matsu

memory vs. storage architecture

The clear distinction between memory (RAM) and storage (HDDs) has suffered an evolution in the past few years, in part because of the technologies, but also because of the way the operating systems work. The current state of affairs goes something like this:

• Cache memory – a faster copy of parts of the RAM of the system, keeping the most used instructions and data “nearer” (in terms of access speed) to the processor
• Main RAM – memory where the whole current state of the machine is kept and manipulated
• Video RAM – memory where the image display state resides (which might reside in the main memory in terms of hardware)
• Virtual Memory – the opposite concept of cache memory – the operating system moves less used data into the slower non-RAM storage system, to free the main RAM memory for the most used data.
• Storage – bigger, slower, non-volatile memory where the non-working state of applications and data is stored, even when the computer is turned off, which is divided into three main technologies:
  • Hard disk drives – magnetic media on spinning platters – cheaper than RAM, more compact by capacity, much slower, less reliable (to impact and wear and tear), but non-volatile
  • Flash memory – solid state memory (like RAM but non-volatile), more expensive but faster than HDDs*, more reliable, some limitations on the total number write cycles possible, the best energy efficiency of all the available storage technologies
  • Optical storage – CDs, DVDs, much cheaper than any other kind of storage, not very energy efficient, slow

Although there has been a constant evolution of all these technologies, the evolution of price/size/speed of flash storage has brought it to the point where HDDs can be replaced by flash storage, although still at smaller capacities and much higher prices. But the trend continues to march on, and based on that evolution I think a possible evolution of the state of the art in memory vs. storage architecture design could go in the following direction, both for mobile and non-mobile applications (with the tweaking of the wanted ratio between capacity/performance on one side versus the need for portability/energy efficiency on the other):

• processor cache memories, main, and video RAM as they are today
• virtual memory to be ran on slower but cheaper RAM (although faster than storage of any kind), at the same price point as flash storage, but volatile – the virtual memory contents could be copied into storage on suspend/hibernation states – which would in part reduce the problem of the potentially limited number of write cycles on flash storage
• HDD spinning platter storage to be replaced by flash memory, in two stages:
  • at first, there could be an hybrid system, where the main OS and applications would work from a flash storage based partition, and the main data storage would still be on traditional HDDs (there are hybrid drives today, but I propose a less integrated and less OS dependent technology)
  • the second stage would be when the price per MB of flash storage achieved a level compatible with the needs of users (to be achieved for mobile applications first), and traditional HDDs would be phased out almost completely
• Same line of thought for optical media vs. flash storage, or the appearance of large capacity solid state ROM storage (although my opinion is that the WAN or LAN online transmission of free or paid data will replace the distribution by physical media completely, eventually)

To give a real life example, the default partition arrangement of general usage Linux installs (boot/swap/data) would be based on three different technologies – flash storage for the boot partition, cheaper RAM for the swap (virtual memory) partition, and traditional HDD technology for the data partition, until flash technology is able to overcome its limitations both in terms of price and no limitation on write cycles, and replace the other two technologies.

* in some applications fast HDDs in sequential data read/write operations can be faster than flash storage