SSD Reliability...is it better than HDD? (Answers!?)

[EarthDog]

I seemed to have ran out thanks. THANKS! Was hoping you would chime in here..Im struggling a bit!

 

[Psycogeec]

Thats actually the main point of this thread... may want to reread the highlighted portions. That sure was MY thinking too before this...


@ Pyshco - If I see SDD one more time... ... SSD!

I would bet my life that most applications would open faster on an SSD vs 2 of any mechanical drive in Raid 0. :

and i would bet my life that some of thier tests systems are way slower than your system too even after they put in a SSD.

I was just indicating that a person can do some good computer managment , and get a lot of nice speed. if the new method to fix program belotion is to throw even more money at it. it will become just like the OS, the more we got the more they will waste it.
i can hear ADOBEE now tell me that the reason "Creative suites 2025" is so slow loading, is because i dont have a SSD, not because it has 4gig of newly added fluff

I Will go back and correct all the SolidDiskDevice referances
.

 

[johan851]

Much Ado about nothing Pyscho. What I think you are attempting to describe is normal use and would of course be included with the data written. There is no need to eliminate those types of writes (and Page File) as I believe those are small files where SSD's actually excel in.

This is exactly the right way to think about it. How often something writes to disk isn't really of concern - the primary metric is total data written per day. Since the SSD handles distributing the write load over all of the NAND cells, the only way to wear out single cells is by writing a very large volume of data to the drive.

 

[EarthDog]

LOL, you dont have to do that, just having fun...(SSD v SDD)!!!

While you can benefit from tweaking this or that to your system to get a lot of nice speed, there is no amount of tweaking one can do to that will make a standard mechanical drive perform like an SSD.

 

[Mr Alpha]

Also, another interesting fact: Due to the way the physics of flash wear out works the warmer the SSD is the more p/e cycles it will endure. So unlike much other hardware an SSD lasts long the warmer it is (up to a point, at least).

Don't go cooking your SSDs though, they get more volatile as they heat up, that is to say they loose the data stored on them faster.

 

[cw823]

Stickied.

 

[johan851]

Also, another interesting fact: Due to the way the physics of flash wear out works the warmer the SSD is the more p/e cycles it will endure. So unlike much other hardware an SSD lasts long the warmer it is (up to a point, at least).

Don't go cooking your SSDs though, they get more volatile as they heat up, that is to say they loose the data stored on them faster.

I'll have to turn off the intake fan and build a 12v electric blanket.

 

[Psycogeec]

Heat is a non issue really with these drives. Even when beating on them for hours at a time they are barely warm to the touch.

For some devices, the user has been observed complaining, that a device "Gets really hot" because they can feel it. Those devices are usually better cooled, extracting internal heat to the outside quickly.
I am of the other school, where i recognise that a Tablet with an alum back heat synced off the processor is not nessisarily "hotter" but is probably the only one that is cooled

I dont know if the controller is in there burning up, i am trying to indicate that they didnt do squat, and some failures reported are not write cycles at all, but "board" failures.

being a modder (often for stupid reasons) i would want to peek inside and see how it is doing. we have had enough failures over the ages due to them sealing up things that process a lot of data.
.

 

[EarthDog]

For some devices, the user has been observed complaining, that a device "Gets really hot" because they can feel it. Those devices are usually better cooled, extracting internal heat to the outside quickly.
I am of the other school, where i recognise that a Tablet with an alum back heat synced off the processor is not nessisarily "hotter" but is probably the only one that is cooled

I dont know if the controller is in there burning up, i am trying to indicate that they didnt do squat, and some failures reported are not write cycles at all, but "board" failures.

being a modder (often for stupid reasons) i would want to peek inside and see how it is doing. we have had enough failures over the ages due to them sealing up things that process a lot of data.
.

Also, another interesting fact: Due to the way the physics of flash wear out works the warmer the SSD is the more p/e cycles it will endure. So unlike much other hardware an SSD lasts long the warmer it is (up to a point, at least).

Don't go cooking your SSDs though, they get more volatile as they heat up, that is to say they lose the data stored on them faster.

Not sure what to tell you there...

The warmer it is on the outside, the warmer it has to be on the inside to get that heat on the heat sink!!

 

[Psycogeec]

The warmer it is on the outside, the warmer it has to be on the inside to get that heat on the heat sink!!

yes over enough time, if a person worked it hard , and no heat ever departed it , then no heat is being produced. and its another miracle of modern science.

but if a part was cooking everytime it got revved up, and there is minimal conduction to the outside , then the buildup heat could still spread across the whole case and eventually depart. That doesnt stop the part from cooking, because they sealed up the thing inside a case, when they could have cased the thing in its own heat sync.

then we go next door, and some Lab is turning the processor into cooler it just doesnt make sence, when all it would take is case conduction for the hot stuff, which they are finnaly now doing in a Few tablets, and phones.

oops, i hope i said SOME, because many of them are designing the case/cooling better.
.

 

[neo668]

When you have a SSD that is troublefree you will swear that it is more reliable than a HDD. Once you get one that fails on you you will curse it to no end and tell everyone to stick to HDDs.

 

[Ivy]

There is still to less experience in the long run, its still to new technology. There are 20 year old and working HDDs around who have been running almost every day. Now 20 year after i want to see such a SSD aswell, future will tell. They say that there is no moving part, thats indeed true, however, there isnt only the mechanical wear level, there is aswell a electronical caused wear level which is affecting all nand and even the controller chip. That wear level sometimes seems to cause a more huge wear than ever expected (previously it was stated that after only 1 year there is already a higher failure rate due to wear), so i truly wonder how it works out in the long rung. Especially, it was already mentioned, that the manufacturers arnt trying to build with max quality, they try to build in the most effective way and thats not the way which is providing highest durability. Additionally people forgot to say that a SSD is extremely sensitive to power issues. In term it gets a bad power supply (or MB), it can aswell instantly burn down, while a usual HDD may handle much higher power instability, so flashing chips is indeed a sensitive progress and will always be. A HDD is working by magnetizing a plate, that progress is lesser vulnerable to power issues. With that said, the endurance of a SSD surely is very hard to predict.

So, finally, SSDs are not more reliable than HDDs, at least not at the current time. So i would recommend to get a mix of SSDs and HDDs, both got its weakness and strengths but none of those will have higher durability.

 

[rrohbeck]

Reliability of enterprise class SSDs today is comparable to some not-very-reliable HDs. That is, the failure rate is of the same order of magnitude, but worse than HDs on average.
That's absolutely no surprise if you look at how many transistors, ICs, solder joints etc are in a SSD compared to a disk drive and that it's usually not the head/media that causes a HD to fail but the electronics (unless you go back 10+ years.)
Conversely, Flash wear is a non-issue, at least in our application (often 24/7 random IO with varying block size, maybe 10% to 20% duty cycle.) We monitor the SMART wear parameters and I have yet to see a drive reach 98 (I have seen a couple with 99.) Our worst case expected life time was 5 years when we did the design but today we don't expect any significant number to wear out, if any. The SSDs that failed in the field were all normal random failures - none due to wear.
This is from a couple thousand in the field and about two years.

 

[diaz]

What I found curios about the Tom's hardware article is the claimed 10GB/day of writes since I've been doing way more than that. Now, I know I'm not an average computer user, but I'm surprised I would be so much outside the norm.

Still, I'm not in the least bit worried. Firstly, it is a 300GB SSD, so there is a lot of flash to wear out, and secondly, what everybody forgets is that the 3000-5000 p/e cycles of modern 25nm flash is the minimum number of cycles it can handle. The average can be significantly higher than that.

100G a day writes is pretty high.. I don't even dowload that in a month. Where do all the writes come from?

 

[Mr Alpha]

100G a day writes is pretty high.. I don't even dowload that in a month. Where do all the writes come from?

Well downloads are usually a small fraction of the writes anybody actually does. A big part is various temporary data written by programs. A big culprit is Carbonite, which logs all the changes to any files it is keeping backed up, and can produce 10-20GB writes on a day I use the computer a lot. Firefox's caching, and all the rest usually does only 1-2 GB in 24h.

The biggest culprit is still the system process, which includes various Windows functionality, like pagefile. I've since doubled my RAM to 16GB and I seldom anymore actually hit 100GB per day.

Another thing was that I spent some time playing with Bitcoin. It produces a constant slow stream of 1byte writes. But since you can't write 1 byte at the time it gets padded out quite a bit, even with write caching. This can produce a lot of data to write.

For the record: currently I'm at 19.71TB total data written.

 

[diaz]

Is there a way to track which programs use that much write?

 

[Mr Alpha]

Is there a way to track which programs use that much write?

Process Monitor has a column tracking disk writes. It only tracks from when it was started, and it is pre-write-cache so it is not perfectly accurate.

 

[Ivy]

At current knowledge, the reliability for SSD is lower, but there is certainly a huge difference between the drives. SF based drives are for example much less reliable than others, especially those with newest controllers.

While the NAND is rarely responsible for, its usualy dead controller or a complete firmware shutdown, so it cant access NAND anymore. Thats the facts we are facing. So for data security, a HDD is still superior, unless they can increase reliability. There isnt truly such a thing such as dead NAND, most of them just wont be able to flash anymore, so the writtes are lost. But it still would be able to be read and backup the data on another drive. Even if some NAND is lost, it usualy got spare NANDs to backup dead parts with, so NAND is the lowest of all issues and not the highest. The truth is, the tech just isnt mature enough, they still have endless issues to deal with, and the new SF controller tech is just to advanced to fully tackle, they took to many step at once by building it. Ofc its sad because the most expensive parts of a SSD are the NANDs but the cheap parts finally can render the most expensive parts useless, it just wont work anymore, and all the NANDs... wasted.

Ultimately, people should stop caring about NAND overwritte. Its close to no issue because most likely it will die for another reason. I know, no one is telling those facts, thats why i do it.

The reason those manufacturers worry about "not to use to much benches on a SSD" isnt the wear of NAND, forget that fairy tale!! Just wipe it out of mind because its just foolish to believe it, we know the maths and a 120 GB SSD for example, at correct wear levels can handle up to 1200 TB of overwrittes (calculated from 10 000 overwrittes each cell, so pretty much 120 GB multiplied by 10 000), thats so much that not even a daily benchmark would shorten the lifetime to a amount lesser than 10 years. If we take into account that every benchmark takes 20 GB data (a raw estimation, it can be very different), resulting in 600 GB a month, 7.2 TB a year, 72 TB in 10 years, and we cant even nearly cause the NAND to wear out, in term the wear leveling was done proper.

So whats the issue behind? They simply worry to get a dead, burned up, controller, to get malfunctions on software and what else... thats the truth and nothing else.

I had endless malfunctions, even corrupted data on a SF drive that came with extremely expensive super high tech Toshiba 32 nm NAND, made in Japan. I did gently ask the NAND, whats up? Why is there corrupt data? The NAND was laughting at me and telling "we are at 100% health and can handle 9999 more overwrittes, oh dear", then they are pointing at the controller "he is so mean, he is sending us weird stuff all of a sudden, and that weird firmware too". Alright, finally got some enlightenment.

 

[rrohbeck]

Agreed. That's why I use my SSD only for caching. It's almost as fast as the real thing and it covers the entire disk capacity.
Ideally the caching code should reside in the file system driver - it knows best about what data is hot and random so it won't waste cache space on sequential accesses. Can't wait for btrfs to get that feature.

 

[Cloudarok]

Usb 3.0

Was It A Fail? My MB and USB are both 3.0 but yet I can hardly notice a difference if I used a usb 2.0?

thnx for future replies.


_________________________________________
CPU -AMD FX @ 4.6Ghz 1.443v|
MB - Asus Crosshair V Formula|
GPU - Gtx 470x2|
RAM -16GB G.SKill Sniper 1600MHz|
SDD - OCZ 60GBx2 Solid3 -SSD Raid 0|
PSU - Coolermaster 1000w|
Cooler - Coolermaster V6 Gt|
Case - Thermaltake Level 10gt