Conventional Hard Drive Obsoletism? Samsung's 32 GB Flash D

[pschmid]

Samsung has provided us with a new 32 GB Flash Solid State Hard Disk. How does performance compare to conventional hard drives and what are the implications for desktop and mobility users?

 

[irishjd]

One item I noticed that was not discussed in the article is MTBF. Aren't flash chips limited to a finite number of writes? If so, wouldn't this equate to a low MTBF for the unit?

Jon

 

[Doughbuy]

I never heard about that, I would think that flash could do as many number of reads or writes... what would limit them?

But then again, I should pay more attention in my solid-state classes...

Either way, I think a hybrid HDD is a good idea for now, but what I'm wondering is the maximum capacity of flash... can it reach above 1TB?? Can it surpass PMR?? Will it be cost-efficent?? Any thoughts?

 

[Talon]

For the time being the biggest hurdle I see is the cost-efficiency you mention. It is currently very inefficient meaning the cost per GB for solid state is way the heck above HDD cost per GB. That may eventually change if they can sell enough of it to bigger industry but will likely take years.

For now I think its safe to say if you want a good bit or storage or especially mass storage HDD is the only way to go. If you had the bucks and didn't need a lot of space I think an application in moblie PCs or laptops are the best place to start for these new "drives".

 

[sintekk]

Yes... I too am wondering about the MTBF on one of these drives. In the past I have worn out both CF cards and USB flash drives by running OS's off them. From my experiences flash memory doesn't give you warning sings of failure as a machanical hard drive does... It just fails along with your data. It's just a matter how many r/w cycles can this drive take compared to its cheaper counterparts. My personal opinion of a device like this is that it would make a perfect OS drive in conjunction with "classic" hard drives for data storage. I've used over 50 Samsung desktop drives in years past and have only seen two fail -- one because I dropped it.

Paul AKA Sintekk

 

[03flat4]

For the time being the biggest hurdle I see is the cost-efficiency you mention. It is currently very inefficient meaning the cost per GB for solid state is way the heck above HDD cost per GB. That may eventually change if they can sell enough of it to bigger industry but will likely take years.

For now I think its safe to say if you want a good bit or storage or especially mass storage HDD is the only way to go. If you had the bucks and didn't need a lot of space I think an application in moblie PCs or laptops are the best place to start for these new "drives".

Yeah, this is a perfect item to sneak into the mobile market, and hybrid drives, and as costs go down, and capacities increase, it can seep slowly into the mainstream. I think if a 30 gig drive becomes affordable, I would get one for my OS and swap file like they mentioned in the article, and just leave everything else on the hard drives

 

[Orville]

Patrick,

Thank you for this very interesting article. Please tell Samsung to go home and come back with something more definite that is priced around $100 and is SATA. It's taken a long long time to move past PATA. We're trying hard to get rid of those behemoth PATA connectors that dominate mobos and block the air.

If this unknown priced NAND flash is mature enough, maybe it will be usefull for speeding up boot and application loads, reducing access time to the paging file and maybe even lower power draw. All of those attributes are commendable.

Orville

 

[neocristi]

Hey, samsung will probably use SATA for products that are comparable to actual HDD drives

I am glad they are making these first attempts.

In a few years we will have SS Disks in our PC's that are connected SATA or sommething else

 

[Synthetickiller]

Patrick,

Thank you for this very interesting article. Please tell Samsung to go home and come back with something more definite that is priced around $100 and is SATA. It's taken a long long time to move past PATA. We're trying hard to get rid of those behemoth PATA connectors that dominate mobos and block the air.

If this unknown priced NAND flash is mature enough, maybe it will be usefull for speeding up boot and application loads, reducing access time to the paging file and maybe even lower power draw. All of those attributes are commendable.

Orville

Have you seen the price of SSHDDs? 900 for 32 gigs would be pretty good. Check it out...

http://www.dvnation.com/nand-flash-ssd.html

I like the idea of the hybrid drives as well. I'd consider getting one if the price point was there. I guess we'll find out in 2007.

Hopefully in a 3 or 4 years, solid state drives be competatively priced. I'd pay 100 bucks for a 200 gig drive if it had those responce times.

Does anyone know if you flash drives have a finite number of times they can be written to? I didn't think they had a life span in those terms. Can anyone validate this?

 

[Doughbuy]

Would anybody buy a 32 gig flash MP3 player?

I would say that we can probably only except hybrid HDD's, but the mobile market will move solely to flash... no more micro-hdd's, useless and cause enough headaches when they die.

 

[sphericaline]

Has anyone tried to boot Linux off one of these drives? I've tried to with 16Gb Super Talent Flash drives and wasn't able to though Windows installed and booted fine. The Linux install would freeze trying to partition the flash drive and we tried fc4, fc5, ubuntu, and debian. If any of you have been able to get Linux running off the Samsung 32 Gb Flash drive please let me know.

 

[CJK79]

According to wikipedia:

"most commercially available flash products are guaranteed to withstand 1 million programming cycles"

 

[Sokolum]

Ofcourse, for powers users 32GB isn't much. But i really would like to use that sollid drive for my proffesional useage. Nomore being afraid for a HD crash and loose all my precious data.
Besides from that, imagine placing sollid drives for specific apliances, like routers, firewalls. Perfect.

It's a begin and you could add more pcb's on top of eachother (i guess).
I am for sure there is a market, only i you could go encrypt the data. They could let the drive boot first to it's own system and type your password to continue... My data will always be save.
I am for sure they have already something like this... I want that...!
Secured data protection. If they have not, my idea

 

[Doughbuy]

1 million, thats not that many...

 

[theaxemaster]

According to wikipedia:

"most commercially available flash products are guaranteed to withstand 1 million programming cycles"

And "programming cycles" is very ambiguous. Still, if the I/O performance is worth the limited life, companies will buy it. Just back up the flash drive to disk/tape every so often and get some error control on the drives and you're good to go.

Drive die? swap it out and rebuild it from the backup.

I suspect they kept the IDE interface to keep it compatible with current server systems. The only place you really won't find those connectors is a) gaming systems (VERY recent ones) or b) SCSI systems. a is a small market. b is bigger, so you might see these for SCSI in the future as well.

 

[Ram11]

SATA 300 would be a minimum for serious adoption. However hats off to Samsung for the attempt. I am sure the uses, especially for critical apps, will be many.

If the read write rates were to be increased to 100MBps or more,( which I guess it will eventually) the potential is huge, even with a big price premium.

One application is that such a drive could replace the Tape media in all but the smallest camcorders.

Panasonic already uses Flash media called the P2 ( 80 MBps) in a few of their cameras and the workflow benifits are reported to be phenomenal.

ftp://ftp.panasonic.com/pub/Panasonic/Drivers/PBTS/papers/P2-WP.pdf

 

[killerliz]

Using that 2.5' HDD form factor, i think 2 PCBs worth of flash memory can be fitted into it. So instead of 32Gb they have now, it becomes 64Gb!

But nobody's gonna buy it, because ingnoring those filthy richies out there, it will be very bloody expensive!!!!!

But think about the peformance, especially if they have changed their interface to SATA II 300.... ...., and the noise of nothing that these HDD makes

I will wait for the future when 160GB Flash Disk sells for $100. By then who knows what bizzare new memory technologies would have come... .... MRAM?

 

[joebob2000]

Can someone explain why the data transfer diagram (the first chart) shows 0 to 100GB? This is a 32GB disk, right? Not that it matters since the chart shows a flat performance level anyway, but I would like to know why the numbers are off. Oh and to the OP, your link is wrong, too many dotcoms

 

[WarMace]

I understand SATA is probibly on the way, but i still cannot fathom why they dident start development with it.

Maby they thought they would start with the speed of the majority of notebooks?

Wow, i love the web server demo graph! http://images.tomshardware.com/2006/09/20/conventional_hard_drive_obsoletism/image019.gif
It needs its own theme song like "Nand Drives, F**k Yeahhhh!"

 

[WarMace]

Digg this artical at http://digg.com/tech_news/Conventional_Hard_Drive_Obsoletism_Samsung_s_32_GB_Flash_Drive_Previewed

 

[niz]

One item I noticed that was not discussed in the article is MTBF. Aren't flash chips limited to a finite number of writes? If so, wouldn't this equate to a low MTBF for the unit?

Jon

Flash has a limited no. of writes, usually a maximum of 100,000 operations in the same location.

You can easily end up doing 100,000 write operations in the same location of an HD very quickly. especially if you're running windows off of that drive, as its paging algorithm is very inefficient to the point of being broken.

The whole idea of using flash with HD's is a terrible idea for this reason. its GREAT for drive manufacturers though, as you'll have to frequently replace your hard drive if you move to this technology.

I'm dissapointed that Tom's didn't even think to mention this fact in the article as its key to this technology.

 

[kigmatzomat]

I'd like to see a performance test comparing:

1. the OS & swap on the SSHD with the apps on a conventional drive
2. the OS & app on conventional drive with swap on SSHD
3. OS & swap on a conventional drive with the apps on the SSHD.

I've got a number of friends in the creative fields (graphic design, video editing, etc) who might get a decent boost out of a reasonably cost-effective performance booster. Option 2 probably requires the smallest SSHD with 1 & 3 varying with the applications in question. I figure 2GB of SSHD is pretty cheap and would be a decent swap partition. 8GB may hit that cost/performance ratio of letting you have either the OS+swap or swap+key apps on the SSHD.

 

[kigmatzomat]

Flash has a limited no. of writes, usually a maximum of 100,000 operations in the same location.

You can easily end up doing 100,000 write operations in the same location of an HD very quickly. especially if you're running windows off of that drive, as its paging algorithm is very inefficient to the point of being broken.

Most flash drives have a built in system that shifts data so that writes do not occur at the same point. So if your memory looked like this:

AAAA..................BBBBBB.....................CCCCCCC..................DDDDDDD

and you rewrote BBBBBB you might get:

AAAA..................---------...bbbbbb........CCCCCCC..................DDDDDDD

With the old data (-----) still being there but marked as "available" in the FAT table (or equivalent) and I believe a counter is included to indicate how many times the cell has been written to, or at least what order the cells were originally written. The available space with the fewest writes/lowest order is used the next time space is needed. If a cell fails during a write cycle it is marked bad, like blocks on a hard drive, and the data located at a different cell.

I also believe that some non-volatile memories will copy data to a new memory cell after so many read operations to ensure the cell gets tested periodically by a write cycle to avoid "invisible" data corruption.

Which means that a 1M write MTBF will require something around 1M writes/total memory cells. If the cells are 8K a 32GB drive would have some 4,000 cells so you'd need to make around 4 billion writes before you reach 1M writes on any given cell.

Assuming a worst case scenario of writing 8.01K/incidence (requiring 2 cells) results in something like 16TB of data writes.

At which point you fail out some number of 8k cells. The exact number you cannot predict without having more than just the Mean Time Between Failures. However the system will likely then continue to function at its new degraded size for another (1M writes x remaining cells).

 

[niz]

Flash has a limited no. of writes, usually a maximum of 100,000 operations in the same location.

You can easily end up doing 100,000 write operations in the same location of an HD very quickly. especially if you're running windows off of that drive, as its paging algorithm is very inefficient to the point of being broken.

Most flash drives have a built in system that shifts data so that writes do not occur at the same point. So if your memory looked like this:

AAAA..................BBBBBB.....................CCCCCCC..................DDDDDDD

and you rewrote BBBBBB you might get:

AAAA..................---------...bbbbbb........CCCCCCC..................DDDDDDD

With the old data (-----) still being there but marked as "available" in the FAT table (or equivalent) and I believe a counter is included to indicate how many times the cell has been written to, or at least what order the cells were originally written. The available space with the fewest writes/lowest order is used the next time space is needed. If a cell fails during a write cycle it is marked bad, like blocks on a hard drive, and the data located at a different cell.

I also believe that some non-volatile memories will copy data to a new memory cell after so many read operations to ensure the cell gets tested periodically by a write cycle to avoid "invisible" data corruption.

Which means that a 1M write MTBF will require something around 1M writes/total memory cells. If the cells are 8K a 32GB drive would have some 4,000 cells so you'd need to make around 4 billion writes before you reach 1M writes on any given cell.

Assuming a worst case scenario of writing 8.01K/incidence (requiring 2 cells) results in something like 16TB of data writes.

At which point you fail out some number of 8k cells. The exact number you cannot predict without having more than just the Mean Time Between Failures. However the system will likely then continue to function at its new degraded size for another (1M writes x remaining cells).

Your logic is skewed.

Firstly, you've assumed a 1 M write MTBF, whereas nearly all flash is actually 100,000.

Secondly the rest of the system won't suddenly be able to do another full-life of writes just because you've marked already-failed cells as bad. They're all about to fail at that point (assuming the writes have been roughly evenly spaced, which is true because flash drivers do that on purpose as you said).

 

[meekj]

Samsung has provided us with a new 32 GB Flash Solid State Hard Disk. How does performance compare to conventional hard drives and what are the implications for desktop and mobility users?

These things are called DOM or DiskOnModule. I am very excited with these prospects. Industrial users use them all of the time. The real problem is with NAND Flash is read and write cycles. Typically they have 1,000,000 to 2,000,000 R/W cycle ratings. Some places claim tha they have 1,000,000 hours MTBF for Disk on Modules. At the base of most flash ram is a flip-flop transister segment. They say it should hold for about 10 years.

Below someone asked about Lynux, this is perfect for Lynux applicaiton for discrete purposes.

My vision is having three or four of them with an old time Promise IDE Raid 5. Often the Flash Ram has an ECC or error correcting scema imbeded, but I would like to use Raid 5 for security.

I would like to see ABIT or other manufactures add a slot for two or there flash drives, and with the help of BIOS, they could be USB units, 1, Boot, 2. Disk Cache. and 3. Validated Operating system. Setups to boot from these devises. The Validated Operating system is like Microsoft Update a place for validated dlls etc. Comparison of Valided Operation System Files with Validated Operating system files.

The 1,000,000 read write cycles bothers me, fo Raid 5 is the solution. Raid Controlers should be set up for this to, for example, DOMs should not be defragmented. This would wast read write cycles and gain nothing in accesss times.

I would like Toms Hardware to have a contest on how fast a computer can boot. A RAID 5 Disk on Module solution could be very fast. http://dom.pqimemory.com/ talks a lot about the different Flash Disk replacemnt. You have 40 pin, 44 pin, and USB solutions.

Take a Promise FASTTRAKSX4060 put 4 Flash Devices, stipe or Raid 5 and test performance.

The mechanical nature of hard drives are a big minus and weakness in our computer systems, they are not enegery effecient, and they seek times and limited transfer speeds. This is the future. Now when will it be afordable