What do you see as the role of solid state drives today and how do you see that changing over the next five to 10 years?
SW: I've been a very big fan of solid state storage. I had dynamic RAMs backed up with power in huge tower-like structures that I'd connect to early Macintosh computers. I'd use that as my disk and it would double my speed. I've been a fan of solid state disk in my laptops - I always ordered them and paid extra for the solid state disk. I'm just sort of a purist. With music devices, even before the first iPod, I preferred to use solid state storage when it cost a thousand dollars to have enough solid state storage for an airplane flight. So, I'm just a big fan of that storage media.
SW: It's just really interesting to see at the enterprise level that it's outperforming and solving a lot of problems we're having with spinning media. It's a very different approach to how you solve your storage problems. I like to say solid state storage, not solid state disk, because a disk implies it's on a regular cable, plug-in structure that a disk could be plugged into and this is one alternative. I like skipping that and just going to just one protocol in communication.
SSD appears to be finding great success in enterprise-class server systems, but not so much in the PC and laptop market. Why is that?
SW: Hard disks over a certain size are still more economical than solid state storage. You have to pay a premium to have solid state storage, and depending on the way it's architected, it may not be faster in all ways. I have flash in my laptop and it boots up and loads programs much faster than before, but the general operation doesn't seem that much faster. So why would you pay extra when you don't see [the added performance]?
I mean I'd pay extra ... because I like the whole concept of reliability [with flash memory], and I like to be a purest and like to be a little bit in the lead in doing something. I like to have an early jump. So I had this pension for solid state storage before the fact that it now has a place in enterprise storage. That's what really surprised me, and of course now the whole world is recognizing that's a big part of the future of enterprise storage - at least on the first tier.
Why has solid state storage done so well in the enterprise?
SW: There was a vacuum. The solution hadn't been recognized early on. Even the personal computer could be thought of but it couldn't be designed until the price of chips due to Moore's Law came down to a certain amount of power in chips for a certain price. In flash memory, that wasn't obvious at first - how low it would get. But as flash memory went into consumer devices [MP3 music players, USB flash drives, for example] and became such a huge product ... its price came down.
When I first talked to the Fusion-io people I asked how would it compare to a bunch of disk drives in a RAID array. And, how about in an enterprise server? And they said we beat them 2 to 1. I'm a practical person and I look at what really works and what doesn't, and it's incredible how much equipment they save and how much complexity they save and how much power they save. Almost everywhere I go ... I speak to CEOs of companies and when I tell them what our product does they say, "Oh, my god. We've got to get onto this right away." We've got the solution to so many problems that people are going through right now.
How do you see the PCIe form factor addressing the bottle neck issue between the CPU and the application as opposed to using SSD in a 2.5-in or 3.5-in disk form factor?
SW: I look at it as a more efficient, more direct connection to the high-speed bus. That's where you want it. The processors are so fast now, you've got so many cores, that they're no longer the bottleneck. It's always been the disk channel [that's been the bottleneck]. Sure, you can come close in bandwidth if you plug a Fibre Channel board into the same slot. But why do you want to transfer all your data into the Fibre Channel mode and then transfer it back. Why do you want to go through extra boards to cables to more boards and cables and then transfer it back? Skip all that middle-man stuff. That's the same way I thought when I was designing computers, storage and disks. Re-visualize the entire problem.
You want to get a little data from here to there, but you don't want to go through extra chips, extra parts, extra pieces. That's unreliability. The number of pins and connections you have is more related to reliability than the overall number of transistors these days.
I also like the fact that you just plug a board into a computer. I'm one of these guys that doesn't like to see bundles and bundles of cables. You know, where does this one go, and did I plug that one in the right way? Then you have to test both ends of it. It becomes a mess. I've got racks of equipment in my home and I don't like that. It's an overall mess that you look at and think, this scares away the common person. It scares away a child. I really believe that when you make something so simple that anyone can look at it and understand it, it even makes it easier for the engineers to understand what's going on where. That was one of the big things that turned me toward Fusion-io - using this form factor compared to the solid state disk approach.
Do you see a day when solid state storage will kick all spinning disk out of the data center?
SW: No. I don't see it kicking all spinning disks out. In computers we have so many tiers of storage for cost efficiency. Even when you have a hard disk drive it has its own cache built into it. Then we have caching systems in operating systems. Then we have different speeds of memory from your RAM to your L1, L2, L3 caches. This is an in-between one, but I think it's going to be huge - a lot bigger than people think. It cost more money per bit to create NAND flash ... [but] in a lot of places kick out spinning storage. I can see certainly in a netbook you don't want spinning storage. If you're talking 64GB or less, it's less expensive to have flash solid state disk now.
But in a big enterprise-class data center there are huge amounts of data that aren't accessed very often. It's just mathematics. You take stuff that's not accessed very often, it can be accessed slowly. And then you bring it into a faster form of storage when it is being used a lot.
So I can see solid state storage in the enterprise as a type of cache, without the same programming and design structure as cache, but serving that same purpose.