I talk with a lot of enterprise and cloud computing companies, and regularly get the question “Just what is Software Defined Storage (SDS)?” and what’s hardware’s role in it?”. It’s an unclear and potentially imprecise topic to someone who isn’t an expert in storage technology (and doesn’t want to be). The reality is I have seen some unbelievably elegant solutions SDS, and I’ve seen a lot of train wrecks due to cheap hardware. At the end of the day, when it comes to Software Defined Storage – what is critical to recognize is that the hardware still really matters.
The lines that separate different functional/physical layers have been blurred by all of the Software Defined players, and storage technology has its own design quirks as well.
Ok. So which one is the Ferrari 308… the car that the television show “Magnum P.I.” made famous in the 80s? Would you believe they both are?
One description “back in the day” of the Ferrari 308:
“From its sharp nose incorporating a slim bumper and a deep air dam, to its retractable headlights and row of black louvres that vent air from the radiator, the line flows up the windshield and out around the flanks to reunite with itself at the buttress C-pillars, ending in a very subtle rear lip spoiler. The design is so beautiful and effective that it has been a basis for exterior styling of every subsequent V8 Ferrari and an object of study for design students the world over.”
The thing about this picture is that this abomination isn’t a Ferrari, it’s a Pontiac Fierro with a Ferrari 308 “body kit”. From the lights, to the gear shifter, to the emblems in the leather; (and as long as you didn’t look too close…) it looked just like the one on TV. This is absolutely parallel to the approach that most are taking to Software Defined Storage these days. It’s about software that has all of the enterprise features and options, on the cheapest commodity hardware the vendor can find. While this is a great idea for something parked on the beach, it’s a horrible idea for basing your business on.
The companies that I talk with aren’t doing mid or enterprise computing to get the girls or to make their friends jealous (Ok, some are, but that’s a tangential argument). They are in a race to make money. Just imagine what would happen if you really tried to race one of these counterfeits with other real exotic cars… You might be able to stay on the track for a little while (and get passed a lot), but that engine is going to burn up and the transmission is going to drop out before you even get through the first tank of gas. You better have a truck load of those things just to finish the race, and that’s exactly the same design methodology that many storage vendors/SDS solutions are using. When it breaks just throw it away and get another one.
Never mind that you are going to need a TON of them to finish the race (horrifically expensive for the vendor, not all will survive), it’s going to be extremely painful while you switch out the cars (for you and your customers), and that’s going to drive your total cost of ownership through the roof. Performance? Not so much. And, it’ll work for a year or so, and then start having problems, but by then it’s too late to return it.
The problem here isn’t the Ferrari body kit, it’s the Fierro. To get in the race and win, let alone compete, you need a powerful, efficient, engine with the capacity to do the job without pulling into the pit stop every other lap. Hardware really does matter, and it’s the hardest thing to fix. It’s fine for make believe movies like Magnum P.I. but in the real world you can’t just patched with more software redundancy.
UNDER THE HOOD
One could make the argument that only companies such as FalconStor, Datacore, and Sanbolic offer true SDS solutions, as they can be hardware agnostic (server/storage) and configurations can be tailored to the end user requirements. I would also include ZFS/CEPH/Hadoop/ Lustre/etc and other open-source software here as well. The underlying hardware can be chosen by the end user, giving lots of customization options (but this can also be a bad thing). From the storage management/ presentation/ administration side, it’s designed to be the front end for different kinds of storage systems with all of the features provided by the software. This approach basically comes down to hardware (server/storage) and software (SDS).
But if one could “fix” the server configuration (standardize server hardware), then that would solve the server problems that can come from a DIY approach. Here is where I would say that solutions such as the Hitachi VSP, USP-V, IBM SVC and other “appliances” are just hardware versions of SDS. These solutions are designed to virtualize 3rd party storage equipment behind them. Standardization on hardware can be an advantage here, but these companies still have margins to make (and usually with the cheapest hardware that’s good enough’ish).
If you go a little further down this path, then you not only lock in the server that has the SDS software on it (a storage controller), but the storage media and interconnects that the software will be talking to. This is where I would include the recent entrants to the storage market such as Nimble, Tintri, Pure, SolidFire, and Tegile. One could also make the argument that the VMware EVO:RAIL configurations are just SDS software from VMware, with cookie-cutter hardware configurations controlled from various vendors. Costs are a major issue for the vendors in this space, and the usual causality is the hardware platform these solutions are based on. So the software has to make up for all the same problems of traditional enterprise storage, and add to it the common use of SATA, NL-SAS, and consumer grade flash technology in a device that’s going to get pounded on 24/7/365. Some of these companies are in “growth mode” with marketing eating up a TON of money, and all other divisions get squeezed for resources. (Someone spent almost $37M in sales and marketing last quarter, almost 2x more than R&D, while the business was losing $0.39/share! There’s some Liquor and Guessing going on…).
I started my career doing storage hardware qualification testing with DEC/Compaq (G80 and Z70 controllers anyone?), and I’m even further on the extreme in that every storage system is really just SDS that’s “Gorilla-Glued” to vendor “specific” hardware (Custom boards/ASICs). This is where pretty much where all other traditional storage solutions sit. Hardware development costs are astronomical and have not been reasonable for a while now. These are the devices that are getting “long in the tooth” in your datacenter. Everyone has pretty much adopted the concept using commodity components that are specified and sold through the vendor.
VMware is even taking this a step further with the development of the vSphere APIs for IO Filtering (VAIO). This approach has potential written all over it. Any vendor could write a plug-in for this architecture to provide data services. Need to replicate your VMs for Disaster Recovery? Don’t mess with a storage vendor! Just download the plug-in from Amazon, Google, Microsoft, Rackspace, etc. and replicate to them with no hardware costs. Need storage to be encrypted from end-to-end? Get a plug-in for that and it doesn’t matter what storage system the VM exists on, it’s always encrypted from the Hypervisor. The recent acquisition of Sanbolic by Citrix, Seagate eating up the Xyratex/ClusterStor solution, the rapid announcements in the VMware EVO:RAIL, VAIO, and VASA programs should be red-flashing light indicators that there are large movements going on in the market.
Having that conversation is harder than it sounds. Don’t want to have that conversation?
With over $100M in investment from Seagate Technologies, you might want to consider the ISE G3. It’s the latest version of our Mr. Fusion for your SDS “body kit” of choice.
Click here to learn more.