the decline in the storage market is the added spice the new kids like Nimble and Pure bring to the game. Tasty times.
Enrico,
I really enjoyed >80% this article, beginning with the premise and ending just before you got to the point of defining SDS.
IMO the glaring omissions here are the market leaders, HP's StoreVirtual VSA (which evolved from LeftHand networks circa 2009) and DataCore's SANsymphony, (which started in 1998 and began blowing 'hardware defined' storage out of the water in price-performance with their first SPC-1 all the way back in 2003, and continues slaying the 'hardware defined storage' beasts even today). There are many more examples.
I have a slightly different take on what it means to be "Software Defined Storage". Call me simplistic, but in my mind the first criteria is that it must be software-only (i.e. not necessarily a 'hardware appliance' sold by a 'storage software' vendor) and secondly that it must NOT be "hardware-defined" (i.e it runs on any servers, including ones I already own).
Bottom line -- If I can completely separate my storage software purchase decision from my storage hardware purchase decision, including separating the hardware refresh cycle from the software upgrade cycle (like I do when I chose my database software vendor), then I have "Software Defined Storage".
Beyond that, what's the point of limiting SDS to "scale-out", 'Shared-nothing" and "API-based management"? Those are architectural considerations that have nothing to do with whether any given solution is "software defined" or "hardware defined".
Software-Defined Storage is any storage solution that completely disaggregates the software purchase decision from the hardware purchase decision. In this context, since I can't buy Nutanix without buying Nutanix-defined hardware appliances, then Nutanix is no different from Isilon, or any other hardware-defined solution. Nutanix then would not be 'software defined'. As a Nutanix customer, when you realize that your Nutanix software license expires when your appliance server hardware goes EOL, then you realize that Nutanix is not "software defined"...no?
If these seem like rhetorical questions, then they probably are.
Can a Software-Defined storage solution exist without some vendors proprietary file-system implementation? Here, DataCore, Falconstor and Starwind (among many others) come to mind.
What about IBM's decades-old GPFS (General Parallel File System) that runs on any servers/storage how is that not SDS?
Likewise, if IBM really does begin selling XIV as software-only "Spectrum Accelerate SDS" (rather than just marketing it that way), and I can then run it on whatever x86 servers/disks I want -- would that be SDS?
I can't help but notice that all of the companies and technologies you list to constrain the definition of SDS are either venture-capital fueled startup companies or newly minted open-source initiatives.
I'm puzzled by this but IMO, it's all just the Hype-Cycle talking.
Thanks for a thought-provoking article and for inviting the question(s).
Hi,
thank you for chiming in,
I don't usually follow up with comments here on The Register, but this time I asked for it... ;)
I didn't mention everyone in the post, it's impossibile, I just choose to mention a few of the companies I met recently... and I agree that the number of players is much larger than those that I mentioned.
At the same time I tried to find a categorization of SDS (which is hard frankly) and I intentionally left out all "traditional" virtualization solution. Not because they are not good, but because they don't fit exactly with the premise of the article (split of data and control plane, with control plane out-of-band).
But again, you can't go very deep in a short article and it was intended to start the conversation!
Enrico
Following up on the IBM mention.
IBM has probably the oldest SDS in the market (GPFS) and now covers block, file and object -- all in in pure SDS. IDC Research has named IBM #1 in SDS with a portfolio of Software-Defined storage.
You _can_ deploy Spectrum Accelerate, based on the XIV software, on most x86 servers. That includes license portability so one can switch from Supermicro servers to servers on SoftLayer.
IBM Spectrum Scale (the new direction of GPFS) is absolutely SDS. You are right. GPFS has been around for 20 years this June - most of them running on storage that was not just IBM. IBM Spectrum Scale places data according to policy - from Flash to Disk to Tape (in Beta to Object & Cloud) - and metadata is separated from data so you can tune the storage portfolio for the data value. That hits the points in the article.
One of the reasons SDS got so much attention was the rise of HDFS and object storage to build large data storage on storage rich servers. It is a seismic change in the negotiation of storage pricing when you can choose Supermicro, Quanta, Dell, Lenovo, etc. That trend certainly drove IBM's Cleversafe acquisition. IBM Spectrum Scale has an option and license specifically for storage rich servers and a fully transparent HDFS for Hadoop and Spark.
Yes, I work for IBM. No, this is not an official IBM anything.
... is centralized computing.
IBM, late 1950s, early 1960s.
Nothing new here, except marketing leveraging idiots who have absolutely zero clue about computers ... and as a side-note, the marketards obviously have no idea about the technology they are pushing.
Like many other companies, we have a ton of VMWare deployed. SDS solutions like Datera are not integrated with VMware and thus have a while to go before they can replace our EMC or HP storage solution. We have just started with Maxta in a few of our environments as they are much better integrated and thus I can use my current servers and not buy new hardware while joining the Hyper-converged world without busting my budget.
The phrase comes from "Software Defined Radio" (SDR), where a tunable wideband RF circuit combined with a fast DSP substitutes for fixed-function electronics. The Software-Defined Networking guys cribbed it from SDR.
SDS is the same promise as "open": cheap and flexible unlike the proprietary black boxes. Of course, only open-source solutions can be trusted, a vendor will always try to bring back lock-in.
It's a ridiculous "category." It's too easy to misunderstand. Please let it die and stop bringing it up.
Categories should group products by consumption model and function to help shopping. SDS fails this test dramatically. It's being used to describe products which come with hardware, or which require certain hardware. It's used to describe products serving completely different layers of the storage functional stack, so it's not helping sort them for shoppers.
It's not quite dead, but it's pining for the fjords. Please stop trying to fix it. Come up with something that works better. Chris M tried to fix/kill it once before, and 2 years later, it needs to stop.
http://www.theregister.co.uk/2014/11/11/confusion_abounds_in_the_land_of_software_defined_storage/
There are 2 important aspects of SDS - "Software" and "Storage", which distinguishes it from a FileSystem.
The software needs to replace the storage-management functionality of the hardware. Not all Filesystems replace storage-management - esp. the block level functionalities.
So, the SDS test is - Deploy the software on a commodity x86 hardware ( aka -"any hardware") and if one gets all the functionality of an enterprise storage hardware , then the storage software can be called a SDS solution.
Obviously, there is a grey level - storage functionality might range from block, file, object ..resiliency,optimization...etc.
However, a better SDS solution is one - which can add more storage functionality - on the same commodity hardware.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
