Merging Memory and Storage into a Single Unit?

I have always wondered why computers have two separate units for permanent data storage and temporary data access, as the time that it takes for data to be moved from permanent storage is one of the greatest delays in the functioning of modern computers, and has been for the two to three decades since their emergence.

With solid-state drives now becoming reliable and affordable, the delay between storage and memory has greatly decreased, but it seems to me that it would still be more efficient to have a single unit function as both permanent storage and temporary memory, which would allow for nearly-instantaneous access of the data that it contained.

Of course, I understand that the reason for which memory and storage are still handled by separate devices is that there is no inexpensive and efficient manner for developing a device that can perform both functions, but my questions concern a time when such a solution has been realized:

First, what numerical specifications shall be important when measuring the performance of a unit that functions as both memory and storage? I imagine that the important specifications would be the unit's storage capacity and its operating frequency/data transfer rate, which are also the important specifications for both current memory and storage devices, as well.

Second, what type of interface would such a unit use? I imagine that it would be either the PCI-express interface, or a new interface that combined PCIe with the current DRAM interface.

Third, when may such a technology become feasible, if ever? I certainly hope that that task can be accomplished in the near future.

What does everyone else say on this subject? I await your responses.

RAMdisk?

Put the OS on the permanent storage, then run a bootloader to dump it onto RAM. The only time it writes back to the SSD is on shutdown.

Yes, you run the risk of losing important data, but it'd be fast.

Cost has been the main reason historically. I think that we will see that change in our lifetime, but at the same time the interface between memory and the cpu will always be faster than the interface between the cpu and storage, even if the storage controller is moved to the cpu like the memory controller is.

System memory is usually limited (I think some servers allow 380+ GB system memory on one board), but with technologies like NUMA and high speed interconnects, system memory can be huge, but you still lose the speed when addressing other nodes.

Many years ago (1986 ish) Storage Technologies had a product based on memory that acted like a hard drive, similar to SSD technologies today, but the cost to keep the memory contents valid was not worth it most companies.

There was also a motherboard years ago that had a powered memory section, but it had high failure rates and I don't even think it made it to production.

There are also systems like the Texas Memory Systems cards that use the PCIe bus to avoid the storage controllers and increase speed. Maybe one day you will be able to plug a storage device into a DIMM slot....

http://www.engadget.com/2012/06/14/c...nd-reram-unit/

These :)

A hybrid SSD that is 11 times faster than the average SSD and is 93% less power consuming that lasts 7 times longer than pure NAND products.

ReRAM is expensive but at the benefit of the increased life and low power consumption it is likely that it will make it's way into servers soon enough and then eventually into your average desktop with some luck.

I've seen the new SSD cards that fit into a PCIe slot. Anybody have experience with them? How are they for speed and setup?