By looking at the the trend of annual improvement in the number of Megabytes purchasable per dollar for both hard disk and flash memory storage over the last two years and extrapolating these figures into the future brings out a rather surprising conclusion.

Flash memory will be cheaper per meg than hard disk storage in just 11 years.

Another way of saying this is that after 2017 there may not be any compelling reason to ever buy one of those mechanical rotating magnetic data storage devices ever again. They could just end up as one of those quaint curiosities of the late 20^th century like punch cards and paper tape. Sell your shares in Segate and Western Digital if you don't want an investment in companies whose main product will be the technological equivalent of the buggy whip in just over a decade.

The data behind this conclusion can be seen here for the hard disk component and here for the flash memory component. For those people too lazy to do all that extra reading, here is the summary:

Right now your dollar buys about 130 times more hard disk space than flash memory. In almost every year, you can buy more space for your dollar than you could last year. This improvement for hard disks in the last two years was measured at 44% per annum. The annualised improvement for flash storage over the same period was measured at 118%. By simply extrapolating these figures into the future until the megs per dollar figure for flash beats that of hard disks gives the date of 2017 or in just 11 years time.

Flash memory storage has several advantages over hard disk storage that would make it the preferred choice as soon as it becomes economical to do so. These advantages are much smaller size, lower power consumption, solid state toughness and fast random access speed.

There is a common misconception that because flash has (only) somewhere between 100 000 to 1 000 000 write cycles per block before failure that this eliminates them as a suitable media for long term storage or high I/O applications. Most people don't expect their hard disk to wear out just by using it and they will have a similar expectation of what ever replaces it. After all, hard disks are not generally considered to be consumables. A simple comparison with a hard disk of today shows this misconception to be just that. Taking a the example of a flash drive of 200GB with a write speed of 40 Megabytes per second and doing some basic calculations shows that it could be written to continuously for just over 15 years before every block passed the 100 000 write mark. The equivalent of todays 200GB drive some 15 years ago was the 210MB disk. There are not many machines running today with 210MB hard drives, let alone dong the kind of work that requires continuous writing to the disk. And 100 000 writes is often considered a minimum. Average failure figures are often quoted as 1 million writes. Astute readers will observe that it is possible to wear out an individual block much more quickly than 15 years with normal use as they know work is not evenly distributed across the whole disk. Some parts will contain intensive use files like the swap space that achieve high rates of use. The solution to this problem, wear leveling algorithms, already exist.

One area where hard disk performance clearly outstrips that of flash is in sustained read and write speed. In order to get some idea on how much improvement in flash needs to occur, the first thing to do is get some perspective on how far flash trails behind disk in todays performance stakes. Some quick research over at Tomshardware shows the fastest 3.5 inch IDE disk of today averages about 60 MegaBytes per second sustained throughput be it read or write. By comparison, the fastest flash drives achieve a very respectable 23 Megabytes per second write speed with 30MB per second read performance. This means flash will need to triple in speed in addition to keeping up with normal improves that will occur in performance to surpass hard drives in every way. A look back 5 years ago at the throughput of the fastest 3.5 Inch IDE hard disk shows drives have improved about 250% in read speed and 400% in write speed over the period. Finding useful historical performance data on USB flash drives is more difficult. This is because USB 2.0 wasn't introduced until late 2001 and without it, performance tests are fairly useless as performance is constrained by the bus and not the flash.. The other difficulty is that I found no test that claimed to review the fastest drive available at the time. This flash drive review at Ars Technica from 18 months ago shows the fastest tested drive to be only about 9 meg per second in read and write speed. If the fastest drive in that test was indicative of the fastest flash drive available at the time (and thats a big if) then performance has increased some 300% in just a year and a half. If this performance trend continues, it seems safe to assume that flash will catch up in this area and surpass hard disks way before 11 years.

This prediction is based on just a recent 2 year trend. Because we are some 8 months into the survey period that the data comes from, we can take a sneak peek at what the 3 year trend will look like to see if the trend is holding up OK. The improvement to date for flash is about 70% with 4 months still to go. The annualised improvement can be estimated to be around 90%. This would make this year a fairly bland year by flash standards but certainly not the worst nor anything out of the ordinary. To make a comparison to hard disks, the improvement in the year so far is less than 5% and can be projected to be about 5% for the whole year. This might be the worst recorded year in all 14 years of the survey for hard disks. The growth differential between the two then comes out at 85% in favor of flash, which indicates a nice continuation of the trend.

Matts Computer Trends