Conventional Hard Drive Obsoletism? Samsung's 32 GB Flash D - Page 2
Forum Storage : General Storage - Conventional Hard Drive Obsoletism? Samsung's 32 GB Flash D
Last message on previous page:
I think you misinterpreted my post. Nowhere did I suggest that the Samsung drive would lead to the demise of the HDD. The point I made relating to Vista's release spawning more similar products, is thoroughly explained in the following excerpt from the Vista web site:
Windows ReadyBoost
Windows Vista introduces a new concept in adding memory to a system. Windows ReadyBoost lets users use a removable flash memory device, such as a USB thumb drive, to improve system performance without opening the box. Windows ReadyBoost can improve system performance because it can retrieve data kept on the flash memory more quickly than it can retrieve data kept on the hard disk, decreasing the time you need to wait for your PC to respond. Combined with SuperFetch technology, this can help drive impressive improvements in system responsiveness.
Windows ReadyBoost technology is reliable and provides protection of the data stored on your device. You can remove the memory device at any time without any loss of data or negative impact to the system; however, if you remove the device, your performance returns to the level you experienced without the device. Additionally, data on the removable memory device is encrypted to help prevent inappropriate access to data when the device is removed.
Windows ReadyDrive
Windows ReadyDrive enables Windows Vista PCs equipped with a hybrid hard disk to boot up faster, resume from hibernate in less time, preserve battery power, and improve hard disk reliability. Hybrid hard disks are a new type of hard disk, with integrated non-volatile flash memory.
The hybrid disk is intended for mobile PCs running Windows Vista. Your data is written to the flash memory, which saves work for the mechanical hard disk—saving the battery power. The hybrid disk helps Windows Vista resume faster from Sleep because data can be restored from flash memory faster than from the mechanical hard disk. And since more data is written to the integrated flash memory than to the traditional hard disk, you have less risk of hardware problems with the hard disk when you're on the move. Windows Vista takes advantage of hybrid hard disk to save battery life, resume use faster from hibernation, and improve reliability.
Get it now?
The only problem with this technology (hybrid drives), is that the flash portion of the drive is tiny. I honestly dont see WHY they would use a small Flash memory chip in such a situation, unless, as usual, the manufactuers are trying to milk the public for all its worth . . .
I say stick a couple hundred megs of flash on the drives, and mark them up, whatever, but dont stick us with a 128KB flash module on the drive, and expect us to be impressed . . .
You should compare the Gigabyte I-ram with the Samsung 32gig SSD to see how they compare. I can see the I-Ram being much faster in throughput.
I agree. The hybrids should probably shoot for at least 512MB since that's about the size of most laptops' hibernate file (hiberfil.sys). 1GB would be better since this will be for Vista machines, which should be even beefier than today's boxes.
I'm curious if the hybrid drives end up saving power. Theoretically, the flash component could buffer quite a lot of disk writes and let the disk stay spun down until there's a read cycle required.
| Quote :
|
You don't need to rewrite 10TB of data, just a few bytes over and over again will have the same effect. Windows is always doing stuff to your drive even when your PC is just sitting there.
It doesn't matter whether the bytes are "new" data or "rewritten" data as part of swap, the device would have to see 10TB of data writes each day to have an 8-year operating life.
And Maury was being optimistic on the write speed. If you fire the SSHD off at the average tested write speed (29MB/s) 24/hrs a day you get:
24 hrs/day x 60 min/hr x 60 s/min = 86,400 seconds/day
x 29MB/s = 2.5TB/day
So that would mean the SSHD would last about 36 years before reaching MTBF. Again, assuming your computer was continously writing to the drive 100% of the time ofer that 36 year period. Now, odds are it won't really last 36 years. MTBF is an average time to failure and some cells will fail sooner, which increases the load on the other cells as there are fewer of them to split up the writes. However, the device would more than likely last until obsolence made it redundant.
Face it, even if the first cell died at 0.1M writes, it would be at 100% capacity for ~3.6 years of continuous disk writes. After that it would degrade at an ever increasing pace, probably lasting another 20 years or so of writes. I would imagine your drive doesn't operate for more than 10% of the time which means that the useful life is 36 years of full capacity (3.6 write/years /10% write utilization = 36 years).
And the thing to remember is that data-shifting happens local to the drive; the PC itself has no knowlege or intervention. The relocation isn't even perceptible across the IDE/PATA/SATA/SCSI interface. The PC *thinks* it is rewriting the same data blocks but the flash does it's own mojo with translation tables and ignores it.
I've been hearing about the Flash HDDs since 1993/4. They were supposed to replace the real HDDs at that time.... i suppose they encountered a few setbacks. So, 12 years later ...
Kig, your calculation is really... optimistic. Take the example of 32GB HDD that is has 31GB Full. The 1GB free is used as swap drive at it's maximum write rate:
1'000'000 / ((2.5TB / 1GB) * 365.25) = ~ 1 year
The life of the disk is to much dependent of the usage. I am kind of skeptic.
For sale (in case no one posted where to buy...):
Samsung flash drives:
http://www.king-cart.com/cgi-bin/c [...] atch=exact
(same place, previous page)
http://www.dvnation.com/nand-flash-ssd.html
| Quote :
|
Not quite. As I said, it does bit-swapping at times to make sure the data does not silently corrupt. Let's look at this, where it rewrites the As all the time.
1st iteration AAAAAAABBBBBBBBCCCCCCC............
2nd iteration ..............BBBBBBBBCCCCCCCAAAAAA
3rd iteration BBBBBBBAAAAAAAACCCCCCC............
4th iteration BBBBBBBCCCCCCCAAAAAAAA............
5th iteration BBBBBBBCCCCCCC................AAAAAA
etc, etc.
Which means that the 1G of space that keeps getting swapped around actually becomes 2G of writes as it shifts 2 chunks of data but over the entire drive space.
That turns the equation into
1'000'000 / ((2.5TB interface writes * 2 internal writes / 32GB) * 365.25) = ~ 17.5 year
This may seem bad as it causes unnecessary writes but it actually extends the life near max capacity and it means that cells storing data that doesn't change are still cycled to ensure they are still functional, preventing data corruption.
i think a hybrid of the 2 is ideal for the moment untill something more reliable can be created. if they do make hybrids the levels of solid state memore will probably be proportional to the level of the platter. maybe
Perhaps useful as a gaming drive, with a bigger (and cheaper) HD for all the other stuff?
I would say it's perfect for notebooks and HTPC's (only for the whole size factor), except for the ATA interface; WTF?
Wait a few years for capacity increases and cost decreases.
In regards to battery life, I read an article from one of the PC Mag where they put the actual drive in a laptop. They discovered that the battery life increased by a mere 10%. I guess this is because hard drives are not the major power usage on a laptop. I would expect that to be the screen.
The industrial segment of computer usage has been using eproms for data storage for years now. About 3 yrs ago, I updated a cnc lathe to a 32Meg solid state hard drive(try to find a 500Meg drv anymore), and it boots and writes to it every day. We have yet to have a problem with it!
remember, pc's write to Sdram and we have very little failure.
I think the idea is wonderful, and its time has come. I hope they come out with a Pcmcia slot version for laptops!!!!!
| Quote :
|
Not quite. As I said, it does bit-swapping at times to make sure the data does not silently corrupt. Let's look at this, where it rewrites the As all the time.
1st iteration AAAAAAABBBBBBBBCCCCCCC............
2nd iteration ..............BBBBBBBBCCCCCCCAAAAAA
3rd iteration BBBBBBBAAAAAAAACCCCCCC............
4th iteration BBBBBBBCCCCCCCAAAAAAAA............
5th iteration BBBBBBBCCCCCCC................AAAAAA
etc, etc.
Which means that the 1G of space that keeps getting swapped around actually becomes 2G of writes as it shifts 2 chunks of data but over the entire drive space.
That turns the equation into
1'000'000 / ((2.5TB interface writes * 2 internal writes / 32GB) * 365.25) = ~ 17.5 year
This may seem bad as it causes unnecessary writes but it actually extends the life near max capacity and it means that cells storing data that doesn't change are still cycled to ensure they are still functional, preventing data corruption.
I found a really nice read about the flash memory live time. It's from Bitmicro, but i suppose it applies or will apply in the near future to the other flash memory makers.
http://www.bitmicro.com/press_reso [...] sd_db2.php
| Quote : Example #1: Write Frequency in I/Os per day
|
However, it will be good for THG to write an independent article about the subject.
OK, this thread has shed some light on some things for me.
Is it possible to have heavy duty, reliable storage on these types of items? I'm assuming this is non-volatile, but this thread suggests that every so many reads or writes, the whole thing goes kaput? So what good is that? Might as well bank on a hard drive.
I have an application where I need read/writable storage that doesn't have moving parts, that is reliable. Am I dreaming?
Cheers
Every device will fail given sufficient time. Hard drives are not immortal; I've got several clients who will attest to that.
The worst case operating life of a flash device (100% write on a device at 90% capacity) is upwards of 15 years. Under more typical usage (10% write at 90% capacity) results in 150 years MTBF. A quick perusal of enterprise storage gives values of 1.2 million hours MTBF (WD raptor) and 0.62% annual failure rate (Seagate Cheetah 15k) which come out to ~150 years each. Seems pretty comparable.
Let's compare hard drive vs. flash general details.
Longest operating life: tie - but see below
Colder operating temperatures: Flash -25°C vs 5° C
Hotter operating temperatures: Flash 85°C vs 55° C
Hotter non-operating temperature: Flash >85°C vs. 65° C
Shock tolerance: Flash 1000G operating vs ~75G operating or 500G non
Power efficiency: Flash
Cost effectiveness: Hard drive
Largest capacity: Hard drive - currently
Under industrial applications (temperatures 140-180° F) a hard drive will simply not survive for any length of time if it even runs at all. By the same token, an all-weather application that has to operate below 40F cannot use a hard drive while flash can operate at arctic temperatures (-70F).
So if you want reliable and robust go with a flash device if you can afford it and can find one big enough for your needs.
I haven't even seen anyone cover the transfer rates of flash drive vs. sata drives.
Top flash speed of 60 Mbps vs top sata speed of 3 Gbps. UDMA Hard Drives may soon be obsolete thanks to SATA, but expect harddrives to stick around for the near future.
I won't mind having one if the price is around $300.00
| Quote : I haven't even seen anyone cover the transfer rates of flash drive vs. sata drives.
|
There's a problem with your math. Sure, SATA drives have a wider bus than this implementation of flash but that peak speed is actually transfer to buffer, not the platters. WD's 10k-RPM raptor has a max sustained transfer rate of 84MB/s but 1.5Gb/s transfer to buffer. (Note: the WD spec sheet switches between MB/s and Gb/s).
Then you are comparing apples to oranges. Enterprise SATA drives are not what are going into laptops. The biggest, fastest 2.5" WD scorpio has a sustained transfer rate of 75MB/s and a cache transfer of 1.2GB/s.
So SATA disks are faster than the first gen flash in sustained transfer rate.
This is first gen. Reliability is paramount to avoid giving the whole product line a bad rap. Power needs, environmental survivability and near-zero latency ensure this has a market in mobile devices even if it doesn't trump the physical disk in all aspects.
Plus a different controller design would change the tranfer rates. They could add cache memory but that runs the risk of defeating the "nonvolatile" label since a power failure during write could result in several MB of lost data. Or they could use multiple existing controllers and RAID-0 the flashdisk internally; the transfer rate would increase nearly linearly with the new controllers. It would increase the cost, of course, but having a 120MB/s sustained transfer is pretty impressive since it trumps enterprise hardware.
BTW, I'm not expecting flash drives to obsolete physical disks for quite some time. I am expecting hybrid drives to provide a significant boost to the usability of mobile devices and possibly be integrated into server farms where the cost of power+cooling over time pays back the initial expense.
You're quite right. I was looking for the average speed of information being accessed from the platter itself. The closest thing I found was access time. The main point I was trying to make was the transfer speed available to hard drives. If the platter reading speed is improved, the SATA2 bus speed is already there.
And no, I was not comparing apples to oranges. I sent top speed information I could find on both, not "burst" speed on one and average speed on the other. I did not claim that SATA2 drives are going into everything, including laptops. I was making the point that the top speeds among the two TYPES of technology would point to the fact that Conventional Hard Drive technology is not going to be obsolete. I also took special care of what you "Noted". (Note: Both links I gave compared BITS per second, not BYTES--look again)
You're right about the comparing SATA to first gen Flash. But how can I compare SATA to what flash is going to be?? I don't think flash is going away either, but I must compare current speeds. Flash has promise. SATA has promise. I'll leave it at that.
I do fully agree with you about the hybrid drives becoming more useful. It's in the best interest of HD manufacturers to make them ($$$$). Many users have situations and needs that would be better served by hybrids.
If you'll read my comment again, you'll see I pointed out that WD's spec sheet switched between MB/s and Gb/s, not you.
And at this point I'm thinking that a 32GB drive is both too large and too small. Virtually everyone will need more than 32GB of total storage, except for some specific industrial or medical applications. Which means a combination of disks & flash.
And if you are going to have a mix of disks and flash, 32GB is too big (translation: too expensive) to bother with. I think an 8GB drive would be the best price/performance ratio since it lets you have a full Windows install (~3GB) + swap space (~1GB) + a couple of key apps and data files (~4GB). A standard hard drive provides the bulk storage for the stuff you don't use every day.
In a way, this feels like the old SCSI/IDE hybrid systems a lot of people put together, with ultraSCSI disks for OS & swap and cheap IDE for bulk storage.
I have some experience with these devices. And let me tell you they are not new technology. SSDs have been around in the inductrial sector for some time now. They are used extensively in industrial PC applications where vibration and other factors render a disk platter unsuitable.
Advantages:
-better performance than HDD
-solid state. No moving parts, which normally would mean MTBF would be greater than HDDs if it weren't for the life cycle of the NAND/NOR/EEPROM media.
Disadvantge:
*-Life time. Current non-volatile ram chips all lack suitable MTBF values. It is a limitation of the NAND/NOR/EEPROM technology.
What does that mean? It's not suitable for everyday desktop usage (with current OS trends).
What can be done about it? A common approach for industrial systems (in the case of XP Embedded) is Electronic Write Filter (EWF). This essentially disables the flash device altogether after boot. On boot, Windows copies the entire "hard drive" into system memory and the entire operating system operates from there. Upon system shut down the memory is written out to disk. It eliminates all but 2 read/writes for each "sector" on the flash device per boot.
So, that's a great solution for small embedded pc platforms. Think about it, the OS never has to seek a data source outside of main memory! This is great for speed and lifetime, but a nightmare for reliability. If you lose power, you've just lost the last 2 days of information! I'm currently investigating the possibility of issuing mandatory flush commands to the OS to solve this problem for my application.
One difference between this product and the predecessors is size though. Although there are "high capacity" SSD available, the price is terrible. Industrial PCs also use CompactFlash cards plugged into CF adapter cards to interfaces like ATA. As you know, CF cards are getting bigger and cheaper all the time. This seems to be the best of both worlds (HDD and SSD), because you can simply replace cards as they become too small or full of bad "sectors". In this case the interface for the device becomes limitless. Just make a different chipset to translate to the CF card. It's also great for repairs. "Oh, Windows just took a dump? No problem, we'll fedex a new CF card out to you. No, you don't have to take the computer apart."
You'll notice I'm putting " around all the traditional HDD terminology. That's because SSD replicates traditional HDD access methods transparently for the OS.
Now, let me clear up some things. This is based on my experience dealing with CF cards in an *industrial* application (where computers are running 24/7 and performing HDD operations at least every second).
| Quote : The worst case operating life of a flash device (100% write on a device at 90% capacity) is upwards of 15 years.
|
Theoretically you are right, however in practice flash devices do not last that long with close to full capacity and constantly running. Consumer grade flash cards are rated at 100,000 or 1 million read/write cycles. Industrial versions sport much better operating temperature ranges and up to 2 million read/writes.
Take a look at current SSDs out in the market (this is not the first offering, remember). They are sometimes called IDE flash disks. I just came across a 4GB PATA drive for $200 US. No, it's not as fast as this debut model, but it's still ahead of 2.5" platter drives.
| Quote : http://www.bitmicro.com/press_reso [...] sd_db2.php
|
This sounds great! But I'm here to tell you that it's no where near that long in practice. I don't have anywhere near the 50,000,000 I/O operations per day that that site quotes, and some flash cards fail after as little as 1 year. In my application, if I were using the industrial flash cards, I shouldn't see failures until 2 years.
I've also seen a commercial card with 1 million read/write cycles develop multiple bad sectors after just 6 months. It was plugged into my PDA, and that was plugged into my radio to play music while I drove (I drive a lot).
| Quote : About 3 yrs ago, I updated a cnc lathe to a 32Meg solid state hard drive(try to find a 500Meg drv anymore), and it boots and writes to it every day. We have yet to have a problem with it! |
Great! See fellas, it all depends on the application. I'm sure CNC lathes have little reason to write out to disk in the first place. They're mostly controling machine movements.
Now let's talk about what *could* be done.
The next best technology on the horizon is PCM, or Phase Change RAM. See this:
http://hardware.slashdot.org/artic [...] 29/1242219
Or this:
http://en.wikipedia.org/wiki/Phase-change_RAM
This technology *does not* suffer from the failure rates of NAND. Also, it would allow per bit addressing again as with NOR (unlike NAND). Truly random access.
Personally, I see a big future for certain applications with this technology. Imagine what a server could do with a RAID 10 or RAID 0 of 20 something 8 GB PCM flash cards. Whew! I think I found a solution for my database server! Now, imagine that that array could last significantly longer than a HDD, maintain solid state performance, and still be non-volatile!
Impressive. I can't wait. But, what does this mean for the general computing consumer? I don't think SSD is going to take off for the desktop. But if we rethink the OS design, we should be able to adjust to the limitations that current NAND has.
Also, as a previous poster said, HDD manufacturers have managed to stay ahead of all the hype about the next HDD killer. But... have you noticed that they are doing it by:
A. Bumping up the capacity and calling that an advantage (does the average computer user really need 80GB? Come on... And what's with kids walking around with iPODs that store as much as a desktop hard drive? Their collections will never grow that big... But that's for another story...)
B. Improving the algorithms (NCQ, etc)
or
C. Injecting coating chemicals to prolong platter lifetime and performance (http://www.eurekalert.org/pub_releases/2004-08/acs-hd081604.php)
Now there's talk of using flash memory inside the hard drive for cache. It seems hybridization is the way of the future. When you think about it, a computer is nothing more than a calculator with a cascade of memory locations (in terms of access delays) to pull input and output data from. What hard drive manufacturers need to do is reduce access times, not increase capacity! Access time and transfer rates are the biggest problems with today's computers. If we could move more towards faster data storage and retrieval, we could really take advantge of the newer processors. Maybe then I'd consider shelling out over a grand for a processor.
Then the next obvious hurdle is interface bandwidth. Having a RAID of super fast solid state flash memory is great, but it won't do any good over a pci bus or ATA-100. That's why some server mobos employ multiple busses, so that you can dedicate raid channels for better performance.
Summary:
- NAND flash is slow, expensive, and dies too much but it's faster than HDDs
- PCM is going to rock, but it's going to be expensive
- Data storage devices need to be faster, not larger.
- Desktop users have no need to buy this technology. It should be for laptop users who don't care about capacity, embedded/industrial PCs, and high performance data servers.
My 2 cents.
Thank you molafish, for your real world experience testimony and analysis.
This subject really deserves a detailed article from THG.
nice review Molafish, but remember, in cnc controls, there are disk writes. the control in question used a conversational interface, and the replacement of the HD was because it did fail in the industrial enviroment.
In the time period I've had the solidstate drive, I would have went through 3 or more HD's.
And I agree, The desktop segment is not were this technology is going to shine. Laptops are!!! And MS is dictating hybrid drives for laptops running Vista . Not that they won't run it, just would run it better...
You can't depend on MTBF as an operating life but it's the only thing available. Like the old saw goes, there are lies, damn lies, and statistics.
Flash errors will induce cascade failure since the long operating life is based on spreading the load out over as many cells as possible. As the number of operating cells drops, the load increases, resulting in more failures.
I also glossed over interface failures, which renders the device completely inoperable. I considered it a moot point since since flash & disk should have an equal likelihood of interface failure. Its possible the flash cells themselves were still functional on your dead flash devices but the controller kicked the bucket. I'd say that IMO death of the interface happens about 30% I have a disk die.
I wonder how well the SSHD would fair against one of the 150x 8GB CF cards. I know there are simple IDE->CF interface adapters.
| Quote :
|
I have a 20Gb ZEN (real world: 19,056 megabytes) and I'm struggling to fill it up. I have 308 songs which use up 12,335 megabytes (approx. 40 megabytes per song). I just set the rest of the device as a removable disk.
However, my desktop has an 80GB HD and it's nearly full, even though it's not connected to the internet (I'm at work now). When I retire it I plan on putting it in the loungeroom and ripping DVD's onto it, for which 80GB will be too small.
You are right!
Many folks never see what is available in the industrial/military market.
Folks should take a look at www.M-Systems.com. They have been making this stuff for a long time. They hold the patent on the wear leveling technology that you are talking about.
Take a look at the way they calculate life. With active wear leveling, the life depends on the durability of the flash chips and how often they get written to.
Bottom line is you will never wear it out running Windows as hard as you can.
See www.wdlsystems.com for pricing on their products.
This Samsung product will give them competition.
Did you notice the ARM chip that is used as a controller in the "disk" drive?
I just checked the mSystems web site and they have just been bought by Sandisk.
Look at their TrueFFS technology. That is the magic in all of this.
Dennis
32GB is plenty for me (in fact all I need is 2 of these drives). I built my latest system with a 40GB SATA; which was a 20GB Upgrade from my last hard-drive.
You see, I manage my space very well. keeping downloads and music on either DVDs or external drives, so winXP isn't left indexing though that stuff when I need to access something, this method keeps my computer running pretty frisky even if I don't defrag months on end. Plus this makes it easy for me to re-install OSes without much backup (just burn My Documents).
What I REALLY want to know is do these drive decrease load times in video games, and by how much?
I'd love to build a portable storage/music & video player from one of these, as I'm in the market for one right now, preferably one that uses a standard HDD.
PQI has been making these as a dom for at least 3 yrs. I have been using one in an industrial enviroment and it has replaced HD's. they failed quite often.
here is the link if you want more info. http://www.dom.pqimemory.com/
There are 5 identified and unidentified users. To see the list of identified users, Click here.
Please mind
You are about to answer a thread that has been inactive for more than 6 months.
If you still wish to proceed, please ensure that your posting is original and does not duplicate or overlap any prior responses to this thread.
