The Infrastruggle

The Scourge of 'Software-Defined'

Deceptive marketing, including overpromising, plagues the industry.

• By Jon William Toigo
• 02/12/2016

Originally, I liked the idea of software-defined storage and software-defined networks. I had long contended that, in storage and switches, when you got past the layer of proprietary "value-add" code that vendors added to their kit controllers (in reality, just PC motherboards running apps), you were dealing with some pretty commoditized hardware. To my way of thinking, commodity equaled inexpensive: if you bought components separately, they tended to be quite inexpensive. So it wasn't a great stretch of causal logic to attribute the high cost of "finished" products to storage or switch vendor greed, manifested as proprietary software.

When a Network Isn't a Network
What really alienated me from pre-integrated gear, besides price, was the sleaziness of vendor marketing. Storage-Area Networks (SANs) weren't really networks at all, as most vendors will now confess; but vendors called their array products SANs because it was happening and cool and they could charge big bucks for it.

Another example that comes to mind is one of the first deduplication appliances that came on the market a few years ago: it consisted of 300 1TB hard disks in a commodity chassis with a server motherboard head running a dedupe algorithm for an MSRP of $410,000. Since I could buy the drives on Newegg.com for about $80 each and could get the chassis cheap on eBay or Amazon direct from the Taiwan manufacturer (cheaper if you bought it in Hello Kitty pink than in the black matte finish preferred by data centers at the time), I calculated that you were looking at maybe $3,000 for the whole shebang. On what planet, I wondered aloud, was $3,000 of hardware kit worth $400,000+?

Little White Lies, and Big Black Lies
The truth became apparent when the vendor tried to use the "Aladdin" sale (as in Disney's "Aladdin") to convince me to buy the product: "Phenomenal cosmic power. Itty bitty living space." They said they could charge such a markup on the underlying platform price because 1) with their dedupe special sauce, every hard disk would be like 70 hard disks, and 2) because that was what folks were willing to spend. Then they explained that with volume purchase agreements, incentives and discounts, they were selling the kit for a little more than half that amount. Still too much, I decided, since nobody ever got a 70:1 data reduction ratio from the kit.

Anyway, switch vendors also lied, cheated and obfuscated to sell overpriced gear. But, like storage gear, overpriced commodity switches laden with "value-add software" continued to fly off the shelves. People actually bought gear based on the name on the bezel plate, which was often the only feature that really differentiated one product from a competitor's product.

I got to a place where I seriously wondered why consumer IQs had dropped so sharply. Didn't anyone care how much money they were wasting? I quit writing books for a time because I wondered what I could possibly contribute in a world gone insane.

Initial 'Software-Defined' Excitement
When the software-defined thing came on the scene a couple of years ago, my initial reaction was positive. Good, I thought, finally someone is pulling the plug on all of the proprietary, overpriced marketecture and substituting commodity gear with portable software functionality.

Basically, you could take all that dedupe and thin provisioning and replication and backup and RAID, and all of that proprietary switch software functionality that didn't need to be placed on the controller and host it instead on a server. That way its "chewy goodness" (a.k.a. "added value") could be shared across all ports or spindles or flash drives or whatever, instead of being isolated to one set of media or circuits.

The VMware Agenda
That's when I realized that the motivation of some of the leading proponents of software-defined weren't recommending their architectural vision for consumer benefit at all. You didn't have to squint very hard to see that VMware had an agenda to use software-defined to seize control of the entire technology stack, a la IBM in the early 1970s. The cost for the licenses to its preferred software-defined storage (SDS) model, plus the cost for "certified" storage node hardware, for example, came very close to the total cost of the proprietary arrays and SANs they were replacing.

Moreover, making a switch to their SDS model, for whatever that got you in terms of centralized service management, did not address the problem that motivated the change in the first place: poor application performance. Unplugging legacy gear and plugging in 3-node SDS clusters didn't make apps go faster at all. The problems impairing the performance of most applications were linked to raw I/O issues, and not to storage I/O at all. Following their guidance wouldn't solve your storage woes; it would simply lock you into the VMware stack and isolate your hosted data from competing stacks.

Not all SDS or software-defined networking (SDN) solutions are evil, of course, but they do encourage a way of thinking about things that will come back to bite you if you aren't careful. A major problem is that the woo around software-defined -- the whole "hardware is trash" thing -- makes us vulnerable to hardware level problems that nobody will have the skills to solve.

The Dying Art of the Hardware Fix
I recently had a system go down, and took some meters to the circuitry on the motherboard to see if I could find a fault. (Remember when we used to know what we were looking at on the circuit board, or how to use a test rig?)

The problem was a capacitor that had failed, presenting the blistered head that signifies such things. I went online to find a new capacitor and read a ton of articles decrying the shoddy state of capacitor making these days. Blistered caps, resulting from lousy manufacturing processes or bad materials, were impacting everything from monitors to wide screen televisions to PCs and laptops and expansion cards to home appliances. They are apparently the devil's answer to all of that heavenly Internet of Things stuff. Only, very few IT folks today would know what a capacitor looks like, let alone how to find a replacement, de-solder the defective cap without hurting anything else, and replace it.

Just because software-defined sounds cool, that doesn't mean that we can afford to forget everything we have learned about hardware. There is no app for hardware, and simply replacing any node that fails with a new node is a sure road to bankruptcy.

My two centavos.