Solid State Memory on the Rise
skaet writes "CNet is reporting that manufacturers of NAND flash memory are expanding the market for their chips - over the next few years - to eventually replace current methods of storage in media capture devices, mobile phones and even some notebooks as well as car navigation systems and large data storage at corporations and government agencies. From the article: 'The average notebook has 30GB (of hard drive storage). How long is it before the notebook has solid state memory? Five or six years,' according to Steve Appleton, CEO of Micron Technology, one of the world's largest memory makers. 'I'm not saying drives will go away. There will always be a need for storage, but when was the last time you tapped out a drive?'"
Most solid-state memory is pretty darn slow, and the stuff that's fast costs major $$$ ...
I'll buy it when it gets faster & cheaper - but then, flash *is* much faster than the ol' floppy - I was glad to see that go ...
--I gots 99 problems but a new machine ain't one!
AMD! Asus! Whoot! 6 years!
With solid state memory, won't you never have to reboot the OS? Will I still have to reboot Windows every so often even though the machine is capable of instant on/off? This feature of the hardware will put serious reliability requirements on all OSes. MS will have to finally fix the damn blue screen or its lack of reliability will be a serious henderence.
"Those that start by burning books, will end by burning men."
The average notebook has 30GB (of hard drive storage). How long is it before the notebook has solid state memory?
I havent seen a laptop with less than 40GB in I dont know how long. A long time anyway. Maybe this is out of date.
"There will always be a need for storage, but when was the last time you tapped out a drive"
Last week at the parents' place. Two days ago at work. Probably tonight as well at home. You were saying?
No matter how much storage you put in a given system, it will eventually be not enough. I've seen it a million times.
Also, flash memory is way too slow to be used as primary storage. Putting 512MB of MP3s on my SD card takes almost a three minutes. Drive to drive, that's under 10 seconds.
And let's not even mention how quickly a cache partition would die with the 100,000 writes before failure standard of current flash drives...
Most of my hard drives are already full. Let's review.
ShuttlePC Red Hat box: ~120G on a 200G hard drive. (old IDE controller) Full.
G4 Apple Mac, 3 hard drives totalling ~ 620G. Aproximately 60% full, and that's only because I recently added a hard drive.
PC Laptop, 80G hard drive. 25% full. And that's only because the hard drive was recently formatted and reimaged.
120G external hard drive. 75% full
27G external hard drive. Full
60G iPod. Full
So I'm a little shy of a terrabyte of active hard drive space. It would all be full if I didn't have multiple binders full of CD-Rs and DVDs.
But I guess not everyone regularly edits and encodes video on their computers, or routes their entire entertainment system through their computers.
I don't think hard drives will ever be big enough because data files will continue to grow as well. Solid State memory is and will always been a niche technology for areas that suite it best such as high reliability, small packages and extreme environments.
IMHO the market is already awash in solid-state storage microcomputers. They're called PDAs.
That's not what I meant.
I sometimes look at my download folder in awe, it's full of so much useless SHIT that I have no use for (or ever will have a use for, since there are new versions of just about ANYTHING in there) but... sometimes... it's hard to hit that delete key. It really is, I think we have a new symptom of obsessive compulsive disorder. I wonder how long until psychologists start to ask... "How long has it been since you deleted stuff from your download folder?"
"Most solid state memory is pretty darn slow"? May I respectfully ask, What are you smoking? What do you think is in your system (main RAM) and on your processor (cache)? There is no comparison between the worst case access time on a 10ns CAS2 DDR memory bus and an 8 ms seek time hard drive. The memory is *orders of magnitude* faster than the hdd. That's why modern operating systems do read and write cacheing of disk information.
Thinking outside my Head
I am familiar with the current Sun thin client. The paradigm "seems" to make
... except for the
... do you want the government or corporations
sense, until you realize just how fast technology leaps. My conclusion is that
what you get with a thin client is:
* Yots of your data flowing through eveyone's networks.
* Your data residing on someone else's "thing" somewhere.
* A regular fee that someone is charging you to do everything for you.
This makes sense, I don't say it doesn't. But for me, I would prefer to
pay the price of waiting for all of it to become available for me to use
at home.
If they had the thin client, I can have a laptop and more or less
control my data, and applications. It is work, and costs money, but
theoretically the market should take care of that
monopoly in Redmond, it would have been.
So, if they can stop or co-opt open source, then there will be nothing
but the thin client. Imagine what the government will use as an excuse
to keep their control to have to look through all your data.
You will have more and more data, maybe in the future, every moment of
your life could be stored
looking through that? Even if it is "for your own good"???
He was quoting Mooreon's Law...
1. Every eighteen months, the technology that you support will double in capacity.
2. Every eighteen months, the technology that you are supporting it over will do nothing.
Ergo, given that average notebook hard drives are currently around 60gb, rather than 30gb, Moore's Law (as opposed to Mooreon's Law) allows us to deduce he began applying Mooreon's Law 18 months ago - the doubling of average disk space since then has been ignored by him as it's a competing technology (and covered under part II of Mooreon's Law) - it has just taken him this long to get anyone to take him seriously.
Applying Mooreon's law from that point we can deduce:
1. 1GB flash 18 months ago translates to 2GB today, translates to 4GB in 18 months, 8GB in 3 years, 16GB in 4.5 years, 32GB in 6 years.
2. 30GB of traditional 2.5 inch HDD 18 months ago should still be 30GB today and will be 30GB in 6 years.
Therefore, 32GB > 30GB in six years - hence 5-6 years is an accurate figure for when flash should overtake 2.5inch HDDs.
If, you know, Mooreon's Law wasn't for Mooreons.
Everyone keeps mentioning their personal/private computers. Solid state memory will be big in corporate desktops. I'm a system administrator and where I work most of the computers use less than 2 GB. That's because only Windows and Office goes onto the drive. Very few additional programs are installed and documents are stored on a network mapped drive. This is what it's like at most of the larger workplaces.
I and my users would love to swap those 40-80 GB harddrives for 2 GB solid state drives and enjoy the benefits of a computer using less physical space, making less noise, consuming less energy, being faster and cheaper.
While it's generally accepted that harddrive space is cheap there is a minimum cost to pay. The smallest drives of today are still as expensive as the smallest drives of five years ago. With solid state drives I expect the price of the smallest drive to go away because it will be integrated on the mainboard - something that wull never happen with 3.5" drives.
I think this is really interesting.
The role of the home desktop is changing. It used to be the powerhouse. The computer you used when you really wanted to get some work done... but that came at a price: working in an office. Laptops work for me, because when faced with a block the best way of solving it is a change of scenery. Sitting in the same place for hours on end for "fun" is less appealing now I have to do it at work as well.
My G5 is easily twice as powerful as my G4 Powerbook, but I use my laptop 80% of the time. So why have a the G5? It's a home server. I have over 40GB of music, 10GB of photos, 100GB of home movies and PVR, and its incredibly useful to have a single point of access for the whole household, and because its a desktop its always in the same place, always on and permanently connected to the internet meaning that not only does it server the house, it serves us whilst we're on the move as well.
Even if my laptop could match the desktop for storage, I wouldn't want it to be bogged down with running the services, and all the laptops in the house having independant media store is just plain bad management. Also, tasks like media recompression, code compilation and games are still done best on a machine with more RAM than sense and a processor thats designed for performance not low power consumption: you use a push bike to get to work and for fun, you use a car to do the shopping. Sometimes you need the heavy lifting.
In fact I now have a couple of home servers, but thats because I'm a nerd: I have a PIII running debian to provide the low power services like a front end for Azuereus, a few small web apps and LAN facing NFS server. Which is why I can't wait for a 20GB NAND drive that improves the battery life of my laptops. I just don't need that much storage on teh move providing I've got a decent wireless network connection.
As for, when was the last time I topped out a hardisk... yesterday. I hve 300GB of storage available to me and I use all of it. You can never have too much storage, you just don't need all of it, all of the time, providing you can access it from anywhere in the world network latency and speed is more of a barrier than local storage.
Scared of flying, pointy things snce 1979!
I think I have heard this story ever January since 1970, and it was probably around before that.
A brief revue of the literature will reveal that, although its perefectly true that solid state memory follows More's law. HDs appear to as well.
At the time Bill Gates said "640k should be enough for anyone", a 40MB HD was the size of a Bendix washing machine, and cost about the same as a Ford Galaxie 500 with all the extras. 64k of RAM cost about ten times as much as a PC with no RAM.
In 1974, (check your library for old copies of Dr Dobbs) there was a serious debate as to whether the laws of physics made it impossible for memory to EVER cost less than 1c per bit!
And for those of you stupid enough to think solid sate means slow - ask someone what Google store their data on! People who know nothing about history are condemned to repeat it. The rest of us get shiney new USB thumb drives.
Sent from my ASR33 using ASCII
I have a slight problem with this in general and with your scenario specifically.
_ db2.php
Many small systems that I have setup boot from flash card, run in RAM and only write to the flash card when the configuration changes. Logging was forced to ram disk or disabled completely. This was done deliberately to avoid some of the problems with write endurance. I have had small Linux routers/firewalls running like this for 2-3 years now, so it must work. Distributing the wear across the entire flash card in this scenario might work quite well to extend the life since most of the disk is read mostly with only small, infrequent writes. I must assume, however, that there is still a finite life and that the system is doomed to fail at some point. Whether this is measured in 10's of years or 100's of years, I don't know and would have to do some lengthy profiling of the software and calculations to estimate.
In your scenario (i.e. running flash as a swap device) the flash memory is written much more often. In addition, I must assume that most of the flash is being written to almost continuously. After all, since memory limitation is what you're trying to overcome, I must assume that a large part of the 128M swap is being used most of the time. In this case, distributing wear across the whole flash is unlikely to do any good.
This site has calcs for predicting wear-out on flash-based disks:
http://www.bitmicro.com/press_resources_flash_ssd
The results range from almost useless (1.71 years) to who cares (130 years - I'll be dead by then) for 512 byte sectors on a flash disk device.
I design microcontroller hardware for a living and have had to deal with write endurance on EEprom (the technology flash is based on) for many years now. Although write endurance has been going up (it used to be several 1000, now it ranges to 10^5 to 10^6) it is still a concern in many of my designs. For instance, on a data logger I worked on recently, I went to a battery-backed RAM design rather than flash based. The battery-backed RAM design was good for about 10 years (limited by battery shelf-life) and the flash-based design would theoretically have worn out after only 6 months.
.. hard disks have a capacity of at least 2 Tera bytes
30 GB looks cool in 5 years but everybody forgets that disks will grow as well.
So forget about solid state.
That's a good point, but I think there's a way around it. Simply put, on a laptop, at any given time, the hard drive is either spun up or it's not. So we have two cases:
However, I do have to admit there is one wrinkle in this idea: the whole purpose of the flash is to be able to keep the hard drive spun down most of the time. Therefore, with this scheme, you will be more likely to need to have the hard drive spin up, because the odds of needing to spin up the hard drive on any given I/O increase as you increase the percentage of time that the hard drive is spun down. So, in situations where you have a cache miss, you will pay a big penalty, and assuming your flash really is slower at writes than the drive, you will pay a bit of a penalty (a smaller one) when your flash can't keep up with the writes and you have to spin up the hard drive to keep pace.
Still, I think for many workloads, the speeds could be quite good, and they'd be better with flash cache in some cases, and it could definitely result in power saving.