Intel X25-M SSDs slowing down with extensive use?
By Donald Melanson 
posted February 19th 2009 1:09PM
posted February 19th 2009 1:09PM
Intel's X25-M SSD drives have certainly shown themselves more than capable enough in the usual tests and various configurations, but a long-term test report from the folks at PC Perspective is now casting a bit of doubt onto their long-term effectiveness and speediness. More specifically, they found that the sector remapping and wear-leveling algorithms that are designed to increase performance and extend the drives' lifespans were actually doing quite the opposite as the drive became increasingly fragmented after extensive use. Further complicating the matter, it seems that conventional defrag programs only make the situation worse, with the only sure way of resetting the drive's sector maps being a complete wipe of the drive. PC Perspective is quick to point out, however, that many users may well never notice the slowdown with light use, and they do have a few suggestions to minimize the problem until Intel devises a firmware fix, which is the only surefire solution.
[Via Ars Technica, thanks Robert]
[Via Ars Technica, thanks Robert]
Why would fragmentation even matter? It's not as though there's a read head zipping from sector to sector.
Because this techonology is not ready for main stream yet.
Tavis that doesn't make a bit of sense.
It makes perfect sense. SSD's are not really ready for everyone to start using them. There is no real world info about reliability or performance.
As was stated below it may take a generation or 2 for the kinks to be worked out.
Because of the UNKNOWN about lifespan and long term performance use.. (And price) I will stick to standard HD's.
I'm sure it wont take that long. Seems like all that is needed to go mainstream is lower prices, higher capacities, and firmware updates.
I have to agree with Tavis. As a consumer I don't feel comfortable with SSD tech if it has the potential to degrade relatively quickly. It's a great step forward for small portable devices, but I'll stick with PATA/SATA for a few more generations of SSD.
Read the read link, its explained very well.
The shocking thing is that the write speed can drop down to as low as 20MB/s from 250MB/s.
That is a LOT slower than a old 500gig spinner I have.
And also, you cant fix the issues from within the OS, you have to move your whole OS off, then write one large file over the SSD, and even then, sometimes that wont fix the drive. meaning you have to do a deep format to reset it to factory condition.
Also it looks like its just the intel SSDs that are effected atm, the same thing that causes them to have such high speeds is the same thing that causes this problem.
@TravisJohn
How do they get real world stress test and results if nobody in the "real world" is using them? Early adopters are just as much part of the process of refinement as the corporate R&D.
There is no logical way fragmentation can slow down a ssd, an the linked website does not explain how this is even remotely possible. Defragmentation makes a harddrive slow cause the read/write head has to jump all over the disk, however in a ssd there are no read/write heads so fragmentation doesn't matter.
I don't doubt that a ssd can get slower over the time cause of wear leveling or something else but definitely not due to fragmentation and I doubt that the performance can drop that much under normal working conditions.
The better question is, "why does it matter? I have linux." Frag doesn't matter so much...and I usually reformat once every 6 months anyway for the Ubuntu release.
@Xstream You didn't read the article well enough. The algorithms in the SSD create smaller blocks when smaller files written to the drive a lot, which is what makes them so fast (initially). The drive doesn't have to re-write as much data every time that file is written. The problem comes from the wear leveling algorithm that moves those small blocks around. When the wear leveling moves that small block somewhere else, like to a block that's bigger, the algorithm chops the big block into smaller blocks for the small files. That is good when the files being written to those blocks are small, but when a larger file is written to that block is has to be split into smaller fragments than it normally would. Blocks are like CD-RWs in that data can only be added to them; you have to erase the entire disc to erase something on it. Except that you can't just erase a block, the data in it has to be rewritten somewhere else. So when you write a large file to several small blocks, those blocks are moved somewhere else. But, once again, if the small blocks are written to a larger block, that block is split into smaller blocks, further compounding the issue. The write speeds drop so much when it gets fragmented like this because the drive has to do many, many, more operations per write because there are more blocks that have to be written to.
IT has nothing to do with "the technology not being ready".. although it is quickly evolving, SSDs have been used for over a decade in aerospace and military applications...
Secondly, Check out the reviews of OCZ drives like the Vertex.. They seem to hold up well.
Finally, drives that use SLC flash won't have this problem as they write much faster...
[gasp]
Interesting.. I was actually planning on getting some SSD's for my laptop, but perhaps I'll wait awhile till the tech improves.
First you say they're great. Now they're not good for us. Tomorrow you'll say they cause cancer.
Then they'll say SSDs will take over the world and enslave humanity ;)
Ahhh .... the 1.0 blues
For some odd reason, I suddenly want to go out and buy a Canon DSLR.
and for no particular reason, i just felt like running.
that's all i have to say about that.
@ roundgob: I was thinking the same exact thing.
Canon ad revenue earned by roundgob today = $1
Well to new technology, there isn't a lot of information on the devices behavior in real world situations over extended periods of time. Give it a generation or two and I'm sure all of these little hiccups will work themselves out. Its not like they haven't been working on magnetic media improvements for decades or anything.
Sounds about right that this info comes out now... I just received my X25-M yesterday.
@stoatmuseum
I think they are saying that the issue is the algorithms that speed up the drives are impacted by the fragmentation issue, not the drives capability of reading the data on a fragmented hard drive. The algorithm is mucking up the seeks to fragmented data. That said, the sector remapping and wear level algorithms would seem to do the same thing, that is, making sure you don't write too much crap to one area of the drive while not using another area... making defragging first probably pointless as the goal is to prevent those areas from being used in the first place.
Perhaps this is where they could take a lesson from non solid state drives and implement buffering data or only using complete blocks in isolating data... (disk waste, but performance improvements)
Calling what is happening "fragmentation" is bound to cause confusion. It's unlikely this is truly related to blocks in files not being adjacent in adjacent blocks on disk. The only thing that matters is whether as many blocks in a file are stored in a given page as is possible.
I'm a little skeptical that this is a big problem. Yes, it does mean that you won't always experience maximum rated speed with an SSD (which many people thought was guaranteed), but I think that part of the problem is that right after imaging your drive, the system doesn't know which blocks are the "hot" ones. As you actually use the drive, the blocks that are used the most will be treated best and end up grouped properly in pages and run at full speed.
In short, I think what I'm saying is the answer to Engadget's title: 'Intel X25-M SSDs slowing down with extensive use?' is incorrect. They slow down immediately when you start to fill them but will quickly cease to slow down and under extensive real-world use (not necessarily benchmarking), will recover somewhat to nearer to full speed.
I see nothing in the article to indicate that they will somehow recover if you keep going.
The article repeatedly says it is adaptive. If you use certain files a lot, those can be optimized for by the SSD in various ways.
I had a different problem with one of these drives which would have required a firmware update. I was told by an Intel representative that the firmware on the drive was not upgradable, so no fix would ever be available. Either I was lied to then, or this problem could not be resolved with a firmware update either.
Wow that would be a humongous mistake, to release it without a possibility for firmware upgrades, seems quite impossible that intel would be so daft, but we've heard weirder things of course that turned out to be true, but still.
One thing seems sure, intel needs to implement a 'defrag' mode in their firmware and release an utility for doing a SSD-specific defrag (it's obvious you should not and cannot use a HD defragger on a SSD, all calls will be intercepted and all you do is wear out the cells).
Totally agree. I feel maybe I jumped in too soon. There really isn't the kind of software and OS support for these drives yet. Still, my computer is blazing fast and in a laptop these things are gold.
Here is the thing with PC Perspective..
The owner is personal friends with some people who work at OCZ who just so happens to be pushing some new SSD drives out AND who happens to be the only one on the net to do a preview... Seems pretty odd to me.
I have personally worked with both the Apex and Intels, the Intels are hands down better.
Windows 7 disables defragmentation on SSDs
And then ?
Then, it's Microsoft preventing the problem from their end. As mentioned in this post and elsewhere, this is clearly a hardware problem.
I have been using an SSD (128 GB Transcend) in my Macbook Pro for almost 2 months now. I find the drive has been FAST!, reliable and works as advertised. This model does the same wear leveling, but remember folks, these drives are spec'd with a 1,000,000 hour MTBF - 1 million hours.
I like mine. I do, however continue reliable and regular backups.
Hasn't this always been the case though? I swear I've read articles before about SSD wearing out faster if you defrag.
This is different Gundry. The article about SSD wearing out faster with defrag were basically trying to state that defragging the drive doesn't buy you anything from an endurance standpoint and would actually make the drive's endurance worse because you would then be doing nonstop write/erase commands.
This article is showing that your speed will decrease with time as the drive is used because the wear levelling algorithm places so much data in different locations, that when you benchmark the drive later in life, it's nowhere near what it was initially.
Be that as it may, common sense would say as a drive wears out, it will preform worse. This is true of cars, trains, pants, etc. While I'm not saying this article isn't new news (does that work?), but honestly I kinda of assumed this was the case.
And it's Eli, not Gundry. I barely know you.
No need to be rude Gundry. Robert was being helpful... and no, logic does not dictate that performance slows over time for everything... does your RAM get slower after a few months? Doe your CPU lose clock speed with wear?
No pun intended but only an idiot would defragment a Solid State Drive...
It simply doesn't matter if the files are scattered all over that precious flash ram as the access time will always be the specified 1 ms.
Access time is a relative term here. When you look at commands, you have overhead that will make a command that might take 6us on a rotating drive but will take 1ms on an SSD. So when you have tons and tons of these small commands taking a 'little' longer, it eventually adds up.
Has anyone done any kind of testing like this on an OCZ, SuperTalent, Mtron, or any other "smaller company" SSD?
I've heard that if you fill some of their drives with data, you will almost immediately see a substantial slowdown in performance.
Hey - I am experiencing this first hand! I've submitted it to a few different forums and even tried using my company's Experts Exchange.
http://www.experts-exchange.com/Storage/Hard_Drives/Q_24159694.html
If anyone can shed light on this and assist, I'd be grateful.
Thanks!
I knew why I've bought a Seagate Momentus 7200.4 HDD instead of a much more expensive SSD, which also has less capacity and does not work.
Sorry guys, but the SSD is a nice technology but you have to pay for it.
Spend a huge amount of money and get a good but very expensive SLC SSD which may work, but who knows, or spend a lot of money and get a new, beta MLC SSD which does not work, but looks nice on the paper. Or spend only a bit of money and buy a HDD which will work and with a bit of system and OS optimization you won't notice a difference between HDD and SSD (except boot up times, but who boots a system nowadays, sleep mode is the solution)
I don't want to be a beta tester for SSD products. Maybe in one or two years they will release working products, but I can wait.
In order to understand the problem, you have to have a grasp of flash technology. Flash devices are made up of pages. Many pages are grouped together and form a block. You can program a page once in flash technology, but you must erase it before you are allowed to program it again. Unfortunately, you cannot erase a page at a time, you must erase an entire block at a time. So, consider the case where the drive is very fragmented and almost full. You can't just delete the file and overwrite it as in an HDD, rather, you must erase the block. The file, though, is fragmented across many blocks but you can't erase all of these blocks because they have other good data on them from other fragmented files. So, in order to reuse the space from the erased file, you must perform many many copy and block erase operations (think about it for a bit, and you will understand). This problem is only associated with relatively full and fragmented drives (hence the need for an efficient defragmentation routine). So, the more fragmentation done to speed up performance (instead of requiring all of the files data to be put in contiguous pages in a block), the more defragmentation is required. This causes the device to slow down as explained next.
Another problem is wear, as flash devices wear, they take longer and longer to do both program and erase operations (by the nature of the flash device). So, eventually they will slow down, but nowhere near as slow as regular HDDs (notice the article did not mention the amount of slowdown). The problem you want to avoid is one part of the device going bad long before the rest of the device goes bad making the whole device unusable. In order to prevent this, wear-leveling is employed to ensure that the whole device ages at the same rate. The process of wear-leveling results in data being moved around (in the background). This movement, if inneficient, results in the device actually aging faster because of all of the inneficient movement of data.
Also, an HDD defragmentation routine does not understand the block/page issues in a flash device. It just assumes you can overwrite data and group it together. In an SSD, there must be some sort of mapping as a request for data at location X should alway return the same data and a write to location X should always update that data. But from above, you cannot overwrite a page in flash so the data must actually be written to a new location in the flash device and the mapping table will be changed to show that data at location X which used to be at physical location Y is now at physical location Z. So, an HDD defragmentor will see the location X, but the underlying (hidden) structure of the flash is completely different. So an HDD would reorder data, but in the underlying flash, it would probably result in even more fragmentation that the SDD defragmentor would try and fix. Having an HDD defragmento run on a SSD does not make any sense.
And then, there are a lot of other things under the hood which could affect peformance. Since nobody in their right mind would let out their trade secrets, it is unlikely that one would ever get a reasonable response, other than guesses, as to why the SSD is really slowing down. (For example, as the SSD gets more fragmented, are there inneficiencies in the SSD FW or HW that would start to show up, etc.).
Some defragmentation manufacturers PerfectDisk have suggested that one should consolidate free space to ensure optimum performance of the drive is that true?
Also Diskeeper claims to have a routine to defragment SSDs. Perfectdisk claims they are simply consolidating free space.
Should I believe any of these?
What can I do to keep my SSD in shape?
Thanks.
I'm staying away from SSDs like I stay away from "Rev. A" Apple products.
Glad I skipped this and bought a decent SLC SSD instead.