Am Freitag, 12. Februar 2016, 11:48:10 CET schrieb herve: > On 11/02/2016 22:25, Martin Steigerwald wrote: > > I don´t think it is proven that SSDs fail earlier than HDDs. So far none of > > the SSDs I use have failed and one is almost 5 years, still thinking about > > itself that it is actually almost new according to SMART data. And the only > > reason it isn´t older is that it is the first SSD I got. I expect it to live > > on for years to come. > > > > So do you have any factual data to prove your claim? > > > > So far I didn´t see any proof that SSDs fail more often or earlier than > > harddisks. > > > > Thanks, > > There can be any proof or factual data yet because this is a very young > technology. But what we know is that there is a limited number of write > cycle. Of course hdd can fail with mechanical problems ssd don't have. > From what i read, ssd are much much faster, but, at the time i read, > hdd can be more reliable. And the cost of reliable ssd are no comparable > with hdd.
Heise and others tried to destroy SSDs by writing to it. And they found out that it is not easy to do so. The Intel SSD 320 is rated for 20 GiB of writes a *day* for a minimum of 5 years usable life-time. With usable they mean: It remains fast. Now I ask you: Do you have 20 GiB each day of writes? That is 20 * 365 = 7300 GiB or about 7,12 TiB. Now this Intel SSD is almost 5 years old. That will make almost 7,12*5 = 35,6 TiB. I do have: 241 Host_Writes_32MiB 0x0032 100 100 000 Old_age Always - 1000827 Thats 1000827 * 32 MiB = about 30 TiB. I admit thats pretty close and sometimes I wonder about Plasma 5 desktop with KDEPIM + Akonadi + Baloo and before that Nepomuk just writing a tiny bit much onto the SSD. But I also compile KDE Frameworks + KDEPIM + Linux Kernel from source and that adds more I/O. But my main point is: The SSD wrote about 30 TiB. Yet, according to smartctl -a it thinks: 233 Media_Wearout_Indicator 0x0032 098 098 000 Old_age Always - 0 this media wearout indicator which according to Intel is related to write cycle consumption start at the value 100. Now it is 098. So the SSD considers itself to be almost new. Intel writes in a PDF about the media wearout indicator[1]: > The value of the E9 SMART attribute is defined by Intel as the media wearout > indicator of an Intel SSD. When the value reads 1, this indicates that the > SSD is reaching the predetermined maximum media wearout limit. Intel > recommends that you replace the SSD at this point or back up the SSD to > help prevent the loss of data. And in addition this: > E9 SMART Attribute > The E9 SMART attribute reports a > normalized value of 100 (when the SSD > is brand new out of the factory) and > declines to a minimum value of 1. > The normalized value decreases as the > NAND erase cycles increase from 0 to the > maximum-rated cycles. Once the > normalized value reaches 1, the number > will not decrease, although it is likely that > additional wear can be put on the device. > Figure 1 shows how the value of the E9 > attribute decreases over time in a sample > usage model with a consistent workload. and the figure shows a linear scale. So here is facts. Please stop spreading FUD about SSDs in general. This Intel SSD 320 did 30 TiB of host writes and it is as fast as ever. There are a tiny bit of erase failures: 172 Erase_Fail_Count 0x0032 100 100 000 Old_age Always - 169 it raised to 169 I think within the first or second year of the SSD and didn´t raise then anymore (I can look it up if desired. So I bet thats just some cells that weren´t up to par with the rest. Still the normalized value was 100 and is 100, so I don´t care either. So a *good* SSD is very reliable. So just don´t buy cheap crap and read some reviews. Good SSDs are built to last. I am not sure about the current generation, but from what I read here even SSDs with TLC chips are designed to last long if not longer than the ones with MLC chips I have here. I wondered a bit about the Crucial m500 480 GiB mSATA SSD. As it reports: 202 Percent_Lifetime_Used 0x0031 091 091 000 Pre-fail Offline - 9 But with 91% of lifetime left and it almost 2 years old as well, I can extrapolate the lifetime to nine times * 2 years that 18 years. Quite good if you ask me! Even it is just lasts for 8 more years thats fine. So while I still recommend to leave some free space (I did) and use noatime and so, I do think except for crazy server workloads you can only destroy SSDs by writing too much if you actually do this with full intention. It won´t happen on normal use, it won´t even happen on heavy use like the crazy things I do on this machine (like storing a maildir with 1,7 million single mail files and compiling stuff all the time). Of course, make regular backups in any case. [1] Monitoring Media Wearout Levels on Intel Solid-State Drives http://download.intel.com/design/flash/nand/325551.pdf > What i know is that usb key that are sold to be 'almost' eternal, > (relatively to computer evolution) dies very often, too 'young'. I can > understand someone wanting to preserve it's lifetime avoiding too much > write cycle, for example system in ssd, /var and /home in hdd. i would > do the same. Your experience is your experience. it can't be an > everybody rule. None of my USB sticks broke so far. I had some takeMS SD card breaking, the electronic broke and it costed me about 150 Euro to have data recovered from it. That teached me two things: 1) also backup my SD cards 2) use quality SD cards. Ciao, -- Martin
