Long term reliability of SSDs

yea but theres still two types out there. the SLC is projected at being 10 times (a whole order of magnitude) better than MLC.

most consumers still have MLC cards cause they are more affordable for the given size of memory. These don't last as long.

Servers have been using SLCs.

but yea they are being field tested in effect on consumers as we speak. and yes I do mean CONSUMER grade products and not the technology concept itself which has been around for some time as many have continuously pointed as blanket statement of its reliability . so i don't agree with blanket statements that they are reliable merely cause of the fact they are 'solid state'. rather i want to know how realistic they are given the known issue they have with limited write - delete

 

I've been using SSD's for about 2 years now and I haven't had any of mine fail on me.

 

were you using the same one though. also is it MLC or SLC it looks like you have a 256gb SSD frm your sig? Thats pretty new.

 

No, there is not. While a badly built piece of electronics can and will break (witness Nvidia's GPUs), one that is well done is orders of magnitude more reliable than a precise mechanical device which operates at high speeds. With hard drives, even if you buy state-of-the-art stuff, some of the best on the market, you still have to worry about constantly backing it up because the failure rate is unacceptably high and the fraction of times you can correctly predict that a drive will fail before it does is unacceptably low. With SSDs, this just isn't he case.

This will not bother an SSD. They are not damaged by Windows constantly writing stuff to them and your temp files won't hurt them.

If you update the matrix every once in a while, the SSD won't care. You'd have to update every element of the array separately for it to make a difference and nobody does that (even with HDDs) because, as I said, this incurs a monstrous performance hit.

It is certainly possible to deliberately perform enough write cycles to wear out an SSD, but there are few devices one cannot damage through deliberate misuse. If you really need to do the kind of computing that requires more RAM than can be installed on a typical workstation-class machine (last time I checked this was 32GB, it is probably more by now) and you'd like to to do it on an SSD, you should probably invest in one that is state-of-the-art (i.e. SLC) rather than consumer class. Most people will never run into a situation where they care about this though.

 

I'm talking about in general for all devices. people seem to think solid state is better than mechanical. but solid state devices operate on physical phenomena that are itself very sensitive to enviornmental factors so reliability is an issue. e.g. some solid state devices rely on electron tunnelling through an epitaxial layer of dielectric material, but then theres diffusion phenomena always and so over extended use and also high temperature, the dielectric layer becomes no longer due to diffusion. this is only one possible problem of many variety of solid state devices. there a huge debate among physics communities about the social misconception of 'solid state' being superior. they know its not because it depends on perspective.

Its not about 'deliberately misuing' SSDs if thats a legitamite specification you need. We're not talking about doing a 3x3 matrix or a 10x10. im wondering how reliable they are for say computationally intensive tasks involving millions by millions matrix or something ridiculous. for example if people want to use SSD for say doing finite element simulation on a workstation, and they basically leave that thing running all day conitnuously, how reliable can they be. most programs write solutions to disk memory for storage of matrix solutions. they may not store it in physical memory which they need for saying running the repetitive calculations etc.

 

The appeal of SSD isn't really in their long term reliability compared to mechanical HDDs. No storage device is stable for long periods of time (unfortunately), thats why we backup. I think the general consensus is that SSDs are at least in the same range as mechanical HDDs in terms of long term reliability, with the possibility of being more reliable (only time will tell). The real benefit of SSDs (besides heat output, battery life, and speed) is that they are much less vulnerable to physical force. So if you drop your laptop, an SSD will mos tlikely be fine, your mechanical drive has a good chance of being messed up.

 

Most of my SSD's that I own are MlC drives but I do have a 32 GB SLC and a 64 GB SLC drive and they are over 2 years old and still going strong.

The two drives are fairly new in my sig but they are using a proven SSD controller from Samsung.

 

i see. So the 2 years going strong is an SLC then and not the lower performing MLC?

 

There is very little difference performance wise between mlc and slc in typical usage. Servers is where slc shines.

Yes slc drives do life longer, but who really cares. By the time a mlc drive actually dies, you would have already upgraded.

 

sorry I mean lower in reliability. MLCs have lower life. I'm not so worried about which gives me a faster computing experience of MLC vs SLC. I'm worried about longevity of the drive. It depends how you use it. If I am running calculations that are iterative and plan on running them all day then it may be an issue cause that may involve many rewrites.

if only i hear reaffirming news that workstations reliably use SSDs for heavy computational work then thats fine. But even if they do they run on SLC SSDs so they are 10 times better anyway. So I would have 10 times the fear if I get the MLC (which is the one offerred by the compnay)

 

If you're so worried, then don't get an ssd. It's that simple

 

I know. Annoying as it is, but I want one too. In anycase I want to know how good they are anywya.

 

I would think a mechanical drive would have a certain number of write/erase cycles before a sector is likely to go bye-bye (can anyone confirm this?). I think you should be just as worried that a mechanical drive will fail, i would say we've all experienced that happen at one time or another.

 

You're not going to get a definitive answer until SSDs actually start dieing. No one knows for sure.

Lets just say you have reached the max amount of writes on every section of the hard drive, is the data gone at that point or you just can't write anything new?

To me, it seems like the data should still be accessible (no limit to reads) even when you can't write to the drive any more. So the worst that could happen is you have to remove the data from the drive and it becomes a paper weight.

The Know Your SSD thread on here claims lifespan of 7-10 years for MLC SSDs. That seems pretty comparable to the lifespan of hard drives so I'm not sure why people make a big deal about SSD lifespan. By the time it fails, must of us will be on a new computer anyway.

 

I've had four SSD's for 2 1/2 years. Two are SLC and two are MLC. None of them have had a sinle issue ever.

Some of you are saying the jury is still out. Are you kidding me?

The only drawbacks to SSD's are their cost per unit of storage and their capacity.

Every day both of those things are improving, and every day they are being offered in computers from more manufacturers.

I find it very funny that so many people are still bucking the tide of superior technology.

At least some of the buggy manufacturers switched over to combustion engine horseless buggies. The big HDD companies really need to follow that example if they want to be around for long.

SSD usage is gaining market share every day and HDD market share is getting smaller every day. That's not going to change.

 

It's a published fact that HDD makers design their products with a 5 year lifespan. Drives will frequently last longer than that, if they experience lighter load, but this is the cutoff that the manufacturers aim for.

For SSDs, just do the math: take a 64GB MLC SSD, with 10,000 writes per cell. That means the entire drive will last (on average) for 1.25 trillion writes (512 bytes per cell). That's a total write volume of 640 terabytes, and given a sustained sequential write speed of 200MB/sec it would take 889 hours of sustained, non-stop writing to hit that limit. That's a little over 37 days of continuous writing.

In real use, the proportion of sequential writes to random writes is pretty low, most activity is random writes, and most computers don't write continuously for 37 days straight. If you take a typical random write speed of ~10MB/sec then you're looking at 740 days of continuous, 24 hour non-stop lifetime. Something over 2 years. If your usage is more typical, only 8 hours a day, then now you're at 2220 days, well over 6 years. In practice you will have upgraded to a newer drive long before the wear limit is reached. With a larger capacity drive the numbers increase proportionally; my 256GB SSD would have an expected wear lifetime of 8880 days with the same usage pattern - 25 years.

Whether or not a drive will last so long has other factors of course, but the wear limit is pretty much irrelevant.

 

Yea that is pretty bad for say a server or computationally intensive environments that in worst case scenario will be running calculations all day. if you have to replace the drives ever month thats pretty bad.

 

Completely agree. I have one of supertalent's earliest MLC SSD's and it still runs fine. Costs since then have halved or more. In a couple of years SSD's will overtake the HDD market no problem.
I think it is natural for people to "buck" the tide of superior technology as they must justify their own purchases that quickly become outdated, take the dual vs quad core debates going on for example lol. We all experience it

 

but why? would be a better question, HDDs have more then one cause of death, I have 2 hard drives hooked up right now, ones dangling, and one is on the carpet, both are over 7 years old, one im sure is over 10.

just a few hard drives diying out before 2 years doesnt mean ALL HDDs die before 2 years

 

Very few sub-millimeter scale devices (electronic or mechanical) are fond of "environmental factors". The point is that SSDs are much better at handling them than HDDs. That's why the military (which needs to be able to operate in all environments) used them even back when they were ridiculously expensive.

Really? I'm a physicist (though I specialize in high energy physics rather than solid state) and I haven't heard of any such debate. It's pretty clear to me that they're superior -- the only problems are that they're too small and too expensive.

I don't know why you want to deal with millions by millions matrices (when such things arise, there's usually a more clever way to do things), but if you really want to do that, you're going to need a custom solution rather than consumer grade SSDs. A 1 million by 1 million matrix is going to have 1 trillion (10^12) entries. Even if each entry is merely a short integer (2 bytes in size), storing something like that will require 2 TB of space. This is not what SSDs (which are relatively small in size) are meant to be used for -- their purpose is to speed up boot and application load times as well as be more robust than HDDs.

In fact, you don't want to do this with hard drives either because such heavy usage will wear them out. You probably want something like dCache which spreads the load.

 

As the name suggests, a 'Solid state' device is any device that resides in a condensed phase during operation. It is technically referred to devices that operate on quantum theory and 'solid state physics' principles, first and foremost (although if you think about it what doesn't). Solid state devices _in_a_computer_ will include your processor, RAM, SSD, etc. They are all solid state devices. Solid state devices are NOT limited to computers. There are also solid state lasers, solar panels, etc. All are also solid state devices.

All devices whether mechanical (i'm more inclined to use the term 'non-solid state') or solid state are influenced by environmental factors. Some more than others. Some more dependent/sensitive on certain specific variables. On the issue of reliability (how long a device survives under operation), it is entirely dependent on operating conditions of a given device.

On the issue of whether solid state or mechanical devices are better? Well, not all of the possible functions a device may desire/require can be achieved by mechanical means. Likewise not all functions can be achieved with blanket application of solid state physics principles. For example, gas/chemical lasers are likely to continue being much more powerful than any solid state laser solution in the forseeable future because of unavoidable theoretical and physical limitations of the materials. Other similar examples of advantages to non-solid state versus solid state devices can be drawn. Also to contrary too, to prove more benefit of solid state devices.

Given these points I agree that solid state devices are not always (1) better, nor (2) more reliable. It entirely depends on application. It entirely depends on environment acting upon the device. What are we talking about first of all? Since his later statement was a general statement regarding reliability of solid state devices versus mechanical, then its true one cannot state which one is superior. Not in terms of function. Not in terms of reliability. This should be pretty obvious.

So on the note of SSDs, I think you cannot define which one is more superior as a blanket statement just yet. SSDs are not good enough today yet in terms of storage capcity or endurance (what does good enough mean? Industry definition needed. Likely tied with economics and performance metrics. e.g. say good performance is lasting 5 years service in consumer operation, this is a definition). Similarly mechanical devices are not good enough today either (again depends on definition). Where does one focus on research anyway between mechanical versus solid state memory? Lets look at potential. To do this we can look at the physical limitations that plague and are intrinsic to both approaches to non-volatile memory devices. The idea is the better storage is one with least limitations (again least is extremely vague because one cannot make blanket statements predicting use of devices in applications. it always plays out that usefulness is always application driven, not concept drive).

Solid state memory are currently all semiconductor devices. SSDs are currently composed of floating gate transistors. The write and delete operations rely on charge storage of the floating gate. The floating gate is separated by thin dielectric layer from the semiconductor substrate that controls charge transfer between a source and drain. They rely on electron tunnelling to charge or discharge the floating gate during write or delete. The oxide is a very thin layer of dielectric material. We are talking nanometers of thickness. And you will need this in order to observe reliable electron tunnelling phenomena suitable for repeatable device operation. These thin dielectric layers suffer from a lot of limitations both theoretically and realistically in terms of engineering/fabrication. You cannot make a perfect thin material layer of desered properties. Quantum tunnelling of electrons is very sensitive to environmental factors also. You will not realistically observe tunnelling if the thickness is too large by even just a little bit. Say there is delamination of the oxide from the substrate or manufacturing errors with thickness of deposition layers and you shift difference by 1nm then device won't work cause electrons cant tunnel through the barrier (not to any appreciable probability that one can observe in a suitable time frame of applications anyway). Too thin and you can even get charge leakage, defeating the purpose of memory. Electrons are also very sensitive to background electromagnetic noise. ALso, floating gates erase by being applied a voltage potential that causes electrons to tunnel through the dielectric leading to charge gain or loss for the floating gate (this is equivalently setting the value for the storage bit). A related proble: electrical erase of charge does not limit itself when applied a discrete potential over finite time. It is possible to lead to a floating gate developing a POSITIVE charge due to overerasing. This will cause problems for future reads of cells. There are also many other limitations of floating gate transistor based flash memory such as gate disturbances etc but thats another topic. ETC ETC ETC. So there are MANY drawbacks. Good engineering overcomes this just like anything else (same with good engineering for mechancial drives to be discussed later). But ok, we are all interested in this thread about limitations of 'write/erase' that we have all heard about. This affects ENDURANCE. Do we know WHY? This is entirely related to solid state physics phenomena at the material level. With extended writes/erase, you will inevitably generate electron holes due to band to band tunneling. A small amount of these holes are injected into the thin dielectric layer holding charge on the floating gate. Hot hole injection during erasing operation over time will change the threshold voltage of the cells in memory arrays and trapped holes in the dielectric will also degrade the charge retention capability of memory cells. In effect, extended usage means LOSS of charge retention capability of floating gate transistors. I read someone asking what is meant in the long term in terms of the drive? Can it still read data? Well the short answer is NO. It just means you get a whole lot of nothing because memory of flash devices are based of charge storage, and material property degradation over time leads to loss of charge storage. This means your data will be completely lost if the transistors lose all ability to store charge due to degradation. Of course thats an extreme statement cause theres lilkely some charge retention so data may still be able to be extracted albiet much harder than extracting from mechanical magnetic drives.

Onto mechanical drives. Mechanical drives have the bad rep because mechanical phenomena are more obvious to us. They have also been in application for longer so we have more experience with them. We understand such things as friction and wearing from our daily lives even if we are not engineers or scientists. These concepts have driven into our mind of the idea of reliability. This bolsters our thinking they are less reliable than 'solid state'. I think it is a slight misconception in this. Why do mechanical drives fail in the first place? One thing is mechanical drives fail because of mechanical wearing. This can occur over the course of a long time or immediate (such as an impact) where a mecahnical part will cause structural damage to a storage medium hence leading to its loss of data and damage to subsequent device function (since the sensor and other devices are likely broken too). Since it is easily intuitive for us to think of reliability as sort of like 'wearing' (analogy of your car break disk in ones mind for example), then we tend to see mechanical devices as 'worse'. Its only worse for the same reason why SSDs have their problems. Ultimately its related to limitation of manufacturing leading to problems due to phenomena at the nanoscopic level. If you can remove the idea of static or fluid 'friction', the common terms practiced in mechanical engineering, and you also remove vibrations, then these drives can technically last a long time. Current mechanical drives are based on magnetic memory. Magnetic storage is actually much less sensitive to common enviornmental factors like temperature, background electromagnetic noise etc, compared to flash memory.

It is possible to need to solve HUGE matrixes. Why not? Finite element analysis with many degrees of freedom will require huge matrixes to solve. There will be no other way. In fact finite element analysis theory will tell you the more discrete elements the better. This gets the finite difference solution to say differential equations closer to an otherwise continuous or analytical solution. This means larger and larger matrixes are more ideal in general. The drawback is in computation time. So methods need to optimize this. Topic is complex all on its own.

Anyway. I am of the opinion that SSDs are useful as are mechanical drives. SSDs are certainly faster than mechanical memory to read/write data currently. Use depends on definition. But on the topic of reliability: I dont think anyone can state SSDs to be superior to mechanical drives always. They will certainly be superior in many critical aspects of a specification for certain applications. They won't for others.

 

Wow is that a paper on SSD's or a post

 

A few reactions:

1) enterprise SSDs use SLC, so multiply all the previous durations by 10. 370 days of continuous writing for 64GB.

2) enterprise applications use much larger storage arrays. E.g., on the order of 512GB at least, so multiply all the durations again by at least a factor of 8. 8 years is a decent lifetime for enterprise usage. Again, the systems will most likely be replaced/upgraded before the wear limit becomes an issue. In many enterprises this is mandatory - all hard drives are replaced the day their (typically 5 year) warranties expire; they're required to do so to keep their support contracts valid.

3) large-scale computation like you're talking about will not be writing continuously to disk. Or at least, not if the software was designed by a competent software engineer. Even the fastest storage drive is still an order of magnitude slower than RAM, so a well-designed compute engine will read a large chunk of data into RAM, do all the computations in RAM, and then write a chunk of results out to disk. That allows the majority of computation to occur at RAM speed; to simply write results continuously to disk will limit all your computation to disk speed, and your users will die of old age before getting the results they're looking for. In a well-designed system the time spent in writes should be no more than 25% of total runtime: data in, crunch numbers, data out.

 

just get off the paranoya, useroflaptops. get one, and love it, and love to kill it once you've made it. we'll see you in some years, then

because in any, unimportant how intensive use case in a laptop, it will last long enough to not matter. chance are, it will survive your laptop easily.

 

And that's why SLC drives exist, as they have literally 10 times the life span of MLC drives.

Agreed.

 

You'd have to define environmental factors first. Then for those factors its win some lose some for each side.

physicists as in phd in physics? then you should have heard of this type of discussion from your solid state physics class or from friends who are electrical engineers. its the kind of thing one would be thinking about especially if one were in the field or related. its a pretty obvious question a curious mind will naturally ask themselves and others. its like how f1 lovers will at some point imagine what its like watching nascar also even if they dun watch or like it

you will if youre doing simulation work. it may or may not be an exageration but its entirely possible. plus its an example to highlight the point about limitation in longevity if lots of calculations are required. plus simulation work are often iterative, meaning the calculations are repeated many times too to get convergence

 

I don't really know what you talk about but in general, every app, even some simulation apps, don't store all their states on the disk, as disks are very slow. if the simulation wants to perform well, it runs on cpu and ram, or nowadays even on gpu. only a small amount of states get stored to disk.

you, sir, are really dreaming yourself something up. a hdd is just as solid state as an ssd. but the ssd doesn't have moving parts. that's why it's called solid state. nothing about physics, about matter, about environmental things.

the ssd doesn't move. the hdd does. but the data on it is stored in a simlar way, it could always get killed in some form.

 

um. sure if you say so boss. i'll let someone fill in.

 

what?


.

 

Yes. But some of them (like acceleration) are a whole lot more prevalent than others so if it is many orders of magnitude better at those, it is definitely more reliable.

I have heard discussions -- what I haven't heard of is a debate. Fundamentally, the SSDs are without a doubt superior because there is a limit to how fast you can spin a disk. What remains is an engineering problem (pick the right materials and make decent quality flash memory) and a programming one (make a decent controller with an algorithm that spreads the writing between cells). As far as I can tell, this is well on its way to being solved and modern SSDs appear to be fine.

Half of my job is "simulation work". I routinely analyze hundreds of GB (sometimes even several TB) of data per day and I can tell you that reading from and writing to storage is the single largest bottleneck and cause of hardware related errors even if all you use storage for is to read stuff in and write stuff out (all of the processing happens in RAM). This is why I am so appalled by the idea of using disk space for processing -- it's a really, really bad idea to take what is already the weakest link and place even more stress on it.

 

yes, as long as we don't get some magical synergy of all the pros of ram and solid state storage (size and non-volatile-ness of hdds, speed and random access of ram), better put as much as possible into ram...

will be a while till we have tb of non-volatile ram.. but maybe memristors will help one day.