Of Disk and Memory

February 24, 2006

Of Disk and Memory

Of Disk and Memory
By Robin Bloor

A major change in computer technology will happen in the next 5 years, driven by the replacement of disk by memory. The shift has already begun – it began last year – and it will have wide-reaching effects as it progresses.

Let?s take a look at some statistics. Currently the retail cost of memory is $60-$70 per gigabyte (GB). By retail cost, I mean the cost if you buy a single memory stick from your local computer store. The retail cost of disk is about 60-70 cents per GB (for a 500 GB drive). So disk storage is about 100 times less expensive.  The cost of both currently falls at roughly the same rate, with disk costs reducing at about 40% per year and memory reducing at about 43% per year.

Given these stats you might conclude that the role of disk and memory in deploying computer technology will remain the same in the near future. But something disruptive happened last year with Apple?s introduction of the iPod Nano. A device that had been disk-based (4 gigabyte iPod) now became memory-based. Memory banished disk in a mass-produced ?computer? product. There was a crossover of usage.

This is significant because it introduced a truly disruptive change to the whole of computing. First of all, it altered perceptions. There will now be a ?march of memory? upwards as 8GB, then 16GB, then 32GB devices become all-memory devices. It won?t be long before laptops are all-memory devices. Because memory, with no moving parts, is more robust than disk, there is a good engineering motive to move to memory ? especially in the consumer market where robustness has the virtue of reducing support costs.

The market for memory will expand as a consequence and this will make the price of memory fall at a faster rate. If you double the size of a market you reduce prices by 30% through natural economies of scale (as a general rule of thumb). Right now there is about 100 times as much disk storage space sold as there is memory (this is my estimate, based on an aggregation of stats so it might be a little inaccurate, but not by orders of magnitude). So if memory gradually evicts disk this added factor will come into play.

What does this mean?  Well if you multiply the size of the memory market by 100, the cost is likely to become $6-$7 per GB ? now only 10 times the cost of disk ? or to put it another way, if the two markets were the same size the cost difference would only be a factor of 10. As the disk market ceases to grow ? and it will at some point ? it will begin to lose some of its economies of scale and then, it?s game over.

It is not a simple matter to predict the consequential effects of this, but we can at least give you a view of the level of disruption. Ever since the advent of magnetic disks, nearly 50 years ago, software has been constructed on the basis of the need to get information from, and write it to, a ?very slow device?. By ?very slow? I mean that if you need to get information from disk then the retrieval is from 100 to 10,000 times slower than if you are accessing information direct from memory. Retrieving information from memory sticks, by the way, is not much faster than disk, because they are implemented as disks (with a device driver, directories, files, etc.).

However, information in memory is accessed in software by direct (or relative) address. It?s pretty much instant. With 64-bit chips and 64-bit software, the largest possible address space that a program could address is 4 exabytes ? which is about a quarter of all the information that exists on the planet. Nobody is going to build a computer that big (at least not this year), but it is possible to conceive of all-memory computers with capacity in the terabyte range. They?ll be here soon.

So here?s the disruptive part: Just about all the software that has been built in the whole history of computing was built according to an architecture that assumed the existence and necessity of disk store. As computers become memory-based all this software will need to be rewritten or transformed to work in what is actually a simpler environment ? but also quite a different one. Operating systems will work in a different way. Specific products ? particularly databases and transaction engines will take on entirely new roles or disappear altogether (either is possible). The very nature of software will change.    It happens in the next 5 years.



Newsletters 2006
About Robin Bloor

Leave a Reply

Your email address will not be published.