If you're one of the (apparently) many out there scared half to death not over what's residing in your closet at night, but that your recently purchased SSD will wear itself out in no time flat, here's a little reassurance. According to Michael Yang, flash marketing manager at Samsung, a flash device rated at 100,000 write cycles can write 100,000 times "to every single (memory) cell within the device," and of course, it won't continuously write to the same cell over and over thanks to a process dubbed wear leveling. In case you still needed a tad more evidence, he also stated that "a pattern could be perpetually repeated in which a 64GB SSD is completely filled with data, erased, filled again, then erased again every hour of every day for years, and the user still wouldn't reach the theoretical write limit." So, now that we've had that cleared up, why not tell us more about those 250GB MLC-based SSDs that could land before 2009 dawns?
posted February 23rd 2008 9:27PM
Eww, don't even talk about memory paging, my HDD is slow to boot, and I need memory paging. Ugh........
So if you partitioned a 64GB SSD with a 128MB partition for say, a swap partition that gets written to all the time. Will that SSD's "wear leveling" algorithm work? Or does the partition table move across the "cells" too, to have an even wear?
I would imagine it wouldn't move data across the 'partition' - but assuming you have a good amount of RAM you won't really need a swap partition, which are basically defunct in these high-RAM days.
Wear-leveling is "hardware" (drive), where-as the partitioning of your drive is software (OS). If you get a 64 GiB drive, set up a 1 GiB partition, and thrash that one with non-stop writes across all sectors, the drive will not fail more quickly than if you simply create a single 64 GiB partition and do non-stop writes on a single sector.
So, partitioning, swap files (urgh - use true swap partitions) and the likes will NOT take away from the wear-leveling, and will NOT cause the drive to die quicker (unless the wear-level algorithm stinks .. then it'll only last 40 years instead of 4000 years with continuos writes)
/G
Wear leveling isn't done in hardware, it is done in firmware on the SSD. But yes, it will spread the writes over the whole range of the NAND storage, so you can't really wear out one area while the rest stayed untouched.
I have no idea why cbhwr says that this means that this won't wear the SSD out quicker, of course it will. Swap writes will wear out the device faster than if it had no such writes.
Replying to Hardcore:
You don't have to wait till year 19531...
I believe the Copyright protection ceases around 120-140years tops...
checkout http://www.copyright.gov/title17/92chap3.html#302
or http://www.copyright.gov/help/faq
This is fine if all you do is use the web browser and send e-mail. If you do any video or music work, that drive will be dead in a year.
Um, didn't you read the article?
well you would have to work out the math, but I'd wager even with heavy use, the drive will last alot longer than one year. Besides, why would you use an SSD for reading/write huge sequences of video data?? That use case doesn't make any sense. I'm sure it will one day in the future, but for now SSDs are great for use where you have many small non-sequential reads and writes, or where shock tolerance , durability, and/or low power draw considerations are more important than raw sequential speed and capacity.
Considering I recently bought a 160gb 7200rpm sata refurb notebook drive from computer geeks for $89, the price of ssd has a long way to go!
You can get high quality 500GB SATA harddrives for $100....
http://www.newegg.com/Product/Product.aspx?Item=N82E16822136110
@Paul
He said NOTEBOOK drive (i.e. 2.5"). We are all aware of the prices of 3.5" drives.
To Paul Fehr:
Not for a notebook. Think the largest 2.5" nowadays are 320GB, and they're closer to $180. Which is surprisingly cheap. ($180 for 7200rpm 200GB drives. When did mobile drives get this cheap?)
Assume that your drive is > 90% full, and you write 4GB a day (in fact, since your drive is working most of the time, and you know your drive's bus speed, you can estimate how much data your computer actually writes on the drive, which is probably much much more than 4GB/day).
Now assume that the writes are distributed uniformly _on the empty space_ and re-calculate:
~ 1 day for covering the empty space in your card.
~ 100,000 days = ~250 years until you're hit.
if you're running anything interesting on your computer that uses a lot of caching, and your disk is pretty full, or if the wear-leveling algorithm is not that great, expect to be hit (much) before that. But, anyway expect to throw your disk after 3 years because you will get a new laptop..
No. That only happens in the US of A (and a few other first-world countries
)... WE (the ones to the south of the US of A) don't change our computers every 3 years... 5 or 6 years is more realistic
Well dont' forget that these drives also have a large number of spare sectors, so even if your drive is 100% full, it still has (let's say) a 10-15% spare cluster overhead (for cells which perform poorly /near failure and have to be remapped, just like in standard hard drives) to allow remapping of very lightly written sectors to heavily written ones when idle (such as OS files and binaries). Running low on disk space will not affect the availability of sectors for wear leveling.
oh yeah, just consider hibernations (writes most of your 2GB RAM to disk) to re-consider your estimate of 4GB written on the disk daily.. this 4GB number is just too small to be real, when your memory is 2GB.
@Miikun
Where do you get the idea that SSDs have lots of spare sectors? That's true for hard drives, but they have a far, far lower cost per sector. I'm willing to bet SSDs actually have relatively few spare sectors.
cheaper ssd's is great. affordable 250+ gb ssd's are looming soon i hear. i hope sooner than later
these days im more worried about samsung as a company not their products. With all the scandals about the CEO and the employees.
Wow, are their SSD's really only rated at 100k cycles? I'd have thought 500k would be pretty standard by now, seeing that industrial flash is already up in the multiple millions.
That's a great cover story. Except MLC is never rated at 100,000 writes. MLC is 10,000 writes or even as low as 3,000 writes!
My company ran test on NAND storage devices, in reality, if you write to them continuously, you'll reach the end of their stated lives in a couple months.
Think of a 64GB SSD with a 32MB/sec write speed. It will take 2,000 seconds (about a half hour) to write it all once. That means it only takes 5,000 hours (about 8 months) to reach the end of its rated life.
The devices don't immediately fall flat when they reach their stated life, but you're living on borrowed time at that point.
So, is it only write's that'll fail once the sectors been worn down? I mean, you'd still be able to read from the disk, right? (I know next to nothing about SS media.)
Actually NAND has a thing called "read disturb". That is, it very slightly damages the data in the sector every time you read it. If you read a sector enough (between writes to it), the sector starts to lose its data and then you need to rewrite it.
So yeah, even reading an SSD wears it out too!
But this effect is rather minor. You'll get thousands of reads (it could be more like tens or hundreds of thousands for all I know) before you have to rewrite the sector, and then of course you have wear leveling and the 3,000 (or more) write cycles in a sector before it begins to die.
So realistically, you're really looking at reading the entire device out minimum something like 10,000,000 times before you could wear it out just by reading. And it's more likely more like 1,000,000,000 (a billion).
So everytime you connect your iPod Touch / iPhone and it trigers a sync - you're killing it slowly...
Here is my Math lesson for today:
Take for example a 64GB device, with a 100,000 cycle write limit: in effect you have a theoretical limit of storing data up to 6,400,000GB (64 x 100,000).
Lets say your laptop/device can actually last past 20 years of usage.. then you have 7300 days (20 x 365) to use up 6,400,000GB of storage by reading from and writing to the drive. Please note that by using the day, we are supposedly talking about using the laptop/device for 24 hr sessions over this period.
So my math is 6,400,000GB / 7300 days = ~877GB per day of usage for 20 years - assuming you have the device on for 24 hrs a day, writing and reading data to it every second of the day.
This is simplistic because it does not take into account possible manufacturing/material defects or poor OS implementation that creates extraneous write cycles etc.
I think in the end your SSD storage should be adequate based on the article's stated 100,000 cycles.
aren't you guys that are calculating forgetting something. if you have a 64GB SSD and have your OS, programs, and files on it, you are left with much less. on my 80gb drive, i only have about 10gb free that i can use daily. so assuming you are writing 3GB a day and with all of the logging/caching, you would be writing that to the same 7-10GB of free space every time. this would alter your calculations considerably
maybe i'm missing something about "wear leveling" but it I can't imagine this process will move around your permanent data.
Ravi,
What OS are you running? Is it really taking up nearly 70GB?
Look for 128G to start coming down in price very soon! I think there's so much competition in the marketplace that the small guys are going to start dropping out soon. Adtron got gobbled up by Smart Modular Technologies recently
Nothing worse than misinformation by armchair experts. But I guess you get that no matter where you go.
Yeah, but the problem is for 2-bit MLC SSDs the write limit is 10,000 per cell not 100,000.
For 3-bit it's close to 5,000.
I'm amazed at how much attention is payed towards the longevity of the SSDs. I mean, how old is the rotating drive in your laptop right now? How confident are you that it won't quit within the next year? While I believe that many of the current customers of SSD are currently paying the premium because of the high reliability, the amount of discussion from people who don't care about reliability is amazing. Also, with the capacities doubling as fast as they have been, you're probably going to get a new drive anyways long before you wear your old SSD out. With or without wear levelling!
SSDs inside have "erase block" which is pretty much the smallest block they can read/write. When you try to write something small (like a 512 byte sector), the SSD firmware will read the erase block where the sector lives, reallocate it somewhere else ("wear levelling") and write it out with the 512 byte sector replaced. So, your small write suddenly got much bigger. Erase blocks are between 128Kb and 512Kb in size. So, naively, with an erase block of 512Kb, it would take 64*1024*1024*1024/(512*1024)= 128K rewrites of that 512b sector to write every erase block on the disk. At 10K write-before-failure cycles, that's 1.25G rewrites. Systems that have databases or have time-of-last-access on files (Linux/Windows) will routinely write out small chunks of data all the time. Assume writing one sector/second (best case) and thats 1K years.
To further confuse things, the entire drive is NOT used for this levelling; the firmware keeps track of erase blocks that have data vs ones unwritten or "spared". If you write live data to fill the disk to 90%, that 1K years goes to 100 years. Still long, but it's a best case.
Two more confusions:
1. Most applications will write out files in their entirety. This means that the firmware can take advantage of appending and not have to rewrite erase blocks in their entirety.
2. Some file systems are able to tell the SSD firmware when blocks are truly free and can be used for sparing/levelling