While most major disk-drive manufacturers have developed or are already selling solid-state disk drives or hybrid drives, which use a combination of flash memory and spinning disk, Fujitsu Ltd. has chosen not to develop a product for market. Joel Hagberg, Fujitsu's vice president of business development, said his company does not plan to launch any solid-state disk-drive products over the next two years because the value proposition of the technology is not compelling enough and won't be until technology breakthroughs change solid-state disk's performance and reliability. Computerworld spoke with Hagberg about the future of solid state disk. The following are excerpts from that interview.
What is Fujitsu's position on solid-state disk? There is a place for flash. Right now, that's random-read performance, such as relational database look-ups, tables, etc. But the big question that's been posed to Fujitsu is why haven't we come out with a flash drive product, and what do we think about enterprise disk drives, because there are executives from companies in the Northeast claiming that enterprise disk drives will be dead within two years? I think that's definitely overstating the capability of solid state.
There's definitely a niche for flash, but how big is that niche and how quickly can solid-state disk manufacturers resolve the performance problems of solid state in sequential reads and writes as well as random writes is really a big gate to expand the market penetration of solid state.
What are the main performance problems with solid state? If you talk to some of the executives of notebook manufacturers, pretty much over the past year and a half, two years, every manufacturer has launched a notebook with a solid-state disk drive. And, almost universally, there has been a customer-satisfaction issue because they're hyped for performance, and people get them and don't realize what [manufacturers] mean is [NAND] flash is really good if you're reading stuff, but it doesn't work very well for large file reads and large file writes, and it doesn't work well for random writes.
Unfortunately, with disk drives, you're going to do a lot of reading and writing and not just random reads. If you look at the last couple of years, there have been a lot of attempts to improve I/O performance in notebooks. One of those has been the advent of hybrid drives, and a second is solid-state drives themselves. But we've talked to a lot of major end users as well as a lot of major manufacturers of notebooks, and the statement is that while these drives promise a lot, they don't deliver a lot outside of benchmarks in real world performance.
We have people in our facility here that have taken solid-state drives, put them in notebooks and found out it doesn't boot any faster. The power savings are negligible because the power drain in a notebook is the screen, the CPU and DRAM. When the hard disk drive is not being written to or accessed, it actually goes into sleep mode most of the time where you're down to below a tenth of a watt in terms of write. So when you weigh battery life of a notebook [with a 5-hour battery life], solid state may get you 5 hours and 15 minutes, but it's not more than 5% battery life improvement.
Then, when you look at the promise of instant boot, load Microsoft Windows, Linux or other things on your notebook with a solid-state disk drive versus a hard disk drive, and you'll find out there's not much difference in the boot time. And you want to talk about application load, I can click on the icon for PowerPoint on my system here and start a presentation, and in less than a second, my application is loaded. So there doesn't seem to be any improvement I can see with a solid state drive. So there's a lot of talk about the performance advantages and power savings that don't translate into real-world experience [for] notebook users and then if you look at sequential and random writes where hard disk drives win today. Then, you can then go to the next step -- reliability -- and question that.
Solid state still needs a year or two to prove itself because reliability is still in question, whereas with a disk drive, you can write to the same spot on the disk millions of times. With [single-layer cell] solid-state disk, you're limited to 100,000 writes as a spec across the industry, and with MLC [multilevel cell] solid-state disk, you may reduce it to one-tenth of that or less -- 10,000 writes per cell with two bits, or maybe even 1,000 writes per cell with three or four bits per cell.
So are you saying solid state is never going to be able to compete with hard disk drives? Solid-state drives are good in some narrow niche applications where you're focused on random reads. They're great for handhelds, cell phones, iPods, MP3 players -- where you enter your songs one time and listen to them thousands of times -- but when it comes to high-performance network servers where you're writing a lot of information on a continuous basis, you need a map that tells you where the data is ... or you'll wear out your flash very quickly.
With solid-state disk you have wear-leveling algorithms [which more evenly distribute data across memory] or you need to track how many times you've written to every cell. So in addition to writing data and keeping track of where the data is, you have to keep track of how many times you've written to a particular cell ... and then translate that to keeping spot space available to move cell data once you've hit a certain limit on writes.
Now, do I think solid-state disk will become a large portion of the storage solution over time? Yes. Do I think it's over the next two years? No. It's still 5% of the market over the next couple of years. Engineers will find ways to overcome write performance -- if you look at phase-change memory or MRAM, where you have magnetic recording memory that's attempting to overcome the write challenges of static storage.
Are you saying these things because Fujitsu is behind the curve on solid-state drives? The statements I'm making are Fujitsu's observations of the real-world product. If you talk to anyone using the products, you should find a consistent message with what I'm telling you. I'm not saying something to slander a technology because we don't have a device. That would be meaningless. If a technology performs better than I'm stating, then my opinion is mitigated.
At this point, Fujitsu has looked at a number of things, such as hybrid drives, where you put flash on a hard disk drive. For the last couple of years, that was promoted as a way to improve I/O and really to improve performance, and Fujitsu took a look at it, designed some products in the lab and checked performance. While some of our disk-drive competitors got very aggressive in hyping it, Fujitsu's internal testing showed negligible performance improvement. Even Microsoft Windows Vista, the OS supposedly geared for the hybrid or flash-enabled drives, didn't have any significant boot improvement. My notebook with a 28-second boot with a standard drive would go down to 21 or 22 seconds with a hybrid drive. Personally, I wouldn't pay for that, adding flash cost to a drive to see a 6-second improvement to the boot.
Two years ago, a lot of our competitors were hyping these drives, Fujitsu elected not to release the product we developed in the laboratory after benchmarking it because there's no value proposition for this drive.
Two years later, the market doesn't see the value in those competitor's hybrid drives.
When you look at solid state, there is tremendous hype. Talking to John Monroe of Gartner, he estimated 2007 shipments of solid-state notebooks were about 98,000. You're talking about a market where you have 120 million or 130 million notebooks shipping. And if you polled all those users, they probably wouldn't buy another one because most of them bought them based on performance statements that haven't translated into real-world performance.
So it's fair to say Fujitsu will wait at least two years before coming out with a line of solid-state disk drives -- other than NOR? We do believe that phase change or other types of memory probably will replace the current generations of solid-state disk drives in years to come. The question is, the companies making a lot of hype with storage and server vendors right now with the announcement of solid state in their systems, can they over come the performance issues of writing? A lot of them are composing combinations of DRAM with battery backup in combination with NAND to overcome the write problems. So I think there will be solutions.
We'll monitor it and look at it from both an internal development perspective as well as with partners that we work with in memory technology to determine the right time to enter this market, but as you said, it's probably a couple years away before there will be mainstream adoption of this technology.
At Fujitsu, we've been in the disk drive business since 1968 ... and we do our own heads and own media and have third-party agreements that also do heads and media as our second sources. From a value in terms of dollars per gigabyte, as well as the I/O read/write data rate, disk drives are still not threatened by solid state in the mainstream enterprise server market or in the notebook space. The vast majority -- 90%-plus -- of shipments over the next few years will still be hard disk drives.