Enthusiasm over flash technology has many being swept along in the inevitable wave of solid-state storage products coming to market, but it still pays to be cautious.
As a curmudgeon when it comes to most technology marketing, my initial reaction to technology trend projections tends to be actively contrarian. I say actively because I'm not satisfied to simply change the channel, turn the page or delete the email that contains the hype. Instead, I get to work: I research, conduct tests and pose questions in articles such as this one to deflate the marketecture. My goal is to help the unseen hand of Adam Smith's capitalism do its work: Give consumers the information they need to make rational choices about the architectures they adopt and the products they buy.
As expected, this mission doesn't get me on the Christmas card lists of corporate analysts and media relations flacks. I receive far fewer invites to vendor soirées for analysts and pundits than would be the case if I simply echoed the positions adopted by vendor spokespersons. I get less swag, fewer whitepaper assignments and fewer paid speaking gigs. In short, if I depended on vendors for my income, I'd be very poor.
The good news is that, while I am mostly unloved by vendor marketing folk, I generally command some grudging respect among vendors' engineering staffs. As much as the marketing folks tend to try to gloss over the niggling problems with their technology, engineers have the difficult task of dealing with product flaws head on and finding ways to work around the deficits. No technology is perfect, and I have enormous respect for the miracle workers who find ways to make a product work reasonably well despite its warts. I usually sense that this view is recognized by vendor engineers, and that I receive at least a modicum of their respect for acknowledging their challenges.
That's a long way of saying I got a bit peeved the other day when I was talking to some storage software engineers regarding their views on flash memory. It was in a meeting to discuss some work they were doing around flash technology that's intended to make it more useful in virtual server environments. My first question was whether they agreed with analysts that flash storage was the "next big thing" in storage.
One engineer looked at me with incredulity in his eyes. You're kidding, he seemed to be saying. He flatly responded that disk is dead. Flash will replace disk storage in the next six months.
His fellow engineer amended that projection. "Actually, it might take another two or three years, but flash will destroy disk."
They bolstered their arguments with some of the familiar marketing messages I've read in vendor marketing brochures, such as flash is less expensive on a per-gigabyte basis than disk when spread over its useful life. It's just like the new fluorescent light bulbs, which are more expensive to buy than incandescent bulbs, but they're also more durable and return their investment over a longer period of time.
They advanced another familiar argument that flash devices consume far less energy to deliver IOPS than spinning rust, both in terms of power for the device and HVAC to evacuate the heat generated in operation. That was more than an eco-plus; it was a real energy cost savings for businesses.
They also told me to forget what I'd heard about memory wear. They cited a prominent flash vendor, who recently insisted that memory wear is exaggerated. Products ship with a lot of extra and unadvertised capacity that's substituted for bad cell groups when cells burn out. Besides, if the flash device burns out, you just buy another one.
Related to the above, you can afford to buy new flash to replace failed devices because the products are poised to get very cheap, the byproduct of an oversaturated market that has way too many vendors in it. The flash market is ripe for a shakeup, and all the vendors will battle it out -- delivering substantial price cuts and virtually giving their wares away -- until the bitter end. Users will be the beneficiaries, and the effect will be to displace all that spinning rust at an accelerated rate.
I was overwhelmed by so much enthusiasm. Hadn't I read about new kinds of memory in development that weren't subject to wear and that might replace flash altogether? They didn't answer. For now, according to the vendors they listen to, flash chips were in such huge supply that you had the choice of riding the wave or being inundated by it.
I wanted to ask other questions, like what applications are best served by flash, or whether it might be better architecturally to establish a multitier storage infrastructure with different storage media, including flash comprising different tiers. But they weren't having any of it; the future was all flash or no storage at all.
Apparently, flash storage will displace all other storage simply because it exists. Not because flash speeds up VMware (it doesn't), not because it's more cost-effective than alternatives (it isn't), not because it's more manageable (it's not), and certainly not because interconnects and processors can support I/O at flash speeds (they can't).
To hear these guys tell it, flash technology succeeds because engineers are bored working around the deficits of disk storage and prefer to work around the deficits of silicon storage for a while. Pundits, analysts and consumers need to just shut up and go along for the ride. The architects have become the marketects.
About the author:
Jon William Toigo is a 30-year IT veteran, CEO and managing principal of Toigo Partners International, and chairman of the Data Management Institute.