Another SSD question.

[Lemon law]

As I read up on SSD's, the more I read the more confused I get.

But please help me understand the following question, with flash memory of all kinds limited to only a fixed number of read write ram state change cycles before they fail, why isn't rapid failure inevitable with a SSD?

On one hand we would think a SSD would be ideal for reading in things like an OS, write the OS once, and it never changes. But given the fact that there are windows updates, the OS becomes anything but a static write once entity. Meanwhile web browsers, and with almost all other programs getting regular updates, old program storage data is always getting deleted and new data must be written. All of which require the ram States of cells to change.

Then there is normal computer activity, and with window XP handling much of that with a HDD page file instead of ram, why does that not mean that even 30,000 ram state changes will occur in a matter of a month rather than a million hours mean time before failure as advertised? Sure Vista and Win 7 handle ram better than Win XP, but still they too require a ever changing set of ram states.

Just take this post I am making on anand tech, its not huge, but as soon as I make this post it will be stored in history for awhile, and at some point, depending on how long I keep temporary files, it will become available for new data. Why does that not hammer a SSD to death?

 

[darkewaffle]

In principal, they will eventually die. In actuality, any normal user will probably upgrade before it wears out or more often never even reach that threshold. The relatively unrefined technology behind current SSDs rates them for trillions of operations and millions of hours of activity already. Further, controllers use compression technology so that not every IO is an individual action, but in many cases a 'pack' of operations, reducing the load and required space.

 

[flamenko]

There are several things that happen to make sure your ssd lasts a very long time, the first of which is wear leveling. Wear leveling is the firmware moving files around to ensure all cells are written to evenly. The next 2 are very similar and are TRIM and Garbage Collection. They clean the invalid data from the drive and work in tandem with wear levelling to ensure that the drive works at peak performance and lasts a long time.

there is another unique feature of the SSD which is completed by GC and wear levelling. They are capable of determining if any cells become bad at which time they are swapped off (remapped) with the over provisioning and you never know it. the drive does not report bad sectors and a lesser capacity.

Oh as well... over provisioning allows for a more even and lesser write level to be completed through the addition of OP which is not accessible by the user.

 

[LokutusofBorg]

As I read up on SSD's, the more I read the more confused I get.

But please help me understand the following question, with flash memory of all kinds limited to only a fixed number of read write ram state change cycles before they fail, why isn't rapid failure inevitable with a SSD?

On one hand we would think a SSD would be ideal for reading in things like an OS, write the OS once, and it never changes. But given the fact that there are windows updates, the OS becomes anything but a static write once entity. Meanwhile web browsers, and with almost all other programs getting regular updates, old program storage data is always getting deleted and new data must be written. All of which require the ram States of cells to change.

Then there is normal computer activity, and with window XP handling much of that with a HDD page file instead of ram, why does that not mean that even 30,000 ram state changes will occur in a matter of a month rather than a million hours mean time before failure as advertised? Sure Vista and Win 7 handle ram better than Win XP, but still they too require a ever changing set of ram states.

Just take this post I am making on anand tech, its not huge, but as soon as I make this post it will be stored in history for awhile, and at some point, depending on how long I keep temporary files, it will become available for new data. Why does that not hammer a SSD to death?

You have some fundamental misunderstandings here. I suggest you go read Anand's original SSD articles (look up the anthology) a few times over until you get the basics down if you really want to understand this technology.

NAND cells have a write threshold, not read/write as you state. They can only take so many writes. With wear-leveling, data is moved around when it's modified so the write wear is spread across the whole drive. When a NAND cell has reached the write threshold the data can continue to be read from that cell just fine.

Despite the fact that Windows (and most programs, and really most regular user data for that matter) does see changes over time, it's relatively minor. 100MB in patches once a month on a multi-gigabyte installation? That's nothing.

I have no idea where you would get the idea that Windows XP uses the page file for "much of" its normal activity unless your computer has 128GB of RAM or something.

You seem really hung up on the idea that "NAND wears out". It's a very minor issue. So minor that I feel safe in saying you will never see a drive wear out on you.