Avoid erasing NAND as much as possible

[jefor139]

Hi there,

I'm kind of new to SSDs and I try to learn as much as I can about them. I read that article from AnandTech and I wanted to get some more details about that part :

http://superuser.com/questions/220291/what-does-4k-random-write-mean-with-ssds

"Each cell in NAND can only be erased a finite number of times so you want to avoid erasing as much as possible."

What exactly does that mean? Are we talking about erasing files directly from our desktop/explorer? Can we put a number or range on "a finite number of times"?

I'd like to clarify this because it kind of scare me out...

Oh and a quick one: Samsung 840 Pro or OCZ Vertex 4?

Thanks.

 

[Wall Street]

Don't worry. Each cell can be written 3,000 - 10,000 times depending on the technology used. The driver also automatically balance the wear, so for a 100 GB drive you can do about 100 GB * 1,000 erase cycles = 100 TB of erases and writes. This was mostly a problem before SSDs had wear level, for example if you try to boot off a compact flash card. Also, things like the page file, hibernate and system restore result in much more reads/erases than moving documents and installing or deleting programs.

 

[AdamK47]

Wear leveling and over provisioning have significantly reduced the problem. Even to the point where Samsung has started using TLC NAND in their base 840 drives.

As for which drive to get, I'd get the 840 Pro if you can afford the extra cost. Both the 840 Pro and Vertex 4 are fast and reliable drives. You can also get 512GB Vertex 4s with Micron NAND for under $400 now. Pretty cheap SSD-wise. Bought 6 of them just recently for a RAID array.

 

[GarfieldtheCat]

Hi there,

I'm kind of new to SSDs and I try to learn as much as I can about them. I read that article from AnandTech and I wanted to get some more details about that part :

http://superuser.com/questions/220291/what-does-4k-random-write-mean-with-ssds

"Each cell in NAND can only be erased a finite number of times so you want to avoid erasing as much as possible."

What exactly does that mean? Are we talking about erasing files directly from our desktop/explorer? Can we put a number or range on "a finite number of times"?

I'd like to clarify this because it kind of scare me out...

Oh and a quick one: Samsung 840 Pro or OCZ Vertex 4?

Thanks.

Look at this link over at XS. They are stress testing SSD's and yuo can see that they last a long time, so that you should have no problems with NAND wearing out.

XS

And you should probably be asking Sammy 840 Pro vs OCZ Vector (newest model).

But be warned that 840 pro vs vector is liking asking people which is better, VI or Emacs!

 

[AdamK47]

Yes, the AnandTech forums are filled with brand loyal or brand adverse people.

I like to think I'm one of the most brand agnostic computer hardware people on these forums by buying whatever I find best suites my needs. That's really how it should be.

 

[Xpage]

Also note that delays between erases on blocks increases longevity, thus if you are not constantly dumping 1's and 0's onto the drive to fill up it and erase it for cycle testing the blocks last longer.

I have had my Agility 3 for some time, no issues, hell my Agility 1 is on notebook duty @ 60 life left, but it will last 10 years as i don't use the notebook much.

 

[kevinsbane]

Hi there,

I'm kind of new to SSDs and I try to learn as much as I can about them. I read that article from AnandTech and I wanted to get some more details about that part :

http://superuser.com/questions/220291/what-does-4k-random-write-mean-with-ssds

"Each cell in NAND can only be erased a finite number of times so you want to avoid erasing as much as possible."

What exactly does that mean? Are we talking about erasing files directly from our desktop/explorer? Can we put a number or range on "a finite number of times"?

I'd like to clarify this because it kind of scare me out...

Oh and a quick one: Samsung 840 Pro or OCZ Vertex 4?

Thanks.

So I've written 880 GiB of data to my Intel 160 GB drive over a year.

I've used approximately.... 0.1% of my total number of erases. I should start worrying about my drive longevity in the year 3000

 

[groberts101]

So I've written 880 GiB of data to my Intel 160 GB drive over a year.

I've used approximately.... 0.1% of my total number of erases. I should start worrying about my drive longevity in the year 3000

nope. Intel has a bug that remains unfixed in regard to the reported life span.

So, that means that your drive will likely only last till about 2050. :'(

 

[kevinsbane]

nope. Intel has a bug that remains unfixed in regard to the reported life span.

So, that means that your drive will likely only last till about 2050. :'(

You sure? I gots a 320 version, so... 160 * 3000 = 480 000 GiB ~500 TB total write/erase available... I was so looking forward to using it as a working family heirloom for centuries.

 

[Kippa]

I used to install my os quite a bit and every time I did that I did a data wipe on the whole drive. My ssd died after 2 years and 10 months, which wasn't too bad as I did hammer the drive and it was a small 120gb drive. This time I am going to wait for a 1tb ssd and NOT datawipe my drive at all.

 

[blastingcap]

Wear leveling and over provisioning have significantly reduced the problem. Even to the point where Samsung has started using TLC NAND in their base 840 drives.

As for which drive to get, I'd get the 840 Pro if you can afford the extra cost. Both the 840 Pro and Vertex 4 are fast and reliable drives. You can also get 512GB Vertex 4s with Micron NAND for under $400 now. Pretty cheap SSD-wise. Bought 6 of them just recently for a RAID array.

One thing I never bothered to look into but am curious about now: what happens when you have a ton of data that never really gets shifted. This is an extreme example but bear with me:

Say you have a 512GB SSD and put 400GB of movies on it that you never move around or delete. They are just sitting there. After how much time will there be unrecoverable corruption of any of that data? I know people say that flash memory can hold a charge for a long time, years even, but I don't know if that is including or not including any ECC or other anti-bitrot type of stuff, and what the longevity differences are between SLC, MLC, and TLC.

 

[hal2kilo]

One thing I never bothered to look into but am curious about now: what happens when you have a ton of data that never really gets shifted. This is an extreme example but bear with me:

Say you have a 512GB SSD and put 400GB of movies on it that you never move around or delete. They are just sitting there. After how much time will there be unrecoverable corruption of any of that data? I know people say that flash memory can hold a charge for a long time, years even, but I don't know if that is including or not including any ECC or other anti-bitrot type of stuff, and what the longevity differences are between SLC, MLC, and TLC.

Good question. That's the only "unknown" that I slightly worry about with SSDs. I don't lose a wink of sleep thinking I'm going wear out my drive before I replace it because it's become technologically obsolete.

 

[Kippa]

I thought that wearing out of ssds only happens mainly due the process of writing to the ssd, not reading. Does actually reading the cell have any realistic factor on the wearing down of an ssd?

 

[yefi]

I'm not sure if the floating gates leak electrons when the SSD is powered, but the wear-levelling algorithm will be constantly shifting data to equalize out writes across the whole drive, so it shouldn't be an issue.

Unpowered, client drives should retain data for at least one year assuming the stated writes cycles haven't been exceeded: www.jedec.org/sites/default/files/Alvin_Cox [Compatibility Mode]_0.pdf, p26.

 

[blastingcap]

Good question. That's the only "unknown" that I slightly worry about with SSDs. I don't lose a wink of sleep thinking I'm going wear out my drive before I replace it because it's become technologically obsolete.

I found http://hardforum.com/showthread.php?t=1502663 which I'd like confirmation of if true, that powered-on SSDs will move data around to "refresh" even stagnant pools of data like my hypothetical example.

Given the way NAND is constructed, it makes sense that SLC > MLC > TLC when it comes to resisting bit rot, because there would have to be much more decay present before a bit flips with SLC, than with multiple voltage level NAND.

My new question is whether or not little USB flash drives of NAND, or SDHC cards, have ARM controllers capable of shifting around data as well. And even if they did, I worry a little because it's not uncommon for me to have some of my USB flash drives unpowered for over 8 months, and those flash drives may already contain TLC.

 

[Mark R]

I thought that wearing out of ssds only happens mainly due the process of writing to the ssd, not reading. Does actually reading the cell have any realistic factor on the wearing down of an ssd?

Reading doesn't wear an SSD, so much as deplete the floating gate of charge (by a very, very small amount). It's not really an issue for SSDs. Although, it is can be an issue for embedded processors, which run their program directly from NOR flash (rather than loading it into RAM at boot time). With the CPU constantly reading data at high speed, the flash cells containing the program can get "discharged" after months or years.

One thing I never bothered to look into but am curious about now: what happens when you have a ton of data that never really gets shifted. This is an extreme example but bear with me:

This doesn't really happen if the SSD is in regular use. Wear levelling algorithms will detect "static" data and will move it around, so that every cell in the SSD gets used evenly. The better controllers try to keep the amount of movement down, as it wastes write cycles. Typically, the controller will monitor how much difference there is between the minimally worn cells and maximally worn cells and start transferring the static data into new sectors when that happens. However, this usually only happens when the SSD is being written to. Not all drives will do this spontaneously when idle - although enterprise level drives, or drives with enterprise-grade flash (which has much lower data retention guarantees) may do so.

 

[bradley]

Everything old is new again...


http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm/page218

Samsung 840 - FINAL REPORT - DEAD - As of Day 52

Drive Hours: 1235
ASU GiB Written (APPROX): 443,309.73 (432.92 TiB)
Avg MB/s (APPROX): 101.10
MD5: OK

Wear Leveling Count (B1): 3556 raw (1 normalized)

Reallocated blocks (B3,05): 659 (79 normalized)
Failure count (B5, B6): 0 program, 0 erase
Uncorrectable Error Count: 0
ECC Error Rate (C3): 0

Drive is dead and does not respond to anything anymore.

The main concerning thing is that the drive said it did not trigger any smart warnings before dieing! I was not able to get any useful screenshots from ASU or crystaldiskinfo after the drive died (as neither would paint their windows trying to access the drive)

 

[VirtualLarry]

I thought SSDs were supposed to turn "Read-only" when they "died" (expired their flash write cycles). It's disconcerting that the Samsung hard-failed.

 

[jwilliams4200]

I thought SSDs were supposed to turn "Read-only" when they "died" (expired their flash write cycles). It's disconcerting that the Samsung hard-failed.

That's a myth. None of the SSDs that have been written until failure have been reported to enter a read-only mode.

 

[Red Squirrel]

Yes this is one big drawback to SSD tech. SSDs are consumable parts (like toner/ink etc) as opposed to HDDs which are not. Yes HDDs fail, but it's not part of their design. In theory a HDD last's forever.

The write limit on SSDs is VERY high though, so if you put your OS on it and are a general computer user chances are it will last you for 10+ years.

If you run lot of VMs and other high I/O stuff then you want a raid array of spindle drives for that stuff.

 

[Jovec]

If you run lot of VMs and other high I/O stuff then you want a raid array of spindle drives for that stuff.

No, we want to change our way of thinking about how we buy storage and start taking into account drive wear and usage lifetimes while enjoying the massive speed-ups available with SSDs.

 

[bradley]

Friends don't let friends buy TLC NAND Flash for their main drives, otherwise they will continue manufacturing them. Not being read-only is a bad end-life trend.

 

[groberts101]

You sure? I gots a 320 version, so... 160 * 3000 = 480 000 GiB ~500 TB total write/erase available... I was so looking forward to using it as a working family heirloom for centuries.

whoops.. sorry.. my bad. I was off by -10 on my model number there. It's the 330 that supposedly has the MWI needing adjustment with newer firmware.

 

[Red Squirrel]

No, we want to change our way of thinking about how we buy storage and start taking into account drive wear and usage lifetimes while enjoying the massive speed-ups available with SSDs.

I rather not have to worry about how much IO I'm throwing at my storage system and throw everything it can handle at it and know I will still get the same amount of years out of it than if it's idling. Especially in a mission critical environment. Performance isin't everything. If it was, we would all have been using raid 0 from the get go.

 

[yefi]

I rather not have to worry about how much IO I'm throwing at my storage system and throw everything it can handle at it and know I will still get the same amount of years out of it than if it's idling.

If you're throwing so much IO at a modern SSD that you are actually wearing it out before 3 years (taking your definition of consumable goods) then you are going to be wasting a significant amount of your time running the same IO on traditional HDDs.

If the time you've wasted is worth less to you than the price of an SSD, then get a HDD by all means, but I know it wouldn't be for me.