Well, to put it quite blatantly, computer memory has literally become cheaper than dirt. A pound of sand or mulch, costs roughly a couple of cents (depending on where you shop). A byte of memory, today, costs one-hundred-thousandsth of a penny. Which means a MiB (2^20 bytes) of memory costs just fractions more than a penny (USD) – about $0.0122 to give you an idea. So we can literally compare 2 MiB of memory to a pound of dirt (as in they have roughly the same market value for consumers).
Historically, memory was quite an expensive investment during the early days of computing. According to John C. McCallum we can see that it would have cost Millions of dollars to produce just megabytes of memory in the 1950’s (back then just one byte of memory could cost nearly $400). Today you can buy 8 gigabytes of memory at NewEgg.com for just $99.99, and this is why – the way we use computers today has changed so drastically.
Before the Internet and the commercialization of the world-wide-web we were quite comfortable storing all data locally and relying on our PCs for memory and I/O quite heavily. Today, because memory is so cheap and bandwidth costs have dropped substantially, we can store our files online and easily reproduce them on any computer at any time. The need to rely on memory is gradually decreasing as HTML5 introduces more innovative ways than ever to build online applications in ways the web was never meant for. This may not seem like we would be relying on memory less and less as more and more people are purchasing PCs, laptops, net books, and even smart phones with more and more memory today, but let’s not forget how much value we actually put on that memory today vs. 50 years ago. We value the worth of a few megabytes of memory just about as much as we do a pound of dirt. It can equally be argued, of course, that while we value greater memory capacities at a much lower price, we also rely on more memory today than ever before. The most significant difference is in what we actually use such memory for and to what gain. The heaviest consumption of consumer-grade memory is normally indicative of high resolution graphics; such as for games, digital photos, video and other media-driven desktop applications. The most common memory-hungry commercial use seems to reside in database-driven commercial applications and SaaS platforms. Even though, it is quite possible for us to do everything our day-to-day needs depend on – such as checking email, viewing and editing documents, spreadsheets, presentations, etc… – with very little memory requirements. Once we remove the bloat of eye-candy from Windows 7, the gamers-extraordinaire of Xbox 360 and PS3, and the leaky FireFox browser – we are left with very little need for memory over the long run. Lighter, faster and more portable software applications are becoming the norm for many users. Perhaps, another indication that we are steering more in the direction of SaaS and relying less and less on the SWS model.
Things like Kryder’s Law, which states that magnetic disk space doubles annually, or Moore’s Law – that depicts computing power doubling every 18 months, have demonstrated just how fast technology has been expanding over the past few decades. So by the year 2020 we will be able to purchase a Terabyte of memory for the same price we can buy 8 gigs today. Imagine how large our hard-drives will be if we can have a TiB of RAM on one computer! Today even 3TB HDDs don’t seem to be enough for some users. I know people that use several 3TB disks and still resort to external storage devices such as the cheaper USB mass storage drives or online storage space. This only demonstrates that we are becoming more and more comfortable with storing our data online (or on “the cloud” as the term has been coined). It’s not only cheaper for large corporations to purchase such massive storage space and make it readily available to the public for free (or at a small price), but it’s also safer and more convenient because they can keep backups a lot better than we can. Also, they can keep a lot better up-time than we can when they’ve learned to manage multiple state-of-the-art data-centers with exuberant redundancy measures.
The point is, we will likely still want to keep high volumes of memory locally to allow us to run graphics-savvy games and memory-hungry applications for very specific uses, but the general public will likely have little use for large amounts of memory at one time and will likely resort to more I/O dependency as I/O technology improves and provides faster reads than ever before. This also takes into account how much processing technology is developing with GPUs – for faster graphics rendering – and even more shared cache.