Swap partition on 19NM Toggle NAND SSD's

[karteria]

In purchasing a new SSD recently, I discovered something interesting, and perhaps problematic for some Linux users.

SSD NAND is getting thinner, and with that, the maximum P/E cycles or endurance for these chips is reportedly getting shorter. For example, one manufacturer went from a maximum of 5,000 P/E cycles to 3,000 P/E cycles to now about 500 P/E cycles for its newest SSD. The newest SSD utilizes a 19nm Toggle NAND whereas the earlier SSD's use a 24nm MLC NAND. As an aside, apparently there are no endurance measures given in P/E cycles for the new 19nm NAND and instead a new endurance measure for these SSD's is given in Total Bytes Written (e.g., for one new 60 Gb 19nm NAND SSD, the advertised TBW is 32 TB.) Total Bytes Written or TBW as defined as "how much total data can be written to an SSD for a given workload before the drive reaches its endurance limits." Doing the math that comes out to about 500 P/E cycles.

I've read the discussions concerning swap partitions, size and whether they are needed. One point seems to have a consensus, and that is, if you hibernate your computer, you need a swap partition equal to 1x to 2x the size of your RAM. In addition for all SSD's, the manufacturers state not to defragment the drive. It shortens the SSD's useful life.

So here's the question. With the trend of SSD's going to thinner NAND and as such, shorter endurances, whether it's measured in P/E cycles or TBW's, isn't a person who hibernates his computer each day going to shorten the life of his SSD, probably more so than defragmenting from time to time? For example with a swap partition where nearly all the memory is written to daily and with an SSD having a maximum of 500 P/E cycles, it seems that the SSD, at least the memory in the swap partition, is going to be unreliable in about 16 to 18 months. A shortened reliability would be the case as well where the swap partition is being heavily used during normal computing. Are these observations valid or will the "wear leveling" features of the SSD controller prevent this from happening?

I do not use "hybernation" and use "suspend" instead. But I do have a typical swap partition for daily computing on the contingency that it's needed although I personally have never seen the swap partition used. As such, I intentionally bought an SSD with an advertised endurance of 3,000 P/E cycles or about 6 times the reliability based on the advertised TBW of the new 19 nm NAND SSD's.

One last point. I've seen different manufacturers give 3 and 5 year warranties for SSD's that use the same new 19nm NAND chips. This difference in warranties is puzzling to me. I would rather have an SSD that I know will last rather than have one where its lifespan is "iffy" but the manufacturer gives a good warranty.

[hurleydood]

I think your talking about the Kingston SSD V300. It has a 2000 P/E cycle rating.

The formula: Total Bytes Written = P/E Cycle rating x drive size including over provisioning / Write Amplification

Kingston's expects for average user worst case write amplification factor is 4.

So for a Kingston V300 120GB drives TBW forumla is
64TB = 2000 x 128GB / 4

Where as the Kingston HyperX 3K 120GB drives TBW formula is
96TB = 3000 x 128GB / 4

Kingston SSD V300 TBW ratings
60GB: 32TB
120GB: 64TB
240GB: 128TB

Kingston SSD HyperX 3K TBW ratings
120G: 96TB
240G: 192TB
480G: 384TB

The better your write amplification factor the higher host writes you can do. Write amplification of 1 would allow you to write 2000 P/E cycles x 128GB = 256TB. But that would be in a perfect world of 100% sequential writes. Small random writes are what will bring that write amplification up. This WAF of 4 means that for every 1GB the host writes, 4GB is written to the nand flash. I recently purchased a V300 120GB and my current WAF is 2. NAND writes 286GB / host writes 143GB. If my usage continues I will see double the TBW rating.

[catweazel]

None of which means anything important when the devices carry a warranty.

[MtnDewManiac]

One last point. I've seen different manufacturers give 3 and 5 year warranties for SSD's that use the same new 19nm NAND chips. This difference in warranties is puzzling to me.

I suspect that it is because with a great many consumer products (especially, but not limited to, consumer electronics), the products are produced cheaply enough that the company can still make a profit even when a fair number of the devices fail within the warranty period. Sadly, it seems that the days when a great warranty meant that a product could reasonably be expected to last a long time... are long gone. Instead of purchasing a higher-quality product when we spend more, we often are in actuality purchasing the chance to return said product when it fails. It is not unheard of for a company to offer similar products at two or three price-points, with each step up in price giving you a longer warranty - but the actual product, underneath its exterior, is exactly the same. This has been reported in everything from electronics to home water-heaters.

Take laptops, for example. I read a .pdf (disclaimer: the pdf is three years old; current products may be better - or worse) by SquareTrade. It had figures that showed that one in three laptops will fail within three years. Discounting accidents, the malfunction rate alone exceeded 20% at the three-year mark. While the specifics of this report are not relevant here, I believe this to be indicative of a general trend of manufacturers building products that just aren't very durable; a practice which is sustained by the buying public when they settle for it instead of refusing to purchase products that do not have a high durability. (Incidentally, the report showed that ASUS and Toshiba had the best figures with a 15.6% and 15.7% three-year failure rate due to malfunction, while Gateway and HP had the worst with 23.5% and 25.6%, respectively.)

None of which means anything important when the devices carry a warranty.

Respectfully, it means a lot. The companies that sell these products do not warranty the user's data - only the hardware. While the distinction may not mean that much to someone who regularly backs up their data and does not mind the thought of losing the data created/modified since their most recent weekly or even nightly backup (and who has an extra drive at hand to use until the one that failed has been replaced), I suspect that a lot of people are not in the habit of making regular backups.

I am not currently in the market for one of these drives, but I hope to be some day and these discussions do interest me and I will continue to follow along. But as I have no experience with them, I will very likely not post again in this thread.

Hoping to learn,
MDM

[catweazel]

None of which means anything important when the devices carry a warranty.

Respectfully, it means a lot. The companies that sell these products do not warranty the user's data

Nothing abrogates a person's responsibility to make regular backups.

I suspect that a lot of people are not in the habit of making regular backups.

That is very true, but it's still their responsibility and there's nobody to blame except themselves.

I am not currently in the market for one of these drives

I bought two more to add to my current four. They arrive tomorrow, along with a hardware RAID kit. I expect to be thoroughly gobsmacked by the speed. I'm expecting to boot at close to 2.4GB/s throughput. That's bytes (B), not bits (b).

When I consider what has been said in this thread, I guess I have an unfair advantage. The consumer laws where I live are very stringent. Last month I returned two very expensive OCZ 2 SSDs, two and a half years after purchasing them; they had a 3yr warranty. The vendor had to give me a credit for the full purchase amount, no pro rata scam. So, I bought a Ducky mechanical keyboard and 16GB of 2134MHz G.Skill Ripjaws Z RAM with the proceeds.

Cheers.