Instant Access Memory

tnielson writes: "The April issue of Wired interviews Stuart Parkin, an IBM scientist developing MRAM; Non-volatile, fast, durable, and cheap. It should be great in an MP3 player, and according to the article, could make all of our computers instant-on! Problem is, five years is a long time to wait..."

Comments

Instant On isn't accurate.

[Effugas]

This is the guy who helped come up with GMR? I bow down to his technical skills. But it seems that this technology is being sold for something that it just isn't.

Really, this just doesn't have much to do with instant on technology.

It's true. As useful as it is to require no power to store a charge, neither desktops nor servers have any serious problem with power--they're both plugged into a wall! There's no reason for mature DRAM memory to not receive the trickle charge it requires to keep its contents from drifting away. Problems come when operating systems (primarily) and motherboard standards fail to build in stasis modes--for all the determinism of computers, I find it rather surprising that the entire system cannot be simultaneously frozen until a given restart interrupt is triggered. But that's the situation we face--it's not that the memory doesn't last, it's that we don't know how to deal with a house of cards we don't need to rebuild every so often.

Where I see this technology being useful is in laptops, or anything else where "power just to suspend" is a real issue. Heck, even for normal operation, memory can be a real drain on power: Witness the effect of increasing from 2 to 8 MB of RAM on a Palm V(it's significant!). So this does matter for pervasive computing, as the article suggests.

But it has almost nothing to do with "instant on". I do forsee it being implemented in systems which don't want to have to "recover state from hard drive" or "implement a trickle charge system to keep existing state", but that's not so much a break through. The reduced power load scene DOES seem interesting, but lets not forget just how mature a technlogy DRAM is. They'll have to do some pretty amazing work with the MRAM to surpass DRAM. By then, where will DRAM be? Remember, Intel has its dominance partly out of the sheer amount of resources they can put into making the horrifically complex x86 fast. 21bil is alot of money to lose to MRAM!

Thoughts?

Yours Truly,

Dan Kaminsky
DoxPara Research
http://www.doxpara.com

Re:I'll believe it when I see it

[troc]

Very true - what tends to happen is that in 5 years time, another technology, or a development of an existing technology has rendered the "amazing new breakthrough" obsolete (or too expensive, complex etc)

These things tend to resurface a few years after they were invented or whatever and become part of technology anyway but without the fanfare and with a few people going "I told you so" and claiming 20/20 hindsight :)

Things like holographic and/or 3D memory storage - was firsat mentioned well over a decade ago as the "next great thing" and was promptly forgotted and has recently resurfaced as working prototypes etc.

Best thing to do is take it with a pinch of salt and wait a few years.

I also like the fact that you can read the submission at the top of the page as indicating an awesome instant-on device that takes 5 years to power up :)

Hohum

troc

You don't get it

[Graymalkin]

Linux weiners tout never having to reboot a system because their OS is oh so stable. Big efing deal. My file server's been up for a week and will probably stay up until I decide I want to upgrade the kernel. This is totally beyond the point of instant-on computers. It would be rad to have a computer with 10 gigs of ram that all your stuff was on, that way you'd have little seek time and could turn the system on and off like a television. Why is this cool? Because some people don't like leaving computers on all the time, in many places eletricity can be pretty pricy at certain times of the year or they may just not want to have their electrical toys running 24/7. Instant on would be great in a corporate environment because you'd get to save time waiting for stuff to start, that time builds up in the course of a year costing you cash. Home users also wouldn't have to deal with boot-ups which they may or may not understand. Oh well.

Persistent operation systems

[srn_test]

This sort of thing is what the persistent operating system groups have been working on doing for years.

It turns out that it's _hard_ to do - keeping the data around is the easy part; what do you do when the OS crashes? How do you recover?

You end up with a huge database like wrapper around the entire OS, and really heavy-weight recovery code to try to rebuild a consistent state of the system.

You've also got the problem that if something is wrong in the OS, when you reboot you'll quite possibly just trigger the same bug again! Makes Microsoft style "reboot to fix the problem" solutions not so good.

See some persistent OS sites, like:

• http://www.psrg.cs.usyd.edu.au/
• http://www.cs.stir.ac.uk/~aol/publicationlist.ht ml

This is just a few I happen to know.

Stephen

Irrelivant

[CAIMLAS]

Would an instant on computer not make computer pogram stability a near-useless feature? For me, loss of work isn't my fear through a crash - it's sitting their on my white skinny butt for X minutes while my machine reboots.

I switched to linux to prevent that. (And to geek around more, but that's another story.) Would such a thing make Linux's main strong point null, or would linux be able to develop it's other fields - digital image/video editing, audio, games, workstation software - in time to surpass wintel products, on a quality based assessment alone?

-------
CAIMLAS

Instant on == instant problems

[wowbagger]

I see two different but related problems with a system composed entirely of non-volatile memory:

1. If you pull the plug, the state of memory is retained, but the CPU state is lost. This is very much akin to an interrupt occurring, except that an interrupt at least records the CPU state on the stack. The problem is that you now have to protect everything from re-entrancy problems: otherwise when the system is abruptly powered off and restarted the CPU has to do a restart and system or application data structures may be in a non-reenterable state.
2. If the memory is corrupted, how do you force the system to clear? You'd need a [button|keysequence|etc.] that would tell the system to do a complete coldstart & purge of memory.

Also, this could be bad if the Men In Black (and I don't mean division 6) kick in your door. Anything in system memory will be in system memory "and may be used against you in court", whether you like it or not. You won't be able to just yank the plug and clear system memory.

That said, I still think this will be wonderful in the main. It's just going to have some implications we'll need to think about.

No...

[Greyfox]

If your OS crashes or becomes unstable you'll probably have to jump through some hoops to zero out all your RAM. So it'll only be instant-on if your system remains stable. This should increase, not decrease, the demand for a stable OS. One which doesn't allow programs to take the OS with it.

In five years..

[Lonesmurf]

Apparently, in five years, we will have 1 Terabyte solid-state harddrives, instant memory (what? 10ns is too long??), 1 bazillion GHz processor rings, and video cards that will spit out realtime images straight into our fucking brains, all running on 1 Petabyte networks.. within our homes. Oh ya, I may have forgotten to mention that all this will be free.. all supported by a little blinking banner on your desktop that you will mentally block out after a week of using the machine.

Then again, the way software is moving, I may need this to play Quake |||(|)||| on my BloatedLinux(tm) ver.100.3.2 system.

I'll believe this stuff when I see it.

Rami James
Pixel Pusher
ALST R&D Center, IL
--

Re:Bubble Memory ?

[EntropyMechanic]

The problem with bubble memory was that it was serial. All the bits were stored as little magnetic domains, but the only way to read those domains were to actually shift them to a particular part of the IC where special circutry existed to read them. The domain storage was logically laid out linearly, and as I recall, the domains could only be shifted in 1 direction due to the physical layout of the IC used to hold them. Thus, if you read a byte and then tried to read it again, all the domains that composed the bits in that byte had to be shifted around the entire domain storage array again.

Now while a lot could have been done with caching and using multiple domain storage arrays, bubble memories were serial devices and their latencies just would not scale up well as you added more bits to them. Bubbles would make a good NV storage device, but could never replace RAM.

Bubble memories were introduced in the late 70's, I believe. I think their big failure was lack of storage space and speed. Their commercial death knell was the ramp-up of HDD storage capacities in the mid-80's. They did have the benefit of having no moving parts and I think a military hardened version was available. If they exist at all any more, I'm sure it's just in a few niches.

JTS

Baldric, you wouldn't know a cunning plan if it painted itself purple and danced around on a harpsichord singing 'cunning plans are here again' - Lord Edmund Blackadder

Doesn't make any sense

[muzzy]

... could make all of our computers instant-on! Problem is, 5 years is a long time to wait...

I don't think 5 years is really "instant-on", this story is contradicting itself.

Looking through the archives . . .

[streetlawyer]

Apparently, in five years, we will have multi-gigabyte hard disk drives, a global network of computers, we'll be able to transmit 58.8Kb over voice telephone netowkrs, wireless data networks and x86 chips running at 300MHz will be cheap. Yeah, I'll believe it when I see it [1995]

Apparently, in five years, we'll all have Xerox PARC style desktop environments, hard disk size will be so big we'll be able to forget about our archive of floppies and we'll have moving pictures on our PCs. Yeah, I'll believe it when I see it. [1990]

Apparently, in five years, we'll all have affordable IBM computers with hard disk drives in our homes. And we'll all be walking round with mobile telephones. Yeah, I'll believe it when I see it. [1985]

Apparently in five years, we'll all have over 512K of RAM and we'll be able to do graphics on desktop computers. {Note: I remember hearing someone around this time talk about a "gigabyte" as if it were an obviously made-up word or at best, a whimsical extension of "kilobyte"}. Yeah, I'll believe it when I see it. [1980]

[....]

"I can see a global market for maybe five computers"