Introduction to hard drive technology

What actually makes a drive fast?

So then, NCQ doesn’t, a fast interface isn’t going to and low access times don’t; what actually makes one drive perform better than another? HDTach is a popular disk benchmark program that will give you a detailed analysis of how fast data can be sequentially read and written to a drive, along with the interface speed and access time, but low-level figures all have relatively little bearing on performance.

IOMeter is another popular disk benchmark program that will give you details results of how many random access requests can be serviced every second, but this is entirely useless for assessing desktop performance, as accesses aren’t random in a desktop system. So then, what program would one use to deduce a hard drive’s desktop performance? Unfortunately, the answer to this question is nothing at all. There is currently no benchmarking suite available that will give an accurate quantitative representation of a hard drive’s desktop performance.

The part of a hard drive which has the most dramatic effect on its desktop performance is unfortunately completely unquantifiable. Firmware differences can mean that two otherwise identical drives can differ vastly in performance. Current generation desktop hard drives all have the same interface, the same rotational speed and similar data densities, giving similar low-level performance characteristics like access time and sustained transfer rates, but performance varies quite a bit from drive to drive.

Read-ahead algorithms employed by a drive predict, given the piece of data just accessed, what data is likely to be accessed next. The drive then reads this data into its onboard cache. If the system then makes a request to the drive for data that was cached by the read-ahead algorithms, it can read this data directly from cache. This is far faster than reading the data from the disk and can speed up things dramatically, making a drive ‘feel’ much faster.

There are, of course, other factors that influence a drive’s performance. The rotational speed, sustained transfer rates and access times do have some bearing on performance, but firmware differences can more than outweigh any of these. For instance, despite having a slower access time, slower sequential reads and writes and a slower interface, the 10k SATA Raptor outperforms current generation 15k SCSI drives in desktop applications.

What could the future hold?

It is quite an exciting time in the world of storage, that is, if you can call storage exciting at all. Hard drives of the future will still be plain and boring boxes, but we’ve already covered that. Just last week, Samsung shipped the first hybrid hard drives.

While they’re only available in the 2.5” form factor, and topping out at a relatively unremarkable 160GB of mechanical storage and 256MB of flash memory onboard, this is simply a taste of things to come. I expect we will begin to see the amount of on-board flash memory on drives increase exponentially going forward, along with moving over to the 3.5” form factor and the terabyte capacity it will soon offer.

Flash memory is indeed the future, but I don’t think mechanical storage is disappearing any time soon. It’s not an unpopular belief that mechanical storage, i.e. the magnetic platters spinning at high speed that have been the staple of mass storage for over twenty years, is on its way out in favour of faster solid state storage. I don’t see this happening in anything but the quite distant future. The cost per GB of solid state storage is many times higher than mechanical storage and it won’t be feasible for affordable mass storage for quite some time.

High-speed storage is just unnecessary for the digital media that is chewing up the masses of space on many of our hard drives. Of course, progress with mechanical storage will begin to slow in terms of capacity, and eventually hit a wall where current designs just can’t get any bigger. Progress in flash memory is moving at quite a rate, will continue to do so, and will eventually start to compare to mechanical storage in terms of price per GB, but we have a while to wait before its feasible to buy a terabyte of flash-based storage without selling off siblings and/or body parts on eBay.

Solid state storage has only recently reached a price point where small drives are economically feasible (though still very expensive). Their low-power characteristics mean that small flash memory drives are already being used in higher-end ultra-portable notebooks. Before long, flash memory will reach a price point where one could have a small system drive for applications and the operating system while leaving the mass storage to a mechanical drive. This idea of tiered storage is nothing new, many enthusiasts today run their OS and applications on a high speed, but low capacity 10,000 RPM drive while throwing hundreds of gigabytes of ‘stuff’ onto a larger, but slower 7,200 RPM drive.

Hybrid technology will eventually move across to desktop drives at which point we will all be able to enjoy lighting fast boot times and that ever-pursued snappiness on the desktop. Flash memory storage can easily be integrated on-board a high-capacity traditional mechanical drive. This solid-state storage, say around 16-32GB (for now), can be used for the things on your system that would benefit most from fast storage.

All of your ‘stuff’ can be thrown onto the mechanical storage of the drive. Any read or write operations to the flash memory section of the drive would be completely transparent, as if you were reading and writing to RAM, dramatically improving the general ‘feel’ of your system. Put simply, flash memory is one of the biggest things to happen to storage for a long time.