Looking to buy a new Thinkpad T Series

[Navck]

Thanks ThinkRob, but I want to know about zhenya's drives, especially if they're two years old as he claims.

Zhenya, could you use remote desktop or VNC to take a quick snapshot of the office machines for me?

[ThinkRob]

Well I've got a first-gen X25-M that's about that old (shipped in 2008, with the original firmware). It's currently in a machine that I've lent out, but I'll have it back in my possession around Christmas-time. If you want to drop me a PM sometime in the middle of December, I'll be more than happy to take a look at the SMART stats for it.

The first-gen drives did indeed have a firmware upgrade for them, IIRC to help prevent the performance drop caused by hammering the drive with very large quantities of small, random writes. Interestingly, the G1 drives actually resist said degradation better than the G2 drives -- but since they don't have TRIM they're only really superior if they're to be installed in a system that doesn't support the TRIM command.

It's important to remember that we're talking about a small difference in performance though. Even in their "degraded" state, both generations of the the X25-M will provide far better random read/write performance than even the fastest mechanical drives.

[zhenya]

I'm now in the field for work. I have my laptop with me now - but it is a relatively new one. One of my co-workers here has a drive that has been in his computer for 8-10 months - I'll see if I can check his later. It will probably have to wait until I get back in the office in a couple of weeks or more before I can check any of the machines that have had the drives in them for longer - they aren't in our servers, so I don't have remote access to them.

Anyhow - here's the values from my drive:

[zhenya]

Ok, I got my co-worker who has the oldest of the drives to install the toolbox and send me a screenshot this morning. Here you go:



Anything else?

[Navck]

On the machine that your coworker uses, could you tell me what type of activity would take place on it?
The power cycles count looks remarkably low for the power on hours.

[zhenya]

That drive was originally mine, as it was the first one we purchased and I used it for testing purposes before accepting the drives for use by others. It was installed in my z61t laptop, and as this was pre-Windows 7, there were a bunch of installs done as I attempted to optimize XP on it, before ultimately settling on Vista, using it as my production laptop for approximately 6 months. It then went into my co-worker's development workstation where it has lived ever since. I rarely reboot my laptop if I can help it, and I know he leaves his workstation running most of the time.

[Navck]

"Development workstation?" It doesn't seem to get that much disk activity, especially for a ~ 2 year old drive.

[zhenya]

I don't know what to tell you man, it's used 8 hours a day, 5 days a week, just like the rest of our workstations.

[Navck]

But what is "done" on it, scripting, compiling, what?

[zhenya]

When it was mine, it was used for everything from use in the field connecting to our tools that we operate on client sites, to sys-admin tasks, managing our network, running and testing VM's, to lots of basic office work - writing reports, giving presentations, etc.

It's now on a software developer's machine where he writes code for our internal use on it. I'm not the one who uses it now, so I can't give you specifics, but it would include writing, testing, and compiling code. Plus all the regular stuff people 'do' on a computer.

Keep in mind that this drive is only 80GB in size - so it means that this drive has been written over approximately 47 times in its life.

[Navck]

Even with "limited endurance" NAND which means you'll be at 1k cycles... 87 isn't a lot
And with older, larger lith process MLC cells... 87 is nothing compared to 10k-100k.

It doesn't seem like that X25M G1 gets "pushed" to its limits in that machine.

[zhenya]

Even with "limited endurance" NAND which means you'll be at 1k cycles... 87 isn't a lot
And with older, larger lith process MLC cells... 87 is nothing compared to 10k-100k.

It doesn't seem like that X25M G1 gets "pushed" to its limits in that machine.

Well, take what you want from it, but I can't see how this could illustrate my point better. It's been in constant use for nearly 2 years, yet has suffered only ~4% 'wear.' The fact that a mechanical drive would last much longer doesn't matter, because these drives will be retired well before they have worn out, yet the users will have enjoyed the benefit of increased speed for many years in the meantime.

If daily use as a development machine at a company that has been extremely successful for over 30 years is what you consider light usage, so be it. To me, these drives are filling a need, and doing it spectacularly well. That's all I've ever claimed.

[Navck]

I don't know, I have a friend who has a bunch of blade servers for his own use when he does simulations for his engineering projects, he was able to go through some sample SSDs within a week just from dumping the results on them.

What I've observed would be taxing to any drive as much as somebody's grandparents browsing the internet with a very large browser cache + then some and maybe saving a few images in terms of "loading" the drive...

That doesn't come close to writes that can happen from copying RAW files from a camera then processing them or re encoding videos...

I thought the X25M G1 only existed for ~ 2 years?

PS: The 60GB Hitachi in my T43 has close to 21k power on hours, still running strong as with the T43 in my personal "fleet."

[zhenya]

I thought the X25M G1 only existed for ~ 2 years?

Yes, I think so to. Not sure what you are asking here?

PS: The 60GB Hitachi in my T43 has close to 21k power on hours, still running strong as with the T43 in my personal "fleet."

T43- that means your drive is probably ~ 5 years old. Extrapolate out the use from this drive, and at 5 years old it will have ~ 16,000 power on hours. That means your drive is seeing ~ 30% more use than this one. Big deal. And it's not surprising for a personal computer compared to one only used during business hours 5 days a week.

So the question from me is, at what point would you consider this drive to have been 'tested' enough? Assuming a relatively linear curve between power on hours, host writes, and media wear indicator value, if this drive had seen 5x as much use, meaning it had seen 30,000 power on hours and 20TB of writes, yet showed a media wear indicator value of say, the mid 80's, would that be acceptable? Where is that point?

[Navck]

My T43 is 5.x years old, turning 6 this summer.

I'm saying that the drives you have practically have the wear level of grandparents on a workstation and at that point, might as well advise you to buy some cheap 160GB HDDs then defragment them when you leave work to maintain similar levels of performance due to Win7 being able to cache to RAM anyways. The typical user also spends more time on Youtube and social networking websites to pile up cache writes onto the drive, not compile code then execute the program to test it.

These systems are definitely not being used more than "advanced grandparents+1" right now and I would really not consider them a gauge of what could be the range of average users in terms of writes.

[zhenya]

My T43 is 5.x years old, turning 6 this summer.

Which only further proves my point. For all you can tell from these numbers, this drive is being used at exactly the same pace as your own. You can't possibly assume anything otherwise from such a limited dataset. Now you are just being offensive to me, and making yourself look foolish.

I'm done.

[Navck]

I have easily written more than 1TB of data to my desktop drives within a day, I hope you realize that 3TB over two years isn't very "heavy handed" use for a SSD at all... That is well within their operational range and then some.

I'm just saying that YOUR personal experience with SSDs has been with very "gentle" use with them and that they will not hold true with the range of users that exists.

I'm also suspicious of what you say about system restore because a lot of the SSD topics on most forums tell people to TURN OFF hibernate and system restore for a reason. (Space+Wear)

[ThinkRob]

I'm also suspicious of what you say about system restore because a lot of the SSD topics on most forums tell people to TURN OFF hibernate and system restore for a reason. (Space+Wear)

Actually, that recommendation came about back when the early JMicron controllers were popular. Since these drives had absolutely awful small random read/write performance, killing *any* source of IO was necessary to achieve acceptable performance. To that end, turning off system restore was indeed a good idea -- but if you're really desperate to reduce writes at all costs.

As for why it's still a going recommendation? Well... as we've seen on these forums from time to time, bad "facts" die hard.

[Navck]

Every time system restore writes to the disk, that is a few (~1 or more depending on what space you give it and when it decides to write...) GBs off your potential writes for whatever you, yourself, wanted to put on that SSD.

There is still VALID reasoning to turn system restore off.

[ThinkRob]

Every time system restore writes to the disk, that is a few (~1 or more depending on what space you give it and when it decides to write...) GBs off your potential writes for whatever you, yourself, wanted to put on that SSD.

There is still VALID reasoning to turn system restore off.

Depends on how paranoid you are of writes. Even if it were to cause 1GB/day of writes (which is a *very* high estimate), that wouldn't be a problem for any decent SSD.

As I said before: it's a good idea if you're desperate to save writes at all costs. If you're a normal user with a modern SSD, there's really no reason that I can think of to turn of system restore.

SSD lifespan's been debated to death here and elsewhere. I can certainly understand your caution with regards to SSD reliability -- I too am quite cautious when it comes to hardware -- but I'm afraid your assessment of modern SSD technology is somewhat divergent from both mine and most of the industry's. If SSDs were really as fragile and as prone to failure due to writes as you imply (and seem to believe), then it is somewhat remarkable that such a widespread issue has not been mentioned in the popular tech. media. Further, were SSD reliability to be such a problem, I would not imagine that system OEMs such as Dell, HP, Lenovo, and IBM would offer them as options in their servers -- especially considering that they make systems designed for workloads that involve far, far more disk activity than even the average developer's usage (e.g. database servers, file servers, etc.)

[dr_st]

If you're a normal user with a modern SSD, there's really no reason that I can think of to turn of system restore.

Well, I can think of one. One that's valid no matter what user you are and what storage hardware you use. And it's the fact that system restore has consistently proven itself to be completely useless in actually providing usable helpful restore points, and thus is just a waste of space. At least that's my experience.

[Navck]

Thinkrob, if SSDs were so durable then why does the SSD industry *stress* internally about good wear leveling algo development?
Could it be that people outside the casual mainstream audience will make a MLC drive die within months? Why, yes it is!

SSDs are ONLY apparently durable as marketing and good rep says it is. Ironically all the flash media forensics, data recovery AND storage engineers say otherwise about it, to the point that storage engineers joke about racetrack memory. For a parallel: "Fusion power is ALWAYS the power of the future."

And those server options? SLC with tightly regulated media, not the MLC that everyone loves for the cheap price.

[uux]

Could it be that people outside the casual mainstream audience will make a MLC drive die within months? Why, yes it is!

Sure, but if you're outside the casual mainstream audience, then it was a bad choice to use a casual mainstream SSD for your purpose. The limited write endurance is a fact. There is no argument about it. It exists. Obviously, it's not a problem for everyone.

Ironically all the flash media forensics, data recovery AND storage engineers say otherwise about it

They have their jokes about traditional platter based drives as well. Anything mechanical will eventually fail. No matter what you use, make backups and then don't worry about it.

And those server options? SLC with tightly regulated media, not the MLC that everyone loves for the cheap price.

That's kinda funny because a few years ago, enterprises wouldn't even use SLC based SSDs due to the write endurance and general fear of early failure. When SLC based technology got cheap and started to perform similar in comparison to other SSD based storage, that fear kind of just went away.

[Navck]

Everyone in the casual mainstream audience thinks that everything can be replaced by a casual mainstream audience SSD for all uses universally, are they right too then? What with all the popular opinion enhanced by massive marketing departments. Actually, do you define a pro photographer who doesn't know how to use their computer besides to load their NEFs/RAWs on and process them in Lightroom as a user of a lower endurance SSD?

... No, I don't think so. I've have had insightful conversations about how there are failure models for harddrives that are abundant while the world of NAND is like "Data retention? Lets stick our NAND media into the oven, pull it out after 2 hours and see whats left. Rinse and repeat until we get some results." Oh right, then when you go into the small lith processes (Low 2#nm) you start encountering fun things like quantum tunneling. Harddrives are predictable at the INDUSTRY level, they know what designs failed and what designs didn't.

Harddrives failing? Proper cooling, clean powersupply and a good design will last over 5 years, given you can say they are "obsolete" but I find that 500GB drives can be still useful in 2010 even while 2TBs are abundant and everywhere. SSDs failing is a GIVEN due to NAND being used as the medium.

SLC technology, cheap? 1.2k for a Siliconedge N1x. Right, "cheap." I could invest in a RAM disk at this point and get UNLIMITED write endurance. Not to mention that the reason their fears are suppressed is because of the maturity that they have gotten SLC NAND to so they can achieve all the high write endurance values per cell. The fear is still there, why else would you have drive wear indicators?

NAND is not perfect, it will not be perfect due to the inherent flaws of the technology. Until you find a way to make the drive "refurbish" a cell then you will still have write endurance issues AND even more data retention issues as you head down into the smaller lith processes WITH lower write endurance.

Oh hey, "write amplification." SSD controllers are not perfect either. Ever wonder why the Sandforce controllers compress data (Besides to deal with lower quality MLC?).

I realize that harddrives are not perfect either, they're however extremely mature and well know by the industry that makes them. SSDs is more of a bleeding edge technology that will come back to hit you later when you encounter all the INHERENT problems of NAND that are covered up by all the money poured into R&D to figure out ways to make it usable.

[ThinkRob]

Well, I can think of one. One that's valid no matter what user you are and what storage hardware you use. And it's the fact that system restore has consistently proven itself to be completely useless in actually providing usable helpful restore points, and thus is just a waste of space. At least that's my experience.

Smartarse.

I happen to agree with you though. If you're experienced enough as a user to know when system restore can fail, you're also experienced enough not to need it in the times when it doesn't.

Moving on:

Everyone in the casual mainstream audience thinks that everything can be replaced by a casual mainstream audience SSD for all uses universally, are they right too then? What with all the popular opinion enhanced by massive marketing departments. Actually, do you define a pro photographer who doesn't know how to use their computer besides to load their NEFs/RAWs on and process them in Lightroom as a user of a lower endurance SSD?

I had trouble parsing that, but basically my response is quite simple: if you regularly write massive amounts (let's say > 20 GB) of data each and every single day, and if you intend for your SSD to last for several years, then you are not a good candidate for a mainstream, MLC-based drive.

Harddrives are predictable at the INDUSTRY level, they know what designs failed and what designs didn't.

Both the solid-state and magnetic storage industries are constantly modifying their designs, developing new technologies, and new approaches to old technologies. And guess what? Both of them have their share of technical and market failures.

Harddrives failing? Proper cooling, clean powersupply and a good design will last over 5 years, given you can say they are "obsolete" but I find that 500GB drives can be still useful in 2010 even while 2TBs are abundant and everywhere. SSDs failing is a GIVEN due to NAND being used as the medium.

I'm slightly confused: you tossed out a fairly specific number, but made no mention of where you drew it from.

As far as SSD failure being a given... well... you're right. They're not suitable for long term (think 10+ years) archiving. Fortunately, SSDs are neither designed nor marketed for such a purpose.

The fear is still there, why else would you have drive wear indicators?

What indicators are you referring to? Surely you're not referring to SMART attributes -- are you?

SSDs is more of a bleeding edge technology that will come back to hit you later when you encounter all the INHERENT problems of NAND that are covered up by all the money poured into R&D to figure out ways to make it usable.

Solutions like wear-leveling and ECC sure do seem like "covering up" flaws -- until you realize that conventional hard drives do some of the exact same things. Ever wonder about that "Reallocated_Sector_Ct" SMART attribute? Ever wonder why WD has invested a fair amount of research into error correction algorithms and into various ways of improving the amount of ECC data they can store?

And those server options? SLC with tightly regulated media, not the MLC that everyone loves for the cheap price.

I'm afraid I must be rather blunt on this: you're wrong.

IBM's Power Systems' SSD offerings are MLC-based. HP and Dell both sell PCIe SSDs that use MLC flash, and the former explicitly markets said drives to server customers.

[Navck]

Quick and incomplete response since I'm not at home:
5 years comes off my personal drives compared to observations of other people's systems using similar harddrives and often different systems.
I realized that the guys who would have their 120GB Maxtors die years before mine (Which are now mostly retired except for the Seagate rebranded 500GB...) often had undercooled systems that would let the drives go past 45C+. Equally I've noticed people who had Seagate drives dying prematurely would have very undercooled systems with el-cheapo questionable powersupplies from special sales. I have yet to personally have a harddrive fail on me except for two Deathstars (One is a 75GXP)

The wear indication SMART value is what the controller believes is the REMAINING lifespan in the drive. Not all early SSDs had this and some users would experience a drive death without warning... Hope you have the most recent backups that aren't a week old.

"Data retention? That’s a lie, you're obviously paid by the harddrive industry to lie to us about how those spindle motors will not spin up after ten years randomly."
People love saying lines similar to that and act like SSDs are these infinitely perfect omnipotent devices. I've seen friends and family with data retention problems on SD cards they loaded up with their favorite movies so they can watch it on a plane. They also seem to be attracted to any valid criticism about SSDs..

SSDs utilize ECC as well. My personal drives do not have any reallocated sectors and I expect them not to gain any for a very long time. NAND however, WILL fail after enough rewrite cycles. People refuse to get the concept that NAND is not the superperfect media that they think it is.
Harddrives do not "need" wear levelling. The defects on the platter are usually found when they're writing the servo to the platter. Reallocated sectors would probably mean that somehow that slipped through QC or that some extreme amount of incompetence was involved (I'm going to attempt to ignore all the bad things that happened with Seagate and Maxtor) OR that someone has been beating ont heir computer case every time the computer is not “responsive” enough to them. I don’t care if you have a SSD in that computer but pounding the computer will do all sorts of bad things (Oh look, heavy mass known as heatsink, would you like to make the planar suffer as you have more inertia?)

I'll ask my friend with the Blade servers on the subject of IBM's Power Systems.

[uux]

Everyone in the casual mainstream audience thinks that everything can be replaced by a casual mainstream audience SSD for all uses universally, are they right too then?

Right, you've spoken to everyone in the casual mainstream audience and received their opinions. Sure, guy. At home, I'm in the casual mainstream audience and I do not have the above opinion. I don't know why you're even trying to make this an argument. You obviously know a flash based SSD isn't suited well to every application. I don't see anyone here trying to tell you that they are. There is no argument here. The limited write endurance of a flash based SSD is factual, not mythical. There are disadvantages to a flash based SSD. Though, sometimes, those disadvantages do not outweigh the advantages.

Lets stick our NAND media into the oven, pull it out after 2 hours and see whats left. Rinse and repeat until we get some results.

I'll try to keep in mind not to use my laptop in the oven. Thanks. Who knew? Yes, a well designed platter based drive can last for years. They can also fail early due to quality issues or electronic failures, it's not just SSDs. They can fail from a piece of dust getting in one of the air holes, or a change in air pressure (use your laptop on a plane?). They can even fail in such a way that the head can scrape and cause abrasions on the surface of the platter. If you're lucky, it'll only be a mechanical failure, which many data recovery specialists are adapt at repairing. Sure, SSDs can fail in spectacular ways not related to the write endurance limitation. So what? You're either lucky or haven't been using computers long to not experience a drive failure of any type. There isn't a single reputable person in data recovery or storage engineering that will tell you drives won't fail. They all say it's not "if" they fail, it's "when". They all fail. Pick what is best for your application and move on. Learn to perform backups religiously.

SLC technology, cheap? 1.2k for a Siliconedge N1x. Right, "cheap."

Yes, cheap. Look at the alternatives in the enterprise market. SLC is the low end of the spectrum.

The fear is still there, why else would you have drive wear indicators?

Yes, and that is precisely why I find it funny that it is acceptable now.

I realize that harddrives are not perfect either, they're however extremely mature and well know by the industry that makes them. SSDs is more of a bleeding edge technology that will come back to hit you later when you encounter all the INHERENT problems of NAND that are covered up by all the money poured into R&D to figure out ways to make it usable.

They are mature, and yes, there are actually advantages to a traditional hard drive in many applications. Despite the downsides to NAND, there are advantages to flash based storage devices as well. SSDs are not bleeding edge, though. The consumer based NAND drives are, but otherwise, SSDs have been around for a long time.

I looked at the SMART attributes on my first gen X25-M. I used this in my laptop for almost two years, everyday, for at least 8 hours a day. It did an average of about 1GB per day. Yes, you probably consider it light usage. And? It works for me. At that rate of usage, this drive will last a LONG time. Far longer than any platter based drive has lasted me. The speed benefits and time I've saved due to the benefits the SSD provided is priceless to me. We're talking saving hours a day. At the same time, the laptop still remained responsive and quick. I can't overlook that benefit.

[uux]

The defects on the platter are usually found when they're writing the servo to the platter. Reallocated sectors would probably mean that somehow that slipped through QC or that some extreme amount of incompetence was involved (I'm going to attempt to ignore all the bad things that happened with Seagate and Maxtor)

Hard drives fail. All of them. SSDs, too. Some in worse ways than others. Have you read the paper published by Google on hard drive failure rates? Do you think people at Google were pounding on the computers in their data centers? Do you think Google doesn't keep their data centers at optimal temperatures, clean, and with well conditioned power? I wouldn't put anymore trust in the reliability of a traditional HDD than I do a consumer based SSD.

[Navck]

Google's study deviates from what the harddrive industry agrees on: Lower temperatures means lower rate of failure.

I'll tell you why it works: Imagine your FDB that doesn't leak heating up to the point it starts leaking fluids... Oh look, you've just put yourself on the fast track to motor failure. Oh right, electronic component lifetime is rated at 25C and drops as you rise in temperature...

I'm sorry that you don't like how the industry likes to test data retention, but you realize that your computer does generate heat, right? Arrandale processors are rated UP TO 105C, which would be that "oven" you mock. Ever consider why your Thinkpad spins the fan even when the system is cool? Circulating stagnant hot air pockets in the system could be a vital function here.

What, you mean a full RAM disk setup with a backup battery? SLC drives are not cheap.

Speed and benefits that people talk about seem to be highly overstated from my personal observations ("Look at my laptop trash your boring black box of a laptop!" type kids). I've practically performed "at the same level" if not faster in silly games such as "lets watch who can load up their Internet Explorer/Firefox/Opera instance up then some other program."

Whatever, for some reason when I make any valid criticism that comes from STORAGE ENGINEERS, every fan on the bandwagon feels the need to elevate SSDs onto the pillar of "best storage option."

PS: Dust in the breather hole? HA! Some drives have multiple breather holes AND they have filters in them, good luck plugging it up somehow. I won't address the airplane issue because of modern fly height adjustment with the head and how it can actually INTENTIONALLY contact the platter for short durations. Didn't know that? Guess what, modern drive platters have a thin layer of lubricant as part of the platter, the head CAN touch the disk without damage ("Hydroplaning") during reads and writes. Go find out some information from actual engineers in the storage industry about harddrives. Also, any damage that PHYSICALLY damages the magnetic media? Anything there is GONE, you cannot "repair" that damage.

To Thinkrob:
My Blade server friend tells me that the "MLC" drives that are used at an enterprise level are "not complete MLC" and "they are based in tiers." As well as how 32GB of reserve and mixed NAND with engineering can go a very long way.

[Colonel O'Neill]

I really don't quite see how you could "save hours a day" by using an SSD. Either you're clearly overexaggerating the benefits, or somehow, this SSD makes your 24 hour day a 30 hour day.

As for speed improvements, a few seconds here or there doesn't really add up to many hours does it? That reduces itself if you take advantage of the computer's ability to multitask; you can preemptively load an application a few seconds in advance: If I want to photoshop something, I go start Photoshop first. By the time I go get the image I'm looking to photoshop, the program will be loaded already.

I also don't quite see how you only manage 1GB of writes a day; I did a rough sampling of file writes using Process Monitor, did some calculations, and ultimately ended up doing approximately 80MB of file writes in 20 minutes just doing some moderately intensive web surfing and MSN chatting. If I extrapolate this to using my computer for about 5 hours a day (on average), I get 1.2GB a day. Those conditions are really light to me. I just did a 7x86 install in VMWare to check out that weird new KB update that was pushed out this morning. There goes another 8GB in an hour.