Flashy Hard Drives

A boot buffer is a good thing to have to start feeding the CPU while the hard drive is slowly waking up, then pass along more boot stuff from the hard drive to complete the process. This at least theoretically ought to cut boot times quite a bit, which is one of the few hard-drive related activities the average person really notices.

After booting, the flash is supposed to act as a write-back cache, primarily to minimize hard drive writes, so the hard drive can sleep most of the time. Not much to write home about on the desktop side unless you hate hard drive noise, but that would be a real power-saver for notebooks.

What about reads? Well, this device is being branded as a Longhorn device, and MS says that Longhorn is going to try to cache data big-time to main memory to minimize reads, but this could be used for reads, too.

Issues?

This isn’t close to ready yet, maybe by the end of 2006. There’s basically two reasons for this:

Speed: Currently, flash memory transfers data much more slowly than hard drives. Microsoft/Samsung presumes that flash transfer speeds will be more like 100Mb/sec by the time they get this out the door. This will take some work.

Reliability Currently, the best NAND available is rated for 100,000 write/erase cycles. Even if the flash were only used for boot/write cycles, 100,000 cycles would be cutting it close for even Joe Sixpack-type activities. Assuming just 50 write/erase cycles a day (which would seem to be pretty close to a minimum for anybody, and probably is grossly optimistic), you have a drive that will normally start failing in about six years.

Yes, for current flash devices, there are such things as load-leveling firmware to increase MTBF, but those devices generally don’t fill up completely, then empty, which is what this would do.

One could cut down the number of cycles used to boot/write by increasing the size of the flash, of course (this device is only supposed to contain 128MB), but cost then becomes a factor.

I think MS/Samsung is going to have to get the NAND up to a minimum 250K cycles, if not 500K or 1M, before people are going to feel comfortable about using these drives, or increase the flash size to at least 512MB, if not 1GB.

Moving Away From The HD

What I think is important about this announcement is not the device, but the intention. The idea is to make the hard drive the last resort in any activity, use memory as the real working repository, and treat the HD as an archive.

For most of those reading this, what will probably have more impact on their computing lives in a few years will be Longhorn trying to cache everything in main memory. This will definitely increase the amount of RAM you buy. Unfortunately, though, you give away in the time required to cache a gigabyte or so in volatile memory what you gain in performance, and separately-powered solid-state hard drives are too impractical (and expensive) for the mainstream any time soon.

We shall see.

Ed