Samsung SSD 840 EVO 250GB & 1TB TLC NAND Drives Tested

MojoKid writes "Samsung has been aggressively bolstering its solid state drive line-up for the last couple of years. While some of Samsung's earlier drives may not have particularly stood-out versus the competition at the time, the company's more recent 830 series and 840 series of solid state drives have been solid, both in terms of value and overall performance. Samsung's latest consumer-class solid state drives is the just-announced 840 EVO series of products. As the name suggests, the SSD 840 EVO series of drives is an evolution of the Samsung 840 series. These drives use the latest TLC NAND Flash to come out of Samsung's fab, along with an updated controller, and also feature some interesting software called RAPID (Real-time Accelerated Processing of IO Data) that can significantly impact performance. Samsung's new SSD 840 EVO series SSDs performed well throughout a battery of benchmarks, whether using synthetic benchmarks, trace-based tests, or highly-compressible or incompressible data. At around $.76 to $.65 per GB, they're competitively priced, relatively speaking, as well."

Comments

Call me old fashion

[Taco Cowboy]

How many effective READ/WRITE cycle can the chip in SSD perform, before they start degrading ?

Has there been any comparison made in between the reliability (eg read/write cycles) of old fashion spinning-plate HD versus that of SSD ?

Re:Call me old fashion

[quarrel]

Ok, you're old fashioned.

This was a thing, yes, but only for that brief period when you actually got your slashdot id. Since then? Not so much ...

--Q

Re:Call me old fashion

[pipatron]

The question is still relevant. Manufacturers talk about erase cycles, but are there any massive-scale studies done by a third part on SSD failure modes?

Re:Call me old fashion

[Anonymous Coward]

Well said.

Nothing lasts forever. If a hard-driven SSD lasts 3-4 years, I don't really care that if it's used up some large fraction of it's useful lifetime, because I'm going to replace it just like I'd replace a 4 year old spinning disk.

And the replacement will be cheaper and better.

And if the SSD was used to serve mostly static data at the high speed they provide, then it's not going to have used up its write/erase cycle lifetime by then anyway.

Re:Call me old fashion

[Anonymous Coward]

Wearout is not a significant failure mode. Nearly all failures are to due to non-wearout effects such as firmware bugs and i/o circuit marginality.

Re:Call me old fashion

[beelsebob]

Yes, many sites have done the maths on such things. The conclusion "finite life" is not the same thing as "short life". SSDs will in general, outlast HDDs, and will in general die of controller failure (something which affects HDDs too), not flash lifespan.

The numbers for the 840 (which uses the same flash, with the same life span) showed that for the 120GB drive, writing 10GB per day, you would take nearly 12 years to cause the flash to fail. For the 240/480/960 options for the new version you're looking at roughly 23, 47 and 94 years respectively. Given that the average HDD dies after only 4 years (yes yes yes, we all know you have a 20 year old disk that still works, that's a nice anecdote), that's rather bloody good.

Re:Call me old fashion

[beelsebob]

The problem with large scale studies on this is that it takes too long to happen to actually study. You need to study real world usage patterns, and in the real world it takes decades before the flash actually wears to death. Controller failure (as is possible with HDDs too) will generally happen long before the flash becomes unwritable.

Re:Call me old fashion

[Anonymous Coward]

I am more worried about what happens when the power is pulled suddenly or a spike/dip comes through. I recall reading an article which showed the drive could be bricked like that...http://www.zdnet.com/how-ssd-power-faults-scramble-your-data-7000011979/
I know some areas are very susceptible to power outages...

Re:Call me old fashion

[bemymonkey]

For an OS drive (page file off, 8+GB of RAM), I don't see any "premium" (i.e. non-cheapo) SSD with 120+GB of capacity failing within 10 years... There are a few forum posts where people have actually tested how much data they could write to SSDs (i.e. permanently writing at max. speed until the drive fails), and the results are pretty good. The few drives that did eventually break just switched to read-only mode... Can't for the life of me find the damned thread though. Maybe someone here knows which one I'm talking about? It was in one of the usual tech forums...

As a cache drive for audio/video or similar "write-multiple-gigabytes-per-minute" applications, well, make sure you've got enough cash in the bank for the replacement drive...

Re:Call me old fashion

[jamesh]

have they solved the corruption-on-power-failure issues yet?

Re:Call me old fashion

[beelsebob]

Yes, they were solved in a firmware patch a long time ago.

Re:Call me old fashion

[beelsebob]

The problem with such tests of writing as much as you can as fast as you can is that they're rather deceptive. They don't allow TRIM and wear levelling to do their thing (as they normally would), and hence show a much worse scenario than you would normally be dealing with. Actual projections of real life usage patterns writing ~10GB to these drives per day show you can get their life span in years (specifically the 840 we're talking about here) by dividing the capacity (in gigabytes) by 10.

Re:Call me old fashion

[Joce640k]

Modern SSDs have big capacitors in them to avoid that (well, some of them do...). They can complete pending writes on capacitor power alone.

Re:Call me old fashion

[bemymonkey]

That's definitely true, but with the drives not showing any signs of abnormally early failure even in the worst-case scenario, I'd say that's a good thing. :)

Re:Call me old fashion

[Anonymous Coward]

Really so ZDNet was testing obsolete drives this March? http://www.zdnet.com/how-ssd-power-faults-scramble-your-data-7000011979/

Re:Call me old fashion

[Gaygirlie]

How many effective READ/WRITE cycle can the chip in SSD perform, before they start degrading ?

They don't start degrading, per se. Performance-degradation is all due to wear-levelling and the amount of free blocks on the drive, and that varies between manufacturers. Generally the advice is to have atleast 20% of the drive free at all times for wear-levelling and TRIM to work efficiently and in such a situation there should be no performance-degradation.

As for reading and writing cells? Well, you can read a cell indefinitely. You cannot write to cells forever, however, and once the limit comes there is 100% degradation -- so to speak -- as in that that cell cannot be written to ever again. It just goes from 100% to 0%, so using the term "degradation" for that still seems useless. I'll repeat, though, that it can still be read from even if it can't be written to.

Has there been any comparison made in between the reliability (eg read/write cycles) of old fashion spinning-plate HD versus that of SSD ?

Plenty, but how much those comparisons actually cover and how reliable they are is subject to debate. Generally the consensus is that SSDs are more reliable nowadays as full-on controller-failures are very rare and since the SSDs can still be read from even if they hit the maximum amount of writes that means your data is quite a lot safer in the long run -- if a regular, mechanical drive can't write to some sector it most likely can't read it either, and that means your data is as good as gone.

Re:Call me old fashion

[beelsebob]

Noting that march is a long time ago in tech terms, and that one of the (incredibly small sample of 2) HDDs suffered issues as well.

Re:Call me old fashion

[maxwell demon]

Given that the average HDD dies after only 4 years

I guess I must have had exceptionally good HDDs. I only had one HDD failure (and that happened after a power source failure, so I suspect it was caused by that), and with the exception of some of those currently in use, all of my HDDs (including the failed one) have been in use for more than four years. And in almost all of my computers, I had more than one HDD.

Re:Call me old fashion

[Anonymous Coward]

They do indeed suffer from this though, at least earlier drives did. See Vertex 2.

Re:Call me old fashion

[beelsebob]

Yes, the 840 did indeed suffer from this, but as I said up the thread, the firmware was patched to fix the issue.

Re:Call me old fashion

[ericloewe]

Some reviewers take popular devices and see if they can kill them by bombarding them with writes.

So far, the consensus is that, for typical consumer workloads, the limits on NAND writes are high enough not be a problem, even with Samsung's TLC NAND.

Same should apply to heavy professional workloads when using decent devices (Samsung 830/840 Pro and similar).

As for servers, the question is a bit more difficult to answer, but even assuming a very bad case, SSDs make sense if they can replace a couple of mechanical drives (Throughput is most important, not the amount of data stored)

Re:Call me old fashion

[AmiMoJo]

It depends on what you use it for. I managed to wear out an Intel XM-25 160GB SSD a few years ago by hitting the 14TB re-write limit.

Modern SSDs so a lot of compression and de-duplication to reduce the amount of data they write. If your data doesn't compress or de-duplicate well (e.g. video, images) the drive will wear out a lot faster. I think what did it for me was building large databases of map tiles stored in PNG format. Intel provide a handy utility that tells you how much data has been written to your drive and mine reached the limit in about 18 months so had to be replaced under warranty.

Re:Call me old fashion

[gigaherz]

Read cycles? many millions. Write cycles? not so many, but a lot. ERASE cycles? around the 10k order of magnitude, for the latest generation TLC. And it will get smaller as they shrink the capacity of the floating gates (electrons get stuck there, and it fills quicker).

Re:Call me old fashion

[dmesg0]

TLC endurance was tested in this article:

http://uk.hardware.info/reviews/4178/10/hardwareinfo-tests-lifespan-of-samsung-ssd-840-250gb-tlc-ssd-updated-with-final-conclusion-final-update-20-6-2013

Re:Call me old fashion

[MobSwatter]

Ok, you're old fashioned.

This was a thing, yes, but only for that brief period when you actually got your slashdot id. Since then? Not so much ...

--Q

All things considered, if they do not shatter the barrier of flash memory 200k R/W bit failure, there is no algorithm that can improve upon standard HDD platters. Sure they can make a flash bit live a little longer and provide better throughput to the drive but the technology overall cannot survive a superior storage technology. Stop kicking a dead horse and move on to something new.

Re:Call me old fashion

[citizenr]

ERASE cycles? around the 10k order of magnitude, for the latest generation TLC. And it will get smaller as they shrink the capacity of the floating gates (electrons get stuck there, and it fills quicker).

haha, try 300-500 for latest smallest TLC

Re:Call me old fashion

[Tapewolf]

Ok, you're old fashioned.

This was a thing, yes, but only for that brief period when you actually got your slashdot id. Since then? Not so much ...

--Q

Technically it becomes less and less reliable each time they do a die shrink on the flash. Adding a whole extra bit level makes things worse still. In the early 2000s you were looking at 100'000 P/E cycles, maybe a million for the really good stuff. Good TLC memory seems to be rated around 3000, with a figure of 1000 being widely quoted, and in some cases, less.

Realistically, they've designed the drive to fight tooth and nail to avoid doing rewrites, and in actual fact it looks like they've put a layer of fast SLC cache in front (i.e. the million-cycle stuff). What could be more interesting is the retention period - if the thing is left powered off for three months it could well be left unreadable.

Re:Call me old fashion

[gigaherz]

With wear leveling?

Re:Call me old fashion

[bemymonkey]

Found it: http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm

It's a bit out of date, but basically: Stay the hell away from OCZ and certain Intel drives, and you'll be fine in nearly all cases.

Re:Call me old fashion

[Rockoon]

Intel provide a handy utility that tells you how much data has been written to your drive and mine reached the limit in about 18 months so had to be replaced under warranty.

You were (amplified?) writing 32.8 GB per day, on average.

Clearly you will run into SSD erase-limit problems at such a rate, but such workloads normally turn out to not be tasks that actually benefit from an SSD to begin with (32.8GB/day = 380KB/sec, so the devices speed wasnt actually an issue for you)

You were either very clever and knew you would hit the limit and get a free replacement, or very foolish and squandered the lifetime of an expensive device when a cheap deice would have worked.

In any event, in general the larger the SSD the longer its erase-cycle lifetime will be. For a particular flash process its a completely linear 1:1 relationship, where twice the size buys twice as many block erases (a 320GB SSD on the same process would have lasted twice as long as your 160GB SSD with that work load)

Re:Call me old fashion

[Anonymous Coward]

Power failure?

You don't have a UPS or other standby power source available? You know its 2013 right...

Willing to spend hundreds on an ultra fast STORAGE device and have no backup power available? really? come on...

That's some messed up priorities there... Spend a hundred bucks on a UPS already.

Then you don't ever have to worry about data corruption. Or the much more common... Loss of unsaved work due to power failure...

Re:Call me old fashion

[Rockoon]

Indeed. I just don't see how the erase-limit issue applies for most people. The most common activity where it might apply is in a machine used as a DVR (dont use an SSD in a DVR), with the next being a heavily updated database server (you may still prefer the SSD if transaction latency is important.)

For people that use their computers for regular stuff like browsing the web, streaming video off the web, playing video games, and software development.. then get the damn SSD -- its a no-brainer for you folks.. you will love it and it will certainly die of something other than the erase-limit long before you approach that limit.

Re:Call me old fashion

[beelsebob]

Clearly you will run into SSD erase-limit problems at such a rate, but such workloads normally turn out to not be tasks that actually benefit from an SSD to begin with (32.8GB/day = 380KB/sec, so the devices speed wasnt actually an issue for you)

Actually, most devices will survive several years at such a rate. GP was unlucky to see failures quite

Re:Call me old fashion

[Anonymous Coward]

TLC is the worst tech though. Bring back SLC, or even MLC :/

Re:Call me old fashion

[Rockoon]

Technically it becomes less and less reliable each time they do a die shrink on the flash. Adding a whole extra bit level makes things worse still. In the early 2000s you were looking at 100'000 P/E cycles, maybe a million for the really good stuff. Good TLC memory seems to be rated around 3000, with a figure of 1000 being widely quoted, and in some cases, less.

Lets not neglect the fact that while every die shrink does reduce the erase-limit per cell, it also (approximately) linearly increases the number of cells for a given chip area. In other words, for a given die area the erase limit (as measured in bytes, blocks, or cells) doesnt actually change with improving density. What does change is overall storage capacities and price.

When MLC SSD's dropped from ~2000 cycles per cell to ~1000 cycles per cell, their capacities doubled (so erases per device remains about constant) and prices also dropped from ~$3/GB to about ~$1/GB. Now MLC SSD's are around ~600 cycles per cell, their capacities are larger still (again erases per device remain about constant), and they are selling for ~$0.75/GB (and falling.)

By every meaningful measure these die shrinks improve the technology.

So now lets take it to the (extreme) logical conclusion, where MLC cells have exactly 1 erase cycle (we have a name for this kind of device.. WORM: Write Once Read Many.) To compensate, the device capacities would be about 600 times that of todays current capacities, so in the same size package as todays 256 GB SSD's we would be able to fit a 153 TB SSD WORM drive, and it would cost about $200.

Re:Call me old fashion

[Anonymous Coward]

No, they don't !
Only some server SSD have capacitors. They are considered too "expensive" to be put on the consumer version.

Re:Call me old fashion

[Rockoon]

around the 10k order of magnitude, for the latest generation TLC.

You are thinking of SLC, not TLC, and are also probably off by a generation.

Re:Call me old fashion

[Traciatim]

Of course that math is done on the assumption that 10GB per day can be spread over the entire drive which isn't the case once you have 100GB of data on it, suddenly that 10 years gets reduced to 1.7 years years and that's the estimated mean time to failure meaning the actual failure rate is probably withing +- 50% of that, so somewhere between 0.85 and 2.55 years is likely. That's bordering on the realm of "Not a reliable place to put data". Of course, your important data should probably be stored in multiple locations locally, as well as an additional copy in another physical location anyway if you really want to keep it, but citing those figures is not anywhere near a reasonable usage pattern of most drives.

Re:Call me old fashion

[hackertourist]

(32.8GB/day = 380KB/sec, so the devices speed wasnt actually an issue for you)

That's an odd way to look at it. You assume that GP spreads out his writes evenly over 24h, and has no desire to speed things up.

Re:Call me old fashion

[AmiMoJo]

In my case having an SSD made a huge impact. I was using offline maps of a wide area build from PNG tiles in an sqlite database with RMaps on Android. Compiling the databases was much faster with an SSD. I was doing it interactively, so performance mattered.

I can only tell you what I experienced. I installed the drive and I didn't think about it wearing out, just carried on as normal. The Intel tool said that it had written 14TB of data and sure enough writes were failing to the point where it corrupted the OS and I had to re-install.

I was using Windows 7 x64, done as a fresh install on the drive when I built that PC. I made sure defragmentation was disabled.

I'm now wondering if the Intel tool doesn't count bytes written but instead is some kind of estimate based on the amount of available write capacity left on the drive. I wasn't monitoring it constantly either so perhaps it just jumped up to 14TB when it noticed that writes were failing and free space had dropped to zero.

It was a non-scientific test, YMMV etc etc.

Re:Call me old fashion

[beelsebob]

Why, once again, actually figuring out the life of these devices shows that TLC devices will live a perfectly acceptable length of time, so why should we use SLC or MLC for consumer devices?

Re:Call me old fashion

[beelsebob]

Actually, no, that maths was done with the assumption that wear levelling would be able to do the average case job for a consumer drive which is reasonably full. If you assume ideal conditions the life span is in fact much longer than that.

Re:Call me old fashion

[Anonymous Coward]

If so, that's a very important claim. Got a link to the Samsung release note or other official documentation that confirms it?

Re:Call me old fashion

[Anonymous Coward]

I work for a company that makes flash arrays. We push them much harder than using one for your c:\ drive, and yes, they DO wear out.

Re:Call me old fashion

[Immerman]

Better yet, have you got a link to an independent test that confirms whatever Samsung may be claiming?

Re:Call me old fashion

[Anonymous Coward]

The conclusion "finite life" is not the same thing as "short life". SSDs will in general, outlast HDDs, and will in general die of controller failure (something which affects HDDs too), not flash lifespan.

This is total bullshit. Every single SSD I had owned has failed within a year. Just from normal development work and use with virtual machines.

On the other hand I have hard drives with over 70000 powered-on hours still going strong. Hell, the primary RAID array in my workstation is made from drives that all have over 50000 powered-on hours on them.

Re:Call me old fashion

[Immerman]

I guess I'm old school, got used to saving regularly back before UPSes were a consumer product. If I lose power 3 times a year I've lost a total of maybe 15 minutes of work, and it's a rare year where I have three power outages while working. So the insurance of a UPS would cost me $100/ 0.25 = $400/hour. Even if it lasts a decade that comes to $40/hour for saving my ass from some inconvenience. Pretty steep price.

Re:Call me old fashion

[Immerman]

Isn't wear leveling mostly orthogonal to erase cycles? I.e. it tries to spread out the erasures evenly among the flash blocks, but doesn't actually change how many times any given block can be erased.

In fact I would think wear leveling would actually *decrease* the total number of user visible erase cycles for a drive since it means that infrequently changing data is continuously being shuffled from low-usage areas to high-usage ones to make the low-usage areas available, and every one of those user-invisible moves is another erase/write cycle lost.

Re:Call me old fashion

[Immerman]

So what exactly is the difference between a write cycle and an erase cycle in practical terms? Yeah I know and erase block typically contains many write blocks, but for any given write block I can't write to it a second time until it's erased, therefore it would seem the number of write cycles can't possibly exceed the number of erasures.

Or is there some marketing mumbo-jumbo being applied to the terms now?

Re:Call me old fashion

[beelsebob]

This is total bullshit. Every single SSD I had owned has failed within a year.

Damn it, I covered off one anecdote approach, and you found a different one.

For those who don't know how anecdotes work –if your sample size is incredibly small, you can not draw meaningful results from it. I don't care if you have a single 20 year old disk, 20 6 year old disks, or 20 SSDs that failed within the first year, none of these give you a full enough picture to actually tell you what's going on.

The actual stats on return rates show that SSD return rates are around 0.5% of all drives per year, while HDD return rates are about 5% of all drives per year. There's one notable exception, which is that if you bought OCZ drives a couple of years ago, you were looking at about a 10% failure rate.

The bottom line is, SSDs are now more reliable, and more long lasting than hard disks in a consumer setting. In an enterprise setting, you need to be careful and make sure that you get enterprise level SSDs because your write patterns will likely involve far more writing than any consumer will ever do.

Re:Call me old fashion

[AlternativeIdeas]

And let's not forget, when using an SSD on a laptop the UPS is "built-in".

Re:Call me old fashion

[gigaherz]

It's a bit like this:

Flash memory has a special transistor with a gate that's disconnected from the circuit. This gate can be loaded with electrons by sending them at a high enough voltage, or "sucked" empty by quantum tunneling.

I don't know exactly which is write and which is erase but: the write operation is less costly, but it can only turn bits off but not back on (or the other way around). When you need to reset the bits, so that they can be turned back off, you have to erase the sectors. The erase operation is more costly, and wears down the "floating gate" a lot quicker. Since it's more costly, it's done on a larger group of sectors at the same time. This means that you can reset a whole group, but then only write a piece of it. Here's where the famous TRIM comes to help: without TRIM, the flash controller doesn't know which sectors of an erase group are still used, so when the OS asks for a sector write that requires an erase operation, it has to read all of the sectors, erase, and then write all of them. With TRIM, it can ignore the sectors that are marked as empty.

The reason why erase cycles are limited, is because in those "inject"/"vacuum" cycles, some electrons get stuck inside the gate, and the smaller the gates get, the quicker they fill up.

Re:Call me old fashion

[gigaherz]

Not quite: in any practical usage, there's a lot of sectors that are only written once, and a few sectors that are written a lot more often. As an example: System files, and the TEMP folder, respectively. With wear leveling, the controller is free to swap them around, so that all the sectors are used more.

Of course, that sector rotation could result in the TOTAL number of writes/erases going up, but without wear leveling, all the sectors corresponding to the often-written files (TEMP, logs, pagefile/swap, ...) would die out quite quickly.

Re:Call me old fashion

[moosehooey]

Hence why we want to see a study that compares overall failure to old-fashioned drives, taking all failure modes into account.

I would like to see some evidence that SSDs are more reliable.

Re:Call me old fashion

[rsmith-mac]

Modern SSDs so a lot of compression and de-duplication to reduce the amount of data they write.

That is only true for SandForce based drives as the tech behind it is LSI's "secret sauce". Samsung, Marvell, and Toshiba do not do any kind of compression or dedupe; they write out on a 1:1 basis.

The latter group could probably create their own compression and dedupe tech if they really desired to, but it's a performance tradeoff rather than something that has clear and consistent gains. SandForce performance is more bursty than 1:1 writing, since the content matters.

Re:Call me old fashion

[moosehooey]

What is it with SSD controller failure?

The processor, bridges, memory controller and memory, and all the other chips in a modern computer, can run flat-out for many years without failure.

What makes the controller chips in a SSD fail so often? (And I don't believe you about the controllers in a HDD failing, I've never had one fail, or even known anyone who had one fail, out of hundreds of hard drives run for many years, but I've heard of several SSDs failing in just the few that my friends have tried). Do they spend so much on the Flash chips that they have to go that cheap on the controller chip?

Re:Call me old fashion

[Rockoon]

So what exactly is the difference between a write cycle and an erase cycle in practical terms?

The difference is that there is no such thing as a "write cycle." The guy that you are replying to doesnt actually know much about what he is talking about.

In regard to the general difference between writes and erases the terminology on the table are supposed to be write amplification, block size (typically 256KB), and page size (typically 4KB.) Write amplification occurs when data smaller than a page is frequently written or "modified."

In practice write amplification is typically below 2x on modern controllers, and obviously always larger than 1x.

Worst case write amplification is horrendous at 4096x but the typical scenario where near-worst-case amplification does occur turns out to be low volume traffic in practice (for every near-worst-case page write the SSD experiences, it experiences thousands of near-best-case page writes.)

The worst cases turn out to be log file writes, but thats very small change compared to other write activity (its like worrying about that $1 surcharge on your $150/month cable bill.)

Re:Call me old fashion

[sribe]

Given that the average HDD dies after only 4 years...

Sorry, not even close.

Re:Call me old fashion

[Kjella]

I hear people say that, but my first SSD I used as a scratch disk for everything since it was so fast, it burned through the 10k writes/cell in 1.5 years. My current SSD (WD SiliconEdge Blue 128GB) has been treated far more nicely and has been operational for 1 year 10 months, SSDLife indicates it'll die in about 2 years for a total of 3 years 10 months. Granted it's been running almost 24x7 but apart from downloads running to a HDD it's been idle most of that time, unlike a HDD where the bearings wear out that shouldn't have much to say for SSD life time only bytes written. I'd not buy any consumer disk unless you consider it "expendable" and will die in <5 years, personally I'll be looking at an enterprise disk next time around.

Re:Call me old fashion

[Traciatim]

You mean this test where almost all the drives are used at very low amounts filled, and the drives that are used with large static data fail in extremely short periods of time?

Re:Call me old fashion

[AmiMoJo]

We get interruptions to our supply less than once every five years. Even at 95% efficiency a UPS would cost a fair bit to run. It would be better if, like hard drives, SSDs were simply designed not to die in the event of unexpected power failure.

Data corruption isn't an issue with modern file systems. I suppose there is loss of work, but the cost/benefit ratio is too low.

Re:Call me old fashion

[Anonymous Coward]

Unless your battery dies.

Re:Call me old fashion

[Joce640k]

Isn't "some" the word I used, AC?

Re: Call me old fashion

[Zeinfeld]

Hmmm I replace my hard drives when I start to see RAID errors. I don't plan to run SSD raid as the on board fault tolerance should be ok.

Would be nice to have hard data on expected failures so that I know whether to plan for a three or a six year lifespan. I generally replace my main machine on a six year cycle as I have a lot of expensive software. Looking to upgrade this year when the higher performance intel chips launch.

1tb is quite a lot. Probably more than I need in solid state. The price is also quite a bit more than the $0.05/gig for Hard drives. But it's getting a lot narrower. And RAID 1 doubles that cost anyway...

Re:Call me old fashion

[beelsebob]

No, I mean this, which has a detailed explanation of what's going on, and why you shouldn't care.

Re:Call me old fashion

[Immerman]

Certainly, but my point is that you can't meaningfully write to a cell without first erasing it, so the number of writes to a cell cannot exceed the number of times it block has been erased.

Perhaps it will be clearer if you consider an SSD where each cell can be erased independently - it would be more expensive to make, but in principle there's no reason you couldn't do such a thing.

Re:Call me old fashion

[Anonymous Coward]

Power spikes can ruin any electronic equipment, it's not limited to SSDs. If you have power issues, get a UPS, they aren't expensive.

Re:Call me old fashion

[Immerman]

Well yeah - which is why I said it's orthogonal: wear leveling has absolutely zero effect on the number of erase cycles a given block can handle, it just spreads the load around since otherwise some blocks will get hammered while others are almost never modified.

Similarly the number of write cycles available has almost no effect on wear leveling beyond setting the "danger limits" for each block.

Ah, okay, I think I see the source of confusion - yes, in most usage patterns wear leveling will dramatically increase the number of erases before the first block failure, but it will also dramatically decrease the number of erases before the *last* block failure. Which would be relevant in some hypothetical drive that gracefully lost capacity as blocks "died"

Re:Call me old fashion

[iggymanz]

you have hard facts. 2007 google study said about six years for all enterprise and consumer grade magnetic disks, however for low utilization disks most fail in only three years (contrary to most people's expectations)

Re:Call me old fashion

[Traciatim]

So you mean on this page where they estimate a life span of a perfectly empty 128GiB drive using TLC nand at 2.5 years... but if it was 75% full then it would be a quarter of that, which is pretty close to what I estimated before of likely to fail between 0.85 and 2.55 years?

Re:Call me old fashion

[Anonymous Coward]

The problem with your extrapolation is that the smaller the feature size the smaller the charge per cell. The cells leak electrons which leads to corruption. This is a far-from-perfect WORM and the data would have to be constantly migrated ahead of the leaking electrons to maintain the data. If the device was ever power cycled it would likely be corrupt before it came back on-line.

Re:Call me old fashion

[Rockoon]

What makes the controller chips in a SSD fail so often?

It isn't the controller chips that are failing (a hardware fault,) its buggy logic in the controller firmware (a software fault) that leaves the data stored within the flash in an incoherent state (garbage in, garbage out.)

Re:Call me old fashion

[damnbunni]

I have, personally, had about eight hard drive controllers fail. In all cases, I was able to replace the controller board on the drive and recover my data. (And generally keep using the drives as scratch disks.) I'd get the boards by buying headcrashed drives.

Most of these were quite some time ago, when I was dealing with a lot of identical, fairly small hard drives. Back when SCSI controllers had an option for drives that took extra-long to spin up. (We called it 'Seacrate' mode.) I've also had my share of head crashes, stiction, and general crap-outs. (The most heartbreaking was when the Micropolis 9GB in my Amiga died. It actually died when power was lost while repartitioning it. Argh!)

I don't necessarily think hard drives have gotten more reliable since. I'm sure they have, though. It's just that I don't deal with enough of them to break a bunch any more.

Re:Call me old fashion

[Anonymous Coward]

What fucking idiot modded this informative?
1. What the fuck is a XM-25?
2. Lifetime write limit for a first gen 160GB X25-M is 140TB, not 14.
3. To write 140TB in 18 months, you'd have to write >250GB/day. to a 160GB drive. For 18 months straight.

Re:Call me old fashion

[TheLink]

Just don't buy OCZ crap?
http://www.behardware.com/articles/881-7/components-returns-rates-7.html
http://www.behardware.com/articles/881-6/components-returns-rates-7.html

It's not a study but it's good enough for me.

Re:Call me old fashion

[Anonymous Coward]

How are you getting your quarter figure?

They are already using a 10x write amplification number.

Even if it's 75% full of static information the drive can move that "static" stuff around so that it doesn't always just write to the same 25% of the drive- that 25% moves about in the available flash. That's why you should try to have smart people making the controllers and software.

And that's also probably why controller failure and bugs are the main cause of SSDs data loss. Unless you buy OCZ crap - in which case the main causes are failure in part of the company and the customer being stupid/ignorant.

Re:Call me old fashion

[gigaherz]

You CAN write to a cell without first erasing it: if it was in its reset state, either because it hadn't been written before, or because the contents were TRIMmed or explicitly erased. With TRIM support, that means any time you write to a previously unused physical sector. Overwriting data DOES require an erase, though.

Re:Call me old fashion

[Anonymous Coward]

The problem is your post put the wrong spin on reality.

e.g. true but misleading.

Re:Call me old fashion

[Tapewolf]

By every meaningful measure these die shrinks improve the technology.

How about data retention? That is also a function of the cell size, since the more electrons you have in the charge trap, the greater the difference between 1 and 0. Intel's drives, for example, are only guaranteed to hold their contents for three months without power. And when they are powered, they keep the data alive by periodically rewriting it, which I strongly suspect amounts to a P/E cycle. (Not sure about flash, but a lot of memory devices use an 'erase' to set the bits high, and then short out the ones they want to be zero, so a lot more actions are liable to count as an performing an 'erase' than deleting a file.)

Re:Call me old fashion

[TheLink]

That link contradicts your earlier post. Few of those SSDs fail in read-only mode.

I'm sticking to Samsung - my 830 is still working - light workload. My guess is something else will kill it than NAND wear.

Re: Call me old fashion

[azav]

OK. I have had an OWC SSD in my Mac for a year and get about 450 MB/s reads and writes. Totally worth it.

And it's "old fashioned", not "old fashion".

Re: Call me old fashion

[azav]

Intel provides* , not Intel provides.

Re: Call me old fashion

[azav]

its useful lifetime, not it's useful lifetime

        it's = it is

Re:Call me old fashion

[jopsen]

Given that the average HDD dies after only 4 years

I guess I must have had exceptionally good HDDs. I only had one HDD failure....

As grandparent said "that is a nice anecdote"... but were you "writing 10GB per day"?

I don't know where the 4 years number comes from... but it's not completely unlikely. Just take a look at the research:
http://static.googleusercontent.com/external_content/untrusted_dlcp/research.google.com/da//archive/disk_failures.pdf

Granted it's a few years old...

Re:Call me old fashion

[sribe]

you have hard facts. 2007 google study said about six years for all enterprise and consumer grade magnetic disks, however for low utilization disks most fail in only three years (contrary to most people's expectations)

Bullshit. That's not at all what the google study said.

In fact, it said absolutely nothing about the six year timeframe, since it only had 4-5 years of data ;-)

Re:Call me old fashion

[Immerman]

Well yeah, but I figure in a discussion of hundreds or thousands of write cycles the very first write is irrelevant to the point - so okay, technically the number of writes can't exceed the number of erases +1.

As for TRIM, it's totally irrelevant to the discussion, those are logical-level operations on sectors, and wear leveling will bounce those sectors all over the drive. I'm discussing the physical flash memory cell that actually stores data. A breakdown from the perspective of the internal control circuitry that's already handled all the logical-level stuff and is now dealing with the actual, physical storage of data: (somewhat out of order so that it can be read easily)

Write cycle 1: I write data into physical cell #27463 in block #6 ("I" being the SSD's internal control circuitry)
[...] lots of other stuff, wear leveling, TRIM etc. None of which directly effect the contents of the actual physical cell
Write cycle 2: I want to put new data in physical cell #27463
          I can't do that without first erasing the corresponding block, so..
          Erase cycle 1: physical block 6 is erased (presumably all other cells have already been released/moved by higher-level processes)
[...] more stuff that has no effect on the contents of physical cell #27463
Write cycle 3: I again want to change the data in physical cell #27463, so I have to trigger
        Erase cycle 2: physical block 6 erased for the second time.
and so on and so forth.

Note that it's impossible* to put new data into physical cell #27463 without first erasing it, so its write count can never exceed its erase count +1.

*okay, not 100% clear on the mechanism, I suspect you erase to all 0s and can then write only 1s, or vice-versa, in which case in some specialty applications you could leverage that behavior for multiple writes between erases. But that's not relevant to a general purpose "the cell used to hold one chunk of random data, now I want it to hold some other unrelated chunk of random data" application.
       

Re:Call me old fashion

[JDG1980]

This is total bullshit. Every single SSD I had owned has failed within a year.

Stop buying from OCZ and start using reputable brands.

Re:Call me old fashion

[JDG1980]

What makes the controller chips in a SSD fail so often?

Intel and Samsung controllers are pretty reliable. Most SSDs from other vendors use either Corsair or Marvell controllers.

Marvell doesn't provide firmware at all, so vendors have to write their own. Many of these vendors are small companies with little in-house expertise, and what effort they do put in to their firmware is often devoted to focusing on speed (so they appear at the top of review sites' benchmarks) rather than stability.

Sandforce is at the other end of the spectrum: they don't provide their firmware source to anyone, and it's pretty much take-it-or-leave-it. (Even Intel had to jump through a whole bunch of hoops to get some stability bugs fixed when they adopted Sandforce controllers for their consumer-grade SSDs.) But the real problem is that the original Sandforce firmware developers were paranoid, and thought every other vendor wanted to steal their precious, precious ideas. So they designed the controller so that if anything electrically out of the ordinary happened, it would be interpreted as a hacking attempt, and the drive would go into "panic mode" (bricking), from which only Sandforce could recover it – and they'd be very reluctant to do this for end users, and even if they did, you'd still lose all your data.

Sandforce has since been sold to LSI, and I think stability is better now than it once was, but I still wouldn't want to risk it.

Re:Call me old fashion

[Anonymous Coward]

You don't have a UPS or other standby power source available? You know its 2013 right...

The vast majority of my power failures have been because of a UPS. Sure the most critical machines have dual input power supplies, but I don't pay that much for all my equipment.

That's some messed up priorities there... Spend a hundred bucks on a UPS already.

Hundred bucks? That's one cheap ass UPS. Are you buying those box-store power strip style ones? My wall outlet is 10x more reliable than those.

Re:Call me old fashion

[0123456]

It would be better if, like hard drives, SSDs were simply designed not to die in the event of unexpected power failure.

About 80% of the hard drive failures on our servers over the last few years have been due to power failures. They run fine for years, then the power goes out and they're dead on boot.

So 15k HDDs don't seem to like power failures either.

Re: Call me old fashion

[KingMotley]

it's == it is, not it's = it is

Re:Call me old fashion

[KingMotley]

And as an additional anecdote, because I can... I have 6 SSDs (2 Intel X25-M bought 2008, 1 OCZ Vertex 2, 1 OCZ Vertext 3, 2 Samsung 830s). One Intel died when it was about 90 days old, and Intel replaced it free of charge, free shipping both ways, and they cross shipped it. All 6 are still running today. The other Intel SSDs is in one of my Internet facing Web/FTP/Email/MySQL servers serving a few thousand clients and has been running 24x7 for 5 years. Last I checked it, it was less than 5% "used up" it's write cycles. This was switched from a traditional HDD that was running at 100% for most of the day. When we switched it to SSD, it was utilizing the SSD at 5% during peak times.

In that same time frame, I started with 6 320GB drives in a raid. 2 failed, I replaced them all with 2 1TB drives. 1 of those 1TB drives failed, I replaced both with 4 2TB drives. 2 of the 2TB drives failed, and I replaced them with 6 3TB drives. Since then, I've expanded it to 10 3TB drives in the raid array, 2 in a different raid, and had 3 of them fail of which I've replaced 2 of the 3. So in 5 years, I've had one SSD controller failure (of 5), and 8 hard drive failures (of 26, many of which were taken out of service prematurely because of the high failure rate). All the mentioned HDDs were my personal hard drives on my main PC. Not exactly stress testing them, but would sometimes push their throughput to the limit for a small portion of the day.

So for my small sample rate, HDDs die much faster than SSDs do, and I do a lot more I/O on my SSDs than I do my HDDs. Contrary to what some people say, my swap file, hibernation file, OS, and temp file location is on the SSD, while the HDD array is mainly for storing large media, and a smaller 2-drive RAID-1 for applications.

Re:Call me old fashion

[KingMotley]

I'm pretty sure your idea of how TRIM works is flawed. Close, but flawed. An erase "cycle" turns all the bits to 0. You can then write to the sector to turn some of the bits from 0 to 1. You CAN re-write the same sector, but ONLY if you are turning on bits, and not turning any off. NAND cells wear out from performing too many erases, not reads or writes. Note that most sectors you aren't usually just turning on bits, so you typically have to do an erase and then write, but not always.

Now as for TRIM. For most disk operations, when a sector/page/whatever is no longer logically used, it isn't told to write 0's to the sector, it's just marked as unused in whatever filesystem you are using. The disk controller doesn't understand filesystem structures so it doesn't know which sectors which are filled with unused data (0's and 1's) are logically unused, so it can't pre-erase them. Without TRIM, eventually all "unused" sectors get filled with garbage, and when the OS asks for the sector to be used again, it must first do an erase before it can write to it, which kills the drive performance on writes. With TRIM, the OS tells the controller than these sectors are no longer in use, and you can erase them when you get a chance, so that I can then write to them later by just doing a write operation instead of an erase and then write.

Hope that makes sense.

Re:Call me old fashion

[Anonymous Coward]

I purchased a Samsung 840 250GB (non pro edition) for my HTPC and dedicated it as the TV Record drive for Windows Media Center. Who cares if it "breaks" in a couple of years. The things are so cheap these days, and the recorded TV data on the drive only has a 2 week turn around before it's encoded by MCE Buddy and stored on a mechanical drive *if* it's not deleted before then.

The speed benefits, power savings & silent operation greatly out weigh "The Sky is Falling!" when it comes to SSD drives.

For people that use their computers *full stop*, get a damn SSD!

Re:Call me old fashion

[WuphonsReach]

Only the "enterprise" level drives. The Intel DC S3500 series, for one. Most consumer level drives do not have capacitors.

Re:Call me old fashion

[ShekharDharShekhar]

Nice story! Do you have any valid trustworthy content to back up your version of the story? As far as I know, "panic mode" issue with sandforce is a history & does not exist with any latest sandforce based SSD. So, move on dude & it is not a good idea to get into too detail even though their products are functioning good!

Re:Call me old fashion

[Anonymous Coward]

It might become a problem for people that put their Windows machine into "hibernate" state several times a day, depending on the amount of RAM in the machine.

Re:Call me old fashion

[Anonymous Coward]

Look up for "exponential decay". HDD failures follow it pretty well, so you only need enough data to fit an exponential curve and you'll know for sure where more than 50% of the drives will have failed. It is exactly the same mechanism that lets us know, for example, that the half life (i.e, the time expected for 50% of a given group of objects to decay) for 238-Uranium is 4.5 billion years, even though we haven't measured 4.5 billion years of decays. Or, even better, 115-Indium has a half-life of 115 trillion years, and we know that number to be right (give or take a few percent) even though the universe hasn't been around for much more than 0.01% of that time.

Re:Call me old fashion

[tlhIngan]

How many effective READ/WRITE cycle can the chip in SSD perform, before they start degrading ?

There are user-done studies on such matters and some of them are quite impressive - to the point where you'll scrap the computer first before encountering failure.

The main reason why SSDs fail prematurely is their tables get corrupt. An SSD uses a FTL (flash translation layer) that translates the externally visible sector address to the internal flash array address. FTLs are heavily patented algorithms and there are many of them. The FTL also handles stuff like wear levelling, bad blocks and garbage collection. It's no surprise that the FTL has an impact on SSD performance.

All this is typically stored in some area set aside for it because scanning the entire array is much too slow (takes many seconds, slowing down the power on boot). In addition, the use of stronger ECC algorithms uses up the available spare area of the flash leaving none for the FTL to manage.

So on power up, the SSD controller gets the tables from flash and starts operating extremely quickly. Of course, the problem is what happens when the tables are corrupt - they're usually held in DDR SDRAM for speed, but they need to be synced to the media periodically, and that's where problems happen (because you don't want to wear out the table blocks, as well... so you need the FTL to manage *that*...).

So if you manage to pull power to the SSD at the wrong time, there's a strong chance it'll corrupt the tables because it's in the midst of syncing them to media.

Good SSDs avoid this by structuring their writes, or more commonly, using onboard capacitance to provide emergency power which can flush out the memory buffers to storage quickly. On some SSD teardowns, you can see a huge row of 10+ tantalum capacitors just for this purpose.

Re:Call me old fashion

[Anonymous Coward]

not as a study, but anecdotal evidence, I've been using sets of kingston SSD and sets of samsung SSDs as RAID0 configurations for photoshop and ptgui scratch, meaning the full capacity of all drives have seen so many R/W that it's not even funny anymore (probably more than 10000 times, working on some monster panoramas). the SSD are now 2-3 years old, and still have good speed and no obvious failures from wear.

but whoever thinks spinning disks are better for their task, should A) stick to them and B) don't bitch about spinning disks not being fast enough

spinning disks are still better for large long time storage (TB to PB of data)

Re:Call me old fashion

[Lord Chaos EOG]

I've been running SSDs as both boot drives, server drives and RAID drives for some 2-4 years and only 1 has ever failed out of the dozens of drives.

Re:Call me old fashion

[cthulhu11]

... and that the workload was write-only.

Re:Call me old fashion

[Whorhay]

Also stuff like your temporary files, libraries (documents, music, pictures, videos), page file and such can all experience quite a bit of churn and hog space. So what I've done is symbolically linked all of the directories that I expect to go through a lot of data churn to a Hard Drive. All of the stuff I expect to be pretty stable like the OS and installed programs go on the SSD, which is then backed up to a Hard Drive.

Re:Call me old fashion

[Whorhay]

From my understanding of HDD technology part of the issue would be that HDD's are most likely to fail within their return period on their initial powerup or very early in their life. So for the most part if your HDD fails it will be early enough in it's usage life to send back, whereas with SSD's the problem is that they have a definite usage limitation. The SSD is very unlikely, compared to the HDD, to be dead on arrival or brick early in it's service life. So the SSD will get used and eventually wear out but it will probably be past it's warranty period and so not be returned.

Also proper use of SSD's is important. If you fill the drive up it will last considerably less time than if you use it sparingly. And of course that then brings up, if I can't use the drive why am I buying it?

Re:Call me old fashion

[jafo]

Worse, a lower rate kind of is *MORE* indicative of a load that needs an SSD rather than *LESS*. SSDs are *VERY* good at random seeks and you can easily saturate a spinning disc at 400KB/sec or less worth of random I/Os. (assuming 10ms average access time, or 100 accesses/sec).

If you are streaming a lot of data, an SSD is "only" around 4x faster than a spinning disc. If you are doing random I/Os, an SSD is more than 100x faster.

Missing disclaimer

[Anonymous Coward]

Wow, MojoKid's lines read like an advertorial. So where's the disclaimer?
I'm missing the "Im not the author of the linked article, or part of the publishing organization, and I do not receive kickback fees for the extra slashdot traffic."

Still put off by price. :(

[Anonymous Coward]

I bought a samsung 840 128 gig drive( not pro ) and Iove it. Newegg had a deal going for like $109 bucks and I jumped on it. It's fast enough and big enough for any OS. I use another rotational drive( seagate 2 tb's ) for my vm's, progs, and files. I really am satisfied with 128gigs using a larger secondary drive for everything else. The price on SSD's really make you think just how much sp ace versus perfomance do I need. I think I'm going to wait a little longer for the prices to come down.

Note: Anyone here about any programs like spinrite or other for drive recovery for SSD's? I'm really curious. Thanks. - A.B

Re:Still put off by price. :(

[Gaygirlie]

Note: Anyone here about any programs like spinrite or other for drive recovery for SSD's?

There is no such a thing except for the few that are just trying to dupe you into giving them money. Why? Well, as long as the drive's controller itself is working and the drive's internal state isn't corrupted you can read the cells indefinitely. You cannot write to cells indefinitely, but all major manufacturers these days promise that even if all the cells failed in the whole drive you should be able to read them. On the other hand, if the drive's controller goes bonkers or the internal state gets messed up there is *no software whatsoever* that can fix it. You'd have to open the drive and work with the actual flash-chips themselves in the hopes of recovering your data, and due to the nature of SSDs where the cells can be re-located at any given time for wear-levelling purposes that'd be one helluva task.

Now, if your filesystems or such get messed up any tool that works on mechanical HDDs works just fine on SSDs. There is no difference.

Re:Still put off by price. :(

[Joce640k]

I bought a samsung 840 128 gig drive( not pro ) and Iove it.

Yeah, we all like the look of the "hot" axis on the graph.

We're worried about the "crazy" axis. Most of us have a long term relationship with our data.

(for those who don't know: http://www.codinghorror.com/blog/2011/05/the-hot-crazy-solid-state-drive-scale.html )

nb. I've been using a 40Gb Intel SSD as my boot drive for a couple of years, it's still going strong AFAICT but there's not too many writes to that drive (swapfile and $home is on a velociraptor).

Re:Still put off by price. :(

[Immerman]

>all major manufacturers these days promise that even if all the cells failed in the whole drive you should be able to read them
and you should take such promises as having the same integrity as all other marketing claims - i.e they're probably not blatant lies.

A Samsung 840 endurance test posted in a comment above: They ran for ~3000 erase cycles until encountering the first unrecoverable read error at which point they declared the drive "dead" - it still seemed to work, but data had already been lost and more losses were inevitable.
http://uk.hardware.info/reviews/4178/10/hardwareinfo-tests-lifespan-of-samsung-ssd-840-250gb-tlc-ssd-updated-with-final-conclusion-final-update-20-6-2013

There's also unpowered data retention to consider - you're storing your data as a partial charge in a capacitor - 1 bit/cap (2 levels) for SLC, 2bits (4 levels) for MLC, or 3bits (8 levels) for TLC. And every one of those capacitors starts losing charge the instant it's written to - it does so very slowly, and typically starts out lasting for years (IIRC), but as the caps start wearing out and the charge levels get "fuzzier" that number can eventually drop to days, though supposedly that's typically well past the drive's rated erase cycles (i.e. folks hammering it 24-7 and hitting 30,000x erase cycles)

But yeah, you're absolutely correct that data recovery tools aren't going to be much use for an SSD where data is stored in discrete "bins", and either it's there or it isn't. Unlike a HDD where data is stored as magnetic "footprints" on a continuous media where even intentionally erasing it will tend to leave a portion of the print behind in the gap between tracks, which can then generally be recovered through repeated head repositioning and statistical analysis.

Not paying for TLC at that price

[Anonymous Coward]

These are "1 year and out" drives. 1000-2000 P/E cycles will fly by in under a year in most prosumer applications. I'll pay hard disk per GB prices for a TLC SSD, but no more.

Re:Not paying for TLC at that price

[beelsebob]

Except if you actually bothered to educate yourself, you'd find that at the capacities samsung is offering you, if you write to them at 10GB a day, every day, they'll last entirely respectable times (12,23,47,94 years respectively for 120,240,480 and 960GB drives).

Re:Not paying for TLC at that price

[Anonymous Coward]

And how am I supposed to edit my HD videos?

10GB??? That's a short video indeed... and if I am editing it I will be using more space... somehow an SSD doesn't sound so good.

I make a lot of videos, and it's not worth killing a drive in months. Make it magnetic media for me, thanks.

Re:Not paying for TLC at that price

[Anonymous Coward]

my 50Gb dev workstation ssd has transfer over 450TB in is lifetime (3years). That drive is from a product line that have a poor reputation for reliability (OCZ-Vertex2). You should really try to edit video on a SSD. If the speedup you gain is comparable to the speed up I got compilation time, you will find it magic. By the way my older dev drive was a WD Raptor 100Gb 15K.

use a XEON next time toms hardware!

[Anonymous Coward]

lol. it's nice to have somebody do all the new hardware testing thru the years.
I suppose it can get abit "tiring" and "boring" after so many years.
but i must say that a z68 chipset and a core i7-2600K are a really bad TEST bed for I/O
tests, non?
the z68 and i7-2600K only have 16 pci-e lanes and pci-e version two.
the "DMI" is 5 GT/sec only.
seriously,it be abit more of a "test" if the test bed weren't constraining the whole I/O.
i'd recommend a beefy recent XEON (which today has a sandy bridge-e core) and a socket
2011 chipset.
this should FOR SURE eliminate all I/O CPU and chipset bottle necks. afterall the xeons for socket 2011
do 40 pci-e lanes and pci-e version 3 and QPI.
i'm sorry if i just blurted out a well keep secret : )

Re:use a XEON next time toms hardware!

[Anonymous Coward]

oh maybe also use the windows SERVER version.
i could imagine M$ doing their little "pay-more" tricks to
artificially limit speeds on desktop version of windblows. (assumption, but should try/test)
also don't install dotNET since it seems to do some
unique hardware compiling in the background to flatten out
the responsiveness of the system so that "slow" chips
and "fast" chips become equal, so that all vendors with
the gazillion hardware combinations
feel regularly "responsive" ... muha^^mono^^haha^^poly^^haha

Re:use a XEON next time toms hardware!

[beelsebob]

You're talking about testing a device that doesn't even saturate 3 PCIe lanes, and complaining the test bed "only" has 16? Really?

Re:use a XEON next time toms hardware!

[Anonymous Coward]

proof is in eating the puddling, sir.
everything is theoretical until physically test, non? oui?
anyways, my anecdote is that a i5-760 will max out at 250 MB/sec.
the ssd is connected via correct sata 3 cable and (advertised) sata 3 port.
the same ssd has done more then 300 MB/sec on another chipset/cpu platform,
as test by toms hardware ...
the ark says that GT/s for the i5-760 is 2.5 GT/sec.
coincidence? i guess not (250MB/sec = 2.5GigaTransfers/sec = 2500 MegaTransfers/sec (= 2500/10)).
my assumption is, that the chip responsible for doing the SATA 3 (in my case), is connected only
via 1(one) pci-e lane, maybe?
furthermore, if during the test ANYTHING else connected via pci-e bus (sata (raid?), video-card, sound, networkcard(s), etc.)
wants to send some data, then at least one pci-e lane is used (and becomes unavailable to the other stuff connected).
connected a GFX titan to a system/cpu/chipset that IN TOTAL only has 16 pci-e lanes
is a very strange thing to do, because the grafic card all by itself can/will/want to use
16 lanes.
to be honest i don't really understand this whole pci-e 2/3, DMI, QPI, HyperTransport thing (and it will
become more difficult to understand because everything is moving inside the CPU (northbridge, memory controller, video)).
good tests will become even MORE relevant in the future.
as a last note: I assume that the GT/sec value is for 1(one) pci-e lane.
if this is wrong and it is a TOTAL value for ALL advertised pci-e lanes (16 in toms ahrdware test case) then ...*holy bottleneck batman*.
further enlightenments are welcome: )

Re:use a XEON next time toms hardware!

[Anonymous Coward]

Another armchair engineer talking about things he doesn't understand. Used to be cute, now it's just tiring.

Re:use a XEON next time toms hardware!

[Anonymous Coward]

i see u cannot explain it either.
of course knowledge is money and can't just be thrown around. afterall the colleg degree has to be paid off
and one cannot just willy-nilly give away valuable knowledge for free.

Re:use a XEON next time toms hardware!

[gl4ss]

excuse me, is this the retard thread?

NAND, or Exclusive NOR?

[smittyoneeach]

The argument went on for about half an hour in EE lab, before the teacher came along and announced: "Yes."

Was the RAPID sw used throughout the test

[Marrow]

One assumes this is windows software. Did the competing drives have their drivers installed too? I would want to see its performance without drivers installed and used as a plain SATA drive. And I would like to see with and without RAPID numbers.

Is RAPID a sophisticated buffer cache that is doing lazy writes to the SSD?

Re:Was the RAPID sw used throughout the test

[fgouget]

I would want to see its performance without drivers installed and used as a plain SATA drive. And I would like to see with and without RAPID numbers.

Is RAPID a sophisticated buffer cache that is doing lazy writes to the SSD?

There you go: http://www.anandtech.com/show/7173/samsung-ssd-840-evo-review-120gb-250gb-500gb-750gb-1tb-models-tested/5

I wonder if RAPID could bring such huge performance improvements on Linux too, or if this just means the Windows cache sucks. Because from the article I still don't see eactly what RAPID does that the OS's cache shouldn't do already.

Re:Was the RAPID sw used throughout the test

[Marrow]

I keep hearing these drives are doing compression. Maybe the driver offloads the compression onto the cpu.

Re:Was the RAPID sw used throughout the test

[fgouget]

I keep hearing these drives are doing compression.

Only the SSDs using a SandForce controller use compression so it's not the case here.

from your kind link, it looks to be doing lazy writes.

The OS's cache should already be doing something like that. However the benchmarks normally force it to flush to disk at key points to ensure they test the performance of the disk and not the cache. So maybe the RAPID driver ignores the flush commands in some circumstances?

Re:Was the RAPID sw used throughout the test

[Anonymous Coward]

Sounds like RAPID fully adheres to the OS and application Flush commands -

"To avoid caching writes in DRAM for too long, RAPID transfers them to the SSD every time the Windows write cache is flushed."

http://techreport.com/review/25122/samsung-840-evo-solid-state-drive-reviewed/3

Is anyone building home SANs out of SSDs yet?

[swb]

In the 2-5 TB range?

I previously would have maybe wanted this but not been willing to spend the money or expose my storage to disk failure with consumer SSD.

I'm thinking now it's getting to the point where it might be reasonable. I usually do RAID-10 for the performance (rebuild speed on RAID-5 with 2TB disks scares me) with the penalty of storage efficiency.

With 512GB SSD sort of affordable, I can switch to RAID-5 for the improved storage efficiency and still get an improvement in performance.

Re:Is anyone building home SANs out of SSDs yet?

[dfghjk]

A RAID array is not a SAN and "home SANs" are moronic. Arrays of SSDs are not a new idea. I've been using SSD arrays longer than I've been using single SSDs.

Applying traditional RAID redundancy techniques to SSD is stupid.

Re:Is anyone building home SANs out of SSDs yet?

[swb]

A RAID array is not a SAN but I have yet to see an actual SAN configured JBOD only, if it's even an option. I'm not sure how you would aggregate the storage of single SSDs without RAID.

And I don't know what's moronic about home SANs, mine has ~7 TB storage and volumes exported via iSCSI and NFS to 3-4 systems.

What's wrong with RAID redundancy techniques for SSDs? Between the aggregation required for larger LUNs, I would think you would want to hedge the risk of a device failure.

Re:Is anyone building home SANs out of SSDs yet?

[Anonymous Coward]

NFS makes it a NAS, not a SAN.

captcha: cheapens

Re:Is anyone building home SANs out of SSDs yet?

[Anonymous Coward]

There's nothing wrong with putting SSDs in RAID, and home SAN/NAS is not "moronic". dfl;hasdo has no idea what he's talking about, like an ever increasing number of /. posters, sadly.

Also most people don't write as much as they think

[Sycraft-fu]

Usually, once you have your computer set up with your programs, you don't write a ton of data. A few MB per day or so. Samsung drives come with a little utility so you can monitor it.

As a sample data point I reinstalled my system back at the end of March. I keep my OS, apps, games, and user data all on an SSD. I have an HDD just for media and the like (it is a 512GB drive). I play a lot of games and install them pretty freely. In that time, I've written 1.54TB to my drive. So around 11GB per day averaged out, though realistically about 500GB of that was done the first day, since I installed the system, put the apps on, then changed my mind with regards to UEFI boot, and reinstalled the system.

I think some people believe that since they have a lot of data, they must write a lot and thus the write limit would be problematic. However the data you have is usually largely static. Your delta is fairly small and thus not that problematic to flash.

So while I wouldn't want to use TLC flash drives in a backup system or something, there really isn't an issue in a desktop. If you do have an atypical situation where you have very write intensive workloads, well you can always have a magnetic (or SLC flash) secondary disk for them. But for desktop usage, you just aren't going to write that much to your disk.

Hot vs Crazy

[bdwoolman]

Here's the thing. SSDs are now more reliable than when this guy logged this report.

But are still maybe not as steady Eddie as a good-quality HDD. But we still want them because having an SSD boot drive changes the whole computing experience due to their awesome speed. And since we are good about backups (Are we not?) we can be relaxed as we ride the SSD smokin' fast Roller Coaster. SSD or HDD then what's the problem if we have data security. Both are gonna FAIL. So what if Miss SSD stabs me for no good reason? It was a helluva ride, Bro. And well worth the stitches. I do wish SLC NAND was not priced out of reach, but, hey, when it comes to hottness we take what we can get. Right?

Okay. This is Slashdot we get no hottiness...no hottiness at all.. No no no hottiness. It's pathetic really. ....

Re:Hot vs Crazy

[Anonymous Coward]

HD
SSD
So the *best* 2TB HD had a 5x higher average failure rate than Samsungs SSDs...

Re:Call me old fashion@beelsebob

[Anonymous Coward]

Do your math... return rates of 5% "of all drives sold" (with less SSD's sold per year than HDD's) is a much higher return rate...(or did you mean "of all drives of the same type?")

I don't buy it when you discount a large number of anecdotal experiences with short lived SSD's... add up all the anecdotes and eventually you reach reality when most of the stories match...

In the end, speed is nice, if you need it... and if you care about your data you will have a backup or three...

Personally I buy new drives to use as offline backup every year... (For some reason I still need more and more space, despite deleting old data/programs that have lost relevancy) Currently 3TB for Data and 2TB for Applications... At this rate I should need near 20TB for a full backup by 2018, which is fine.

If rotating magnetic media were to go away, I don't think I would be bothered.... I'd use SSD's in the same fashion, turn on once a month or so, backup, turn off... The lifetimes of such data should be decades, something no magnetic based drive can match.

I expect the industry will not phase out spinning media until there is a more reliable medium for consumers.

It's AM radio I am HOPING will be phased out and re-purposed very soon... currently it's just a bunch of noise. ;-)

What I find interesting in the OP: The title says drives tested, yet the post has no numbers for read/write speed or metrics other than price per GB....

Been here long enough to know many /. titles don't match the articles well at all...

Cheers, everyone...

EXACTLY what's held ME back from Flash SSD

[Anonymous Coward]

"How many effective READ/WRITE cycle can the chip in SSD perform, before they start degrading? Has there been any comparison made in between the reliability (eg read/write cycles) of old fashion spinning-plate HD versus that of SSD?" - by Taco Cowboy (5327) on Sunday August 18, 2013 @05:24AM (#44598983)

See my subject-line: When I see conclusive & consistent evidence of Flash based SSD's outlasting/showing superior overall longevity + performance over decades? Then, I'll convert over.

Until then, however? I'll keep using the setup I have for @++ decades (since it has lasted for that timeframe across 5 different systems that still run to this very day using HDD's + a "True SSD" NOT based on Flash (DDR-2 currently)):

Since 1992 or so, 1st using separate HDDs (slower seek/access by FAR) & then using software ramdisks per the list below (on a MS-DDK based one I wrote in fact, on how I apply them):

Then applying Software-Based Ramdrives to database work with EEC Systems/SuperSpeed.com to do MOST of the following in the enumerated list below (minus pagefile placement on software based ramdrives, I just didn't have enough RAM onboard to justify it was all though).

Then, once I got a CENATEK "RocketDrive" (PC-133 SDRAM based) with 4gb onboard, & now currently a Gigabyte IRAM 4gb DDR-2 based hardware ramdisk card??

That all changed (to a better way to do it, via dedicated RAM in hardware, NO "flush" from memory pressure by other processes writing ram & the memory mgt. subsystem responding to it, flushing software caches & their FIFO queue like algorithms).

I move the following off my wd Velociraptor sata II 10,000 rpm 16mb buffered harddisks that are driven off a Promise Ex-8350 128mb ECC ram caching raid sata 1/2 controller (which defers/delays writes via said cache, & also lessens physical head movement on disks & this is where I am going to make it even faster via lessening its workloads, read on & reduces fragmentation as well in the same stroke - "bonus") onto my 4gb DDR2 Gigabyte IRAM PCIExpress ramdisk card:

A.) Pagefile.sys (partition #1 1gb size, rest is on 3gb partition next - this I didn't do on software ramdrives though)
B.) OS & App level logging (EventLogs + App Logging)
C.) WebBrowser caches, histories, sessions & browsers too
D.) Print Spooling
E.) %Temp% ops (OS & user level temp ops environmental variable values alterations)
F.) %Tmp% ops (OS & user level temp ops environmental variable values alterations)
G.) %Comspec% (command interpreter location, cmd.exe in this case, & in DOS/Win9x years before, command.com also)
H.) Lastly - I also place my custom hosts file onto it, via redirecting where it's referenced by the OS, here in the registry (for performance AND security):

HKLM\system\CurrentControlSet\services\Tcpip\Parameters

(Specifically altering the "DataBasePath" parameter there which also acts more-or-less, like a *NIX shadow password system also!)

* All of which lessen the amount of work my "main" OS & programs slower mechanical hard disks have to do, "speeding them up" by lessening their workload, fragmentation, and speeding up access/seek latency for the things in the list above too.

APK

P.S.=> HDD's concentrate on program &/or data fetches that are still hdd bound (& not kernelmode diskcaching subsystem cached in 4gb of DDR3 system ram here either yet) done on a media that has no heads to move, & thus, more mechanical latency + slower seek/access as you get on hard disks + reduced filesystem fragmentations due to that all, also & it works!

... apk

Why?

[Sycraft-fu]

You'd need a better network to have any use. A modern 7200rpm drive is usually around the speed of a 1gbit link, sometimes faster, sometimes slower depending on the workload. Get a RAID going and you can generally out-do the bandwidth nearly all the time.

SSDs are WAY faster. They can slam a 6gbit SATA/SAS link, and can do so with nearly any workload. So you RAID them and you are talking even more bandwidth. You'd need a 10gig network to be able to see the performance benefits from them. Not that you can't have that in your house, but you don't because it is damn expensive. Lacking that, you'll notice very little improvement over magnetic drives.

Also to be technically correct (the best kind of correct) you probably don't have a SAN in your home. A SAN is a separate network, purely for storage devices, not connected to your LAN. It is a FC/FCoE/iSCSI/whatever backend that your storage devices talk on, and then there's a different network that your clients use to talk to the storage server (which is on both networks).

Re:Why?

[DigiShaman]

Depends on what he's doing over that SAN via gigabit link. If he's fetching large files, I agree. If it's the latency he's trying to reduce (say, SQL queries), SSDs will help immensely; even over a limited gigabit link.

Re:Why?

[swb]

The core reason why is to avoid the shitty reliability nightmare that contemporary mechanical HDDs represent and to get a bump in performance.

My current environment is 6x2TB Seagate 7200s in RAID-10 and I find with a virtualization workload good throughput dies off pretty quickly. Sure, a single contiguous large write or read can saturate the link, but 2 ESXi hosts and 6-8 busy VMs really brings up the latency and brings down the performance.

After setting this up last fall, I find I made it bigger than I really need and that it doesn't deliver the performance I'd like. With at least one dead disk in a year, 512GB SSDs (5xRAID-5, with one hotspare) seems like a reasonable performance boost for the kind of active disk space that makes sense.

Now I just need to swallow the $1600 and headache to go with SSDs.

Re:Why?

[symbolset]

Virtualization really eats up the IOPs. Generally you see a huge increase in VM performance going with SSD backed storage. It's also great for heavy video editing.

Re:Why?

[swb]

Obviously with FC you're talking a physically distinct network, although I'm sure someone, somewhere uses FC for data networking.

With iSCSI, most places I do work at have converged network infrastructure with iSCSI traffic and "normal" network traffic on the same hardware; the separation of storage traffic is purely logical (VLANs) and not based on a specific physical isolation into discrete hardware. This means less equipment and more flexibility in getting access to storage.

Only in really large setups do you find a lot of dedicated Ethernet for iSCSI, and that's more about maintaining sanity of the physical layout, usually with traffic split over multiple switches, LACP links between and LACP uplinks to the core.

UPSes are usually near 100% efficient

[Sycraft-fu]

Most UPSes these days are line-interactive. That means they are not doing any conversion during normal operation. The line power is directly hooked to the output. They just watch the line level. If the power drops below their threshold, they then activate their inverter and start providing power. So while their electronics do use a bit being on, it is very little. The cost isn't in operation, it is in purchasing the device and in replacing the batteries.

That aside SSDs don't have problems with it (it was a firmware bug, Samsung fixed it) and if your data is important, you probably don't want to rely on your journal to make sure it is intact. When you get in to real high end, reliable, storage, power backup is a big thing. Our Equallogic has dual full redundant power supplies on all units, which they wanted plugged in to separate circuits (one is line only, one is generator backed), redundant controllers, and the NAS has internal batteries backing the cache in case of power failure, ones that last quite awhile.

There's a big difference between "a journal that means the filesystem isn't in an inconsistent state (usually)" and "a setup where one doesn't lose any data."

If you are concerned about efficiency costing you money in your computers (it likely costs less than you think) then your PSU is the place to look. If you didn't specifically buy a good one, it is probably 80% or less efficient. You can get them a bit above 90% if you try, and match them to the load.

Re:UPSes are usually near 100% efficient

[jamesh]

That aside SSDs don't have problems with it (it was a firmware bug, Samsung fixed it) and if your data is important, you probably don't want to rely on your journal to make sure it is intact.

If it was a firmware bug and has been patched then the point is irrelevant, but journals wouldn't have made it better anyway. Because of wear leveling etc, stuff gets written all over an SSD basically at random. And when a block of data gets full the SSD logic will move anything useful in that block to other places on the disk and then erase the original block. If the power failed and a firmware bug meant that pointers weren't saved then you've essentially taken a shotgun to your disk and potentially a whole erase block sized unit of data may be gone. Journals aren't the answer to all problems.

And neither are UPS's for that matter. More and more home computers are coming with SSD's and even if the shop managed to do a UPS upsell, most cheap UPS's don't last very long at all, and the home user isn't going to know until the power fails the first time.

Re:Call me old fashion@beelsebob

[beelsebob]

I don't buy it when you discount a large number of anecdotal experiences with short lived SSD's... add up all the anecdotes and eventually you reach reality when most of the stories match...

The problem is that for each person with anecdotal evidence of SSDs failing, there's 200 other people not commenting about their entirely working SSD.

Only some do

[Sycraft-fu]

New Intel drives do, as they use the Sandforce chipset. However Samsung drives don't. Samsung makes their own controller, and they don't mess with compression. All writes are equal.

Also 14TB sounds a little low for a write limit. MLC drives, as the XM25 was, are generally spec'd at 3000-5000 P/E cycles. Actually should be higher since that is the spec for 20nm class flash and the XM25 was 50nm flash. Even assuming 1000, and assuming a write amplification factor of 3 (it usually won't be near that high) you are talking 52TB if the drive has no internal overprovisioning, which it probably does.

As an example, AnandTech tested a Samsung 840 TLC drive. The 250GB drive was able to take about 266TB of incompressible data, which translates to a bit more than 1000 P/E cycles.

If you have a high write workload, their MLC drives aren't that much. A 512GB 840 Pro drive will run you like $450. That should get you somewhere in the realm of 1.5PB of writes before it fails, maybe more.

Idiotic acronyms?

[K. S. Kyosuke]

Real-time Accelerated Processing of IO Data

Nope, definitely not contrived at all.

TLC not worth it yet

[JDG1980]

I'd be willing to consider TLC despite its drawbacks if the price was considerably lower than with MLC-based drives, but that's currently not the case. The Samsung 840 EVO costs about $185 for the 250GB model, while the 840 Pro (using MLC) is about $230-$250. So we're talking about 75 cents a gigabyte for TLC, and about a buck a gigabyte for MLC. I'm willing to take the 25% cost hit for far better endurance. In my opinion, TLC really needs to get down to 25-40 cents a gigabyte before it would be worth it. If we could get a 640GB TLC drive in the $160-$200 price range, then that would be worth going for. But the current offerings? No.

A different technology is available

[symbolset]

I hate to see this discussion go entirely to the "wearout" issue. Clearly there are some posters here heavily invested in spinning disk. There are more exciting flash technologies in the pipeline.

Samsung has a new flash technology for the Enterprise called 3D V-NAND. By using 24 separate layers of flash on one chip they can keep the feature size up and still keep pace with storage density. They believe they can go to hundreds of layers. There is talk of a 384GB single chip for smartphones and tablets. Not Gbit, Gigabyte.

But no, go back and forth some more about wearout rates.

Re:A different technology is available

[Anonymous Coward]

Slashdot mentioned said V-NAND fairly recently:

* http://hardware.slashdot.org/story/13/08/06/1423215/forget-flash-resistive-ram-crams-1tb-onto-tiny-chip

And for those who want more info (TL;DR version is V-NAND is basically memristor-based, which is something all the industry players should have been looking at instead of NAND from the beginning):

* http://semimd.com/blog/2012/06/27/applied%E2%80%99s-avatar-tackles-v-nand-dielectric-etch/
* http://semimd.com/blog/2011/07/26/vertical-3d-nand-may-be-pulled-in-to-2013-2014/
* http://en.wikipedia.org/wiki/RRAM

Re:A different technology is available

[symbolset]

Whoa. An AC with useful information and references? Who let you in here?

(No, I was not that AC.)

Re:A different technology is available

[symbolset]

Reviewing the links now though I see no reference to Samsung, nor their multilayer developments that are a true departure from the traditional methods, nor product pipeline info. It's all "Crossbar". Also no reference to HP, who had some discovery in the memristor area. Maybe it's time to post a new article. When we see this theoretical stuff we usually think "a decade out, if ever."

Multilayer is a really big deal. One problem with Moore's law is that it has heretofore existed in flatland - dimensions X and Y. While FinFet and other methods have turned transistors on their side, technically moving two or three units into dimension Z that is not a real departure from flatland mode. Moving thousands or millions of units into dimension Z is a really, really big deal as X and Y begin to show problems with quantum effects.

Hat tip to you still AC: this is where the action is at and you've clearly got your eye on it.

posted too soon

[Marrow]

from your kind link, it looks to be doing lazy writes.

Re:Call me old fashion@beelsebob

[Anonymous Coward]

I'm on my 5th SSD. Every time was an upgrade, never a failure.