Saturday, December 16, 2006

Flash HDD

A lot of noise is being made these days about Flash chips and their potential use as system hard drives. I thought I would write about some of my impressions on the subject.

The first issue is write cycles. In the past Flash media was good for less than a million writes or so. This was completely unsatisfactory for most uses as an operating system media because systems generally use the hard drive as a swap, or short term place to store programs and data when they're not busy. In an active system the swap memory can be rewritten thousands of times a day, as programs are swapped into and out of memory very actively.

There are operating systems that don't function in this way. Certainly this process was necessary when computers had little memory, but with cheap PCs able to handle eight or sixteen gigabytes of memory, but it doesn't seem to be as necessary as it once was. Choosing to avoid the issue in this way does limit your choice of OS, but not horribly so.

Recent advances in flash memory have extended the life of the flash memory into hundreds of millions, or more, of write cycles. If you're willing to accept a lifespan of five to ten years for your flash memory, you should be fine with what's available now. Certainly we can expect this trend to continue. Flash will get more durable life cycles until they finally are good long after one would normally consider them obsolete.

Currently available flash media (December 2006) comes in sizes up to 32GB for Secure Digital media. That's a lot of memory for a card that small, and it comes at a premium price. When I first bought a flash drive the largest available size was 64MB, and it was only a few years ago. Since sizes have increased five hundred times in just a few years, and manufacturers are even now working on several generations of denser media. Since a large operating system install should be no more than 8GB currently, and that size is available, Flash Media has cleared the hurdle of being large enough to handle the job. Although larger form factors are available now for 32GB media, the cost of the flash chips is enough to prevent a large market for the devices and so they're not yet common except for those who have no budget constraints.

Speed is another issue when considering Flash media for your system drive. Although Flash currently can be much slower than HDD media, that is changing as ever more chips are added and accessed in parallel. Already you can get media that reads and writes faster than an ATA HDD. Soon they will be much faster. More importantly, since Flash has no moving parts there is no latency to speak of and every file is as close as any other. This simplifies much of the disk access process and makes file access much faster. Speed is about to cease to be an issue for Flash media, and already for most uses it's faster than a hard drive. With Flash a cache might be necessary for some applications, but different from HDD media, the amount of time it takes to flush the write cache is predictable and controllable from the system rather than the controller or on-drive electronics. This makes shutdown issues go away almost entirely.

Power issues are important for storage media -- not only for battery life but also for heat. Here Flash has long been a big winner for cameras, smart phones and PDA's. Because Flash is a static medium, no energy is required to maintain the data stored within or make it accessible. There is no spin up time, no idle power at all. If you're not writing to or reading from it, it uses no power. Naturally devices that use no power generate no heat. Even in their most power hungry use, writing, Flash media doesn't take as much energy as HDD media at idle. The heat issue is an important one because the more thermal energy a device dissipates, the larger it must be to passively cool, and active cooling adds energy costs, size and noise as well.

Flash media has no moving parts. It is utterly silent. This by itself makes it a preferred medium for applications like fanless computers in audio recording environments and low energy entertainment center platforms where even small amounts of fan noise are unacceptable.

There are already many distributions of Linux that can be installed to Flash media. Soon this will be a standard install option across nearly all distributions. Installing to Flash media can be very handy for workstations. Rather than evolve a fancy network system for maintaining each user's settings in a portable way -- that doesn't work if the network is down or the user is offsite, the user can just extract his boot chip and take it with him. Then he can arrive with his full toolset and get to work without worrying about which applications don't install correctly at the new station or aren't integrated properly or don't have his preferred settings.

Price is often an issue. Flash media costs many times what Hard Drives cost for the same amount of storage. Today I can buy a 2GB SD media for about $30.00. It would take 250 of those to make the volume of a 500GB HDD that costs $150.00. For this article however I'm not talking about a volume for storing your media or your database. 8GB should be plenty right now for an operating system and suite of applications for normal use, and that can be had for a reasonable price so I think Flash has cleared the affordability constraint as an operating system install medium.

For these reasons I think we're approaching the day when a Flash Drive is a part of the motherboard on desktop and server systems for the purpose of OS and applications install at least as an option. It seems likely to become the standard for media center PCs as well. I also expect to see more support for installing to this sort of media. It seems reasonable that advances in Flash media will continue to outpace progress in other areas of information technology. In addition for highly portable devices like smartphones richer suites of applications should soon be available to exploit the advantages of having larger static storage available.

Some consolidation in the Flash memory industry seems likely as well. Highly competetive markets like this one erode profit margins and depreciate inventory disruptively. Obviously in this environment purchasing a competitor can be cheaper and more effective in the long term than inventing a new process that increases the storage density or speed of your product line.

No comments: