Slashdot Mirror
Samsung's 64-GB Solid-State Drive
Anonymous Howard writes "Just a couple of weeks ago Sandisk introduced a 32-GB solid-state drive. Now Samsung has one-upped them, unveiling a 64-GB solid-state drive. They are expecting to begin shipping in the second quarter of this year. Samsung says the device can read 64 MB/s, write 45 MB/s, and uses just 0.5 W when operating (0.1 W when idle). In comparison, an 80-GB 1.8-inch hard drive reads at 15 MB/s, writes at 7 MB/s, and consumes 1.5 W when either operating or idle. No pricing yet."
yes. a solid-state hard drive is exactly that - a normal HDD that uses a different mechanism for storing data. Usually its a pinning platter, this uses non-volatile memory chips. The interface and size are the same, so you just use it as you otherwise would.
Personally, I think 64Gb is a bit much for me, I'd stick the OS and swap files on there - which come to about 10Gb on my current XP machine.
The SanDisk 32GB version reports a 2 million hour MTBF... http://www.sandisk.com/Oem/Default.aspx?CatID=1478
That's quite a bit better than typical hard drives these days!
Has anyone found MTBF information regarding the Samsung versions?
the samsung is a 1.8 inch drive and the A-DATA is a 2.5 inch.
I'm under the impression that this is solved by two factors.
First, flash parts have internal controlers that remap the flash to level out the writes. (I remember hearing about some researcher who developed a great flash file system, only to find that it didn't make any difference because of the remapping.)
Second, flash parts can handle orders of magnitude more writes now than they could a few years ago.
The SSDs from Samsung and SanDisk will last for years and have an MTBF of 2 million hours. San Disk claims there device will last at least 5 years.
Last longer? The MTBF on flash storage is about an order of magnitude greater than magnetic storage these days.
TODO: Something witty here...
For comparison's sake, apparently some of my 250GB WD hard drives have a MTBF of only 1 million hours.
They specify 10 years for flash memory to hold it's data, but in practice (e.g. not at the highest temperature or most extreme operating voltage) it is significantly longer. I don't know to what extent the hard drives work around bad sectors, but they probably do it for both flash drives and the traditional magnetic type.
It's not wasting time, I'm educating myself.
Well... that doesn't necessarily mean it's as fast at random access as it is at consecutive access.
Normal computer RAM is also faster at consecutive reads than random reads.
One of the penalties for refusing to participate in politics is that you end up being governed by your inferiors - Plato
One possible reason is an OS that will not run, or will not run correctly, without paging. Linux actually fell into the latter category once upon a time, it would run like crap without swap (unless you made some patches.) SunOS4 was SOLIDLY in that category, but it's old (BSD 4 based - IIRC mostly 4.3, with some 4.4-lite code in the last release of SunOS4?) Windows 2000 would allow you to turn of all paging files, but would blue screen on boot if you did so as it absolutely required some paging (but you could have a teensy, fixed-length paging file.) But in a system that isn't so poorly designed that it requires paging, adding more RAM is a better solution any time you have the slots and the cash. It stops you from paging! That's pure gold right there. Who wants to wait for that?
One solution, of course, is to use a DRAM SSD for your swap if you're out of DIMM slots. Then only cash is the limiting factor. Granted, it still limits me right out of that particular game...
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Fortunately flash fails on writing not on reading so while you can't write to sectors that have failed you can still read them.
If comparing these to 2.5" drives instead of 1.8" drives the advantages aren't as drastic.
* 2.5" drives consume between 0.8W to 2.5W (ok, seeking eats a lot, but during sequential read or write, they consume similar amounts), almost no power consumption when they spin down.
* 2.5" drives give 53MB/sec read and write.
* 2.5" drives are very cheap and have triple the capacity.
The solid state drives are still at an advantage, but it's not quite as large as compared to 1.8" drives.
unless you're planning on treating it as WORM memory, flash has somewhat limited lifespan. the flash we're using at my collage for our integrated processing stuff is rated for 1 million write/erase cycles, but the datasheet doesn't even mention MTBF.
upon the advice of my lawyer, i have no sig at this time
THe Samsung site says these new drives are based on single cell level NAND technology. It doesn't have as high a density as MCL NAND. Bbut each cell can do 100K rewrites as opposed to the 10K rewrites of the more common MCL NAND. See EDN article on difference between SCL and MCL NAND http://www.edn.com/article-partner/CA6319917.html
It's not about the MTBF (the wear with age), yes you can almost indefinitely read data from you flash drive, when compared to harddrives, because there's no mechanical wear.
/var and /tmp, most of the rest of the installation can be read-only), and that support special file systems designed for lower wear (JFFS and such), may fare better : for example there are some Linux distribution that are tested for running from flash, like Damn Small Linux.
BUT!
The flash cells have a limited number of write cycles, which is very small compared to hard drives. If you write too much data on the same sector, the sector get very quickly broken.
If you used a flash card for swap, it won't last long at all (because some sectors get constantly written over).
To limit those damages, flash controllers use "wear level". That means that the small RISC controller that interface between the flash cells and the computer interface (ATA/CF, SD, USB, etc.) dynamically remaps the sectors so the wear caused by write cycles is distributed over several different sector.
Let's say that an OS constatly writes data on the first couple of sectors. Instead of always writing on the first few cell, the controller remaps a different physical flash cell, to the logical disc sector seen by the OS.
This works as a charm for flash media storing files likes used in digital cameras and such.
But doesn't perform as well when used by an operating system.
Windows XP is specially bad at this.
Other OS - such as Linux or *BSD, that already have good support for running on slow read-only media (LiveCDs) for a long time, that don't need writing that much (except
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Not sure what it means but 1 of my 3 flash memory cards has gone bad in 2 years.
Looks perfect- but it reports a formatting error when loaded into my camera and reformatting it doesn't fix the problem.
And hard drives last a lot less than they advertise too (all those google related articles 2 months back).
She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.
These things have a .12ms avg seek time.
Raptors are still at abt 4.5ms.
Additionally they can Read in Parallel, randomly. (ie not limited to the spindle speed and read head physical limitations.
They are NOT just as fast. No sir. They make raptors look like they're standing still, and everyone else like they're in a time warp.
I saw a demo of an XP boot / shutdown on HD vs SSD. (where'd that link go) It's pretty amazing and significant.
Add in an OS / CPU that can HANDLE parallel loading of drivers / OS modules, then it'll be something like the Linux 6 second boot (ok maybe 10 second).
Assuming a million write/erase cycle limit for this SSD, it would take over 46 years of 24/7 writing to reach the limit on the entire disk. Of course that's unlikely as normal usage patterns would not use the drive in such a way, but I thought it was interesting to know.
(((64 * 1 024) / 45) * 1 000 000) / (60 * 60 * 24 * 365) = 46.1807317
8 of 13 people found this answer helpful. Did you?
The flash parts used in these devices can only program approx 10k times before they can be expected to start failing.
Modern NAND flash is in the 100k+ erase/program cycles... from an ST application note on wear leveling: "In ST NAND Flash memories each physical block can be programmed or erased reliably over 100,000 times." Of course, the wear leveling is what gets you in the 1M hour MTBF range...
the MTBF numbers for flash assume that you stay within the endurance limits.
With flash, the weak point is wear of the memory cells, in magnetic disks it's physical components like motors/actuators, heads, etc. Either way, more usage = more wear = shorter lifespan.
Besides, MTBF is not at all a good indication of the expected life of the disk - most drive manufacturers basically cheat and calculate the MTBF based on failures before any components would wear out due to usage (obviously... otherwise they'd be testing the drives for years before shipping them). So it's more of a measure of "defect rate".
An interesting comparison between SSDs and magnetic disks will be their MTBFs vs. average lifespans. I would guess with wear leveling covering the most likely point of failure, SSDs will eventually have a much higher MTBF, but also a much smaller range (I guess "deviation" would be the statistical term) in the average lifespan. It's very possible magnetic disks will have an overall longer average lifespan. But if that SSD lifespan can get into the 5-10 year range, then they are going to become REALLY popular for a lot of uses... (goodbye seek time!)
Since thermodynamics was invented. http://en.wikipedia.org/wiki/Heat
Floppy drive to load RAID drivers?
Try slipstreaming instead. nlite makes it even easier.
"This post is an artistic work of fiction and falsehood. Only a fool would take anything posted here as fact."
IANAL, but I have studied law and I have worked in the litigation field. I have read many letters that have had me wanting to ROFL, and this is in that category. But the best part is also the last bit:
"From there, it should be a short trip to dismissal even if it means getting our clients to mediate Mr. Merchant's positive claims in the absence of an appropriate settlement."
Translation: If you have read this far, you realize that you not only have no case, but that you are entirely out of your league because the standards of evidence in the court system where I have major influence, would procedurally bar you from even entering your case on the docket. Despite this, my client's claims against you are already demonstrated, and our claims will continue to have merit even after your case is dismissed with prejudice (and we have not offered to drop our case.)
This letter is a masterpiece because it manages to hand the plaintiff his ass, in a rather respectful colleague-to-colleague way, while at the same time threatening a counterclaim that could end up with far greater damages than the initial claim!
And the real beauty is that even though the RIAA seems to have withdrawn its claim, the damages from the malice might still hold, if they really want to push it.
Who did they sue? Directors of a Silicon Valley bank? They should do some research before they pull the pin on the hand grenade!
"I would be happy to send the airplane..." (At the plaintiff's expense of course...)
Love it.
-fb Everything not expressly forbidden is now mandatory.
Yes indeed, that's part of all optimisations done in LiveCD. Specially using "union" type of mounting where several filesystems are mounted on the same point (when read, data is pulled from the CD-R, but then, subsequent modification go to RAM disk) and similar solutions.
Also, as I said, the often over-written zone are limited (Linux doesn't write much on disk when it isn't needed) and this makes easier to use such solutions.
RAM used as RAM : and the system could use it even more efficiently.
But swap on RAM disk isn't completly silly... if it's a *hardware* RAM disk :
once you've maxed out all memory slot on your motherboard (say, 2GB DDR-2 dimms in each slot), the only solution to keep adding more memory is to of those "conver RAM dimms into a SATA harddisk" solutions like the Gigabyte's iRAM that was featured on
It won't be as fast and directly usable as the main DDR-2 memory, but it enbales you to add more memory to the system and, if you put your swap file on it, in the end it does extend the maximal memory limit, although in this cas it's *virtual* memory and over a slower connection.
It's kind of "double the number of memory slots, although newer are slower"
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]