Princeton: Replacing RAM with Flash Can Save Massive Power

[palladin9479]

Their talking about energy cost per megabyte of memory which is orders of magnitude better for flash then DRAM. A high end server will come with 192 to 256GB of DRAM, you could add another 1 to 2TB of "flash memory" to that for significantly cheaper then what it would cost for another 1 to 2TB of DRAM (if the system could even handle that many chips on it's memory bus). The tier II memory wouldn't be nearly as fast as DRM but it would be much faster then SSD swap space (I/O and protocol limits) or even regular HDD.

See nothing can be run out of swap space, swap is only used for non-active process's, if a process needs to read something from memory it must first page it from the disk into RAM while paging something from RAM to disk. This is talking about allowing programs to read direction from the flash storage as though it was memory and not dealing with the swap system on the OS. The memory management unit would have to realize that there is a difference between DRAM and Flash RAM (FRAM?) and allocate it accordingly to priority.

This is useless tech for consumers but very interesting for hosting virtual machines where your limit quickly becomes memory and I/O not processor cycles.

 

[supere989]

LOL@Pagefile to SSD.......how about Pagefile to RAMDISK.....PWNED GGKTHXBAI

But seriously, When is Microsoft going to look at how much ram we have installed and say...oh....well okay, so with your 32GB of Ram, maybe we should just turn off the pagefile.....(Yes i know we can manually but windows pitches a bitchfit with certain functions)

So to address this Situation, I have Allocated 8GB of Ram to Pagefile functions via a RAMDISK, and Moved all my Temp Data functions to a second 8GB RAMDISK, Including Temp Internet Files.

My System performance is stellar. I don't have to wait for anything accept for slow web servers.

 

[blazorthon]

[citation][nom]supere989[/nom]LOL@Pagefile to SSD.......how about Pagefile to RAMDISK.....PWNED GGKTHXBAIBut seriously, When is Microsoft going to look at how much ram we have installed and say...oh....well okay, so with your 32GB of Ram, maybe we should just turn off the pagefile.....(Yes i know we can manually but windows pitches a bitchfit with certain functions)So to address this Situation, I have Allocated 8GB of Ram to Pagefile functions via a RAMDISK, and Moved all my Temp Data functions to a second 8GB RAMDISK, Including Temp Internet Files.My System performance is stellar. I don't have to wait for anything accept for slow web servers.[/citation]

This technology isn't about paging files at all.

 

[chazzeromus]

[citation][nom]blazorthon[/nom]This technology isn't about paging files at all.[/citation]
Actually it does, their method for using a log-structured virtual address space extension for allocated object via malloc calls also incorporates page fault interception to find the appropriate areas in the SSDs to materialize and fetch objects into DRAM. What supere999 said is just stupid, everyone knows using RAMDISKs pose latency issues where they are incompetent in applications, and also what he says makes no sense.

 

[blazorthon]

[citation][nom]chazzeromus[/nom]Actually it does, their method for using a log-structured virtual address space extension for allocated object via malloc calls also incorporates page fault interception to find the appropriate areas in the SSDs to materialize and fetch objects into DRAM. What supere999 said is just stupid, everyone knows using RAMDISKs pose latency issues where they are incompetent in applications, and also what he says makes no sense.[/citation]

My bad, it's still not the same paging file system as used by Windows on hard drives, is it?

 

[chazzeromus]

[citation][nom]blazorthon[/nom]My bad, it's still not the same paging file system as used by Windows on hard drives, is it?[/citation]
Actually, I believe you're right, it has nothing to do with the page "file", but uses the CPUs paging mechanism on top of the page file or swap space (system paging on disk unrelated since SSDAlloc has it's own page manager). But I think the technical papers used some form of linux, so there isn't a page "file" that exists in a filesystem but a page partition.

 

[bctande1]

I think focusing on making more efficient RAM. like Samsung's latest 30nm offerings, will yield must more consistent results, far more than this questionable method of progression.

 

[bernardblack]

This would be a good place to store windows paging files tough...

 

[Guest]

"is this what's employed in ASUS's V series, the maximus V gene, maximus V formula, & maximus V extreme with the special m/b header at the top left? that is what the sales people at microcenter are pedaling it as despite my thought it was a SSD from what ASUS reps made it sound like.[/quote

No, that is mPCIe, Mini-PCIe. It's for an addon card for rear IO panel WiFi, vs using up a more valuable PCIe x1 expansion slot.

 

[eralpb]

This will be near to useless in the coming years when SSDs become standart. Since when your ram is not enough, fall-back mechanism can use a partition in SSD which is very fast too!

 

[blazorthon]

[citation][nom]eralpb[/nom]This will be near to useless in the coming years when SSDs become standart. Since when your ram is not enough, fall-back mechanism can use a partition in SSD which is very fast too![/citation]

This is more than just merely having an SSD. It's a different way of letting the CPU use it. This would be even more useful when SSDs are more common. Why would a technology that relies on SSDs be useless when SSDs are more common? Your claim makes no sense.

 

[Guest]

Princeton came up with the cold boot attack. Freezing RAM with liquid nitrogen can make it non-volatile for a short period of time.

This method lets them do away with the liquid nitrogen altogether, so they can do forensic analysis on your working memory to get things like cryptographic keys.

 

[blazorthon]

[citation][nom]booberry[/nom]Princeton came up with the cold boot attack. Freezing RAM with liquid nitrogen can make it non-volatile for a short period of time.This method lets them do away with the liquid nitrogen altogether, so they can do forensic analysis on your working memory to get things like cryptographic keys.[/citation]

Some guy also came up with a way to make printers overheat and fail. That doesn't mean that he's going to attack your printer(s). You're assuming that just because someone at Princeton came up with a cold-boot attack (I'm not sure about that, but you might be right in that Princeton was where it was first figured out) and this new way to complement RAM with NAND, that means that they want to use this to attack you. You do realize that your idea relies on someone having physical access to your computer, right? That's a very unlikely scenario in almost any situation.

 

[Guest]

Governments seize computers routinely. just look at what happened to Jacob Appelbaum, who coincidentally was on the Princeton team that published the cold boot attack.

 

[blazorthon]

[citation][nom]hngcnkugtvkng[/nom]Governments seize computers routinely. just look at what happened to Jacob Appelbaum, who coincidentally was on the Princeton team that published the cold boot attack.[/citation]

So set them up to be able to erase any data that you don't want someone else to have in some way, maybe by remote or something automatically whenever they put power through the storage array. That's just a start to what could be done in the event of your computer(s) being seized.

 

[mrbeanladen]

[citation][nom]sylvez[/nom]Princeton: Replacing Computer with Typewriter Can Save Massive Power[/citation]
I'm not sure, you probably use more energy (your body) by pushing those keys down, I've typed on a typewriter and it isn't as smooth and easy to press as a keyboard, you actually have to put a tiny bit of force, and considering a human body can consume hundreds of watts and a midrange computer has 500w in full load, and like 100w in idle or a few 100-200 in light load like when you browse the internet you could actually use more energy then.

 

[internetlad]

Sure, you theoretically save power, but when it takes you twice as long to get something done, how much power does it save then?

 

[Thomas Creel]

Can't wait till SRAM is standardized. I want me 16GB of Cache!

 

[Guest]

OK, let's try something less "armchair", then.

My laptop consumes something on the order of 10-15w on average when actually being used. It's six and a half years old, now. If it had been used 24/7 for all that time, it would have used no more than about £100 worth of electricity, TOTAL... through the RAM, CPU, screen, hard disk, speaker, USB ports, wifi radios, etc. The compact all-in-one desktop I'm typing from, designed for low consumption, uses about fifty watts, is 2-3 years old, and maybe gets 3000 hours of use a year... So it's used about £50 worth, across all components.

Compare the partial saving you'd make on a small-ish fraction of those totals, vs the cost of replacing a costly SSD long before its time is due, and you'll see that you don't even need to do heavy research to mark this down as a fool's errand.

It might be somehow useful for servers or something, but unless you expect the vast majority of your "dynamic" data to actually be static and rarely written to, you're just exchanging one low cost for a higher one (and one energy sink - production and use of the DRAMs - for another... repeat production and lesser use of the Flash...)

I've got quite a few DRAM chips around the place that have been in use for more than 20 years. How long will flash last when being used as working memory, even if you find a way of greatly reducing the number of writes?

 

[Guest]

Also, SRAM has been used before quite happily... the problem is it's quite pricey. Remember the Amiga (500/1000 thru 3000/4000 and 600/1200 at the very least)? CPU-only "Fast" RAM (vs CPU/GPU "chip" RAM) was added using SRAM cards... on standardised PCMCIA slabs in the case of the 600/1200, too.

Very expensive at the time, and not exactly affordable for 1 ~ 4MB of the stuff even now.

 

[genz]

I don't think anyone truly understand the benefits of this. Either that or you're being blind and pretty stupid.

Think toward the future; it's not going to be long until we have 500Mbps read 512GB SSDs as a standard in a mid range PC. That's around half the speed of SDRAM, but here's the catch... What if the data on that drive didn't have to be copied to RAM before being executed/processed because the entire drive storing the date was seen by the CPU as one big RAM stick? If you do the math, including the wait cycles, it is on average far far slower to read an unpaged bit from a 32GB partition of 500mbit SSD through a stick of 1600 DDR3 than it is to read from 240mbit SSD directly, bypassing the RAM.

 

[genz]

[citation][nom]Anonymous[/nom]OK, let's try something less "armchair", then.My laptop consumes something on the order of 10-15w on average when actually being used. It's six and a half years old, now. If it had been used 24/7 for all that time, it would have used no more than about £100 worth of electricity, TOTAL... through the RAM, CPU, screen, hard disk, speaker, USB ports, wifi radios, etc. The compact all-in-one desktop I'm typing from, designed for low consumption, uses about fifty watts, is 2-3 years old, and maybe gets 3000 hours of use a year... So it's used about £50 worth, across all components.Compare the partial saving you'd make on a small-ish fraction of those totals, vs the cost of replacing a costly SSD long before its time is due, and you'll see that you don't even need to do heavy research to mark this down as a fool's errand.It might be somehow useful for servers or something, but unless you expect the vast majority of your "dynamic" data to actually be static and rarely written to, you're just exchanging one low cost for a higher one (and one energy sink - production and use of the DRAMs - for another... repeat production and lesser use of the Flash...)I've got quite a few DRAM chips around the place that have been in use for more than 20 years. How long will flash last when being used as working memory, even if you find a way of greatly reducing the number of writes?[/citation]

I don't think anyone truly understand the benefits of this. Either that or you're being blind and pretty stupid.

Think toward the future; it's not going to be long until we have 500Mbps read 512GB SSDs as a standard in a mid range PC. That's around half the speed of SDRAM, but here's the catch... What if the data on that drive didn't have to be copied to RAM before being executed/processed because the entire drive storing the date was seen by the CPU as one big RAM stick? If you do the math, including the wait cycles, it is on average far far slower to read an unpaged bit from a 32GB partition of 500mbit SSD through a stick of 1600 DDR3 than it is to read from 240mbit SSD directly, bypassing the RAM.

tl;dr This is NOT a replacement for RAM, its a system to make it so things don't need to be copied from your hard drive to RAM before your CPU can work on them, and is potentially a very big deal, even in the consumer space. It wont use any more read/writes than you usually would with it as a HDD.

 

[Guest]

sounds to me like they reinvented swap space / virtual memory

 

[Duckhunt]

I am not really sold on the idea of Flash RAM. We need some new tech.

We need the ability to turn on the pcs in an instant rather then everyone having them in sleep mode or hibernate mode. I know whole corporations who have all like that.

 

[Duckhunt]

who have all there pcs like that. The damn typos. ahhh.