Did you catch the news yesterday about those chronic
performance problems that cropped up in a trio of Intel
X25-M SSDs under review by
PC Perspective? Intel apparently did, promptly responding and identifying what it believes to be the issue: the testers. The review alleged that the drives' write algorithms, intended to evenly spread wear and extend their lives, result in extreme fragmentation and major decreases in performance. Intel is saying it wasn't able to replicate the results, that "the synthetic workloads they use to stress the drive are not reflective of real world use," and that "the benchmarks they used to evaluate performance do not represent what a PC user experiences." The reviewers were largely just copying files around and installing/uninstalling applications, which sounds fairly realistic to
us. We're thinking these two aren't going to agree to disagree on this one, and that there will be more updates to come.
posted February 20th 2009 9:21AM
good morning, engadget :)
top of the mornin to yah sir =P
for once someone replies! +1 sir, and have a great day
You know, I would almost side with intel here, my experience with flash is that it works 100%, and if it fails it completely goes down. I can see fragmentation slowing the performance, but not the chip overheating or failing.
like a processor... if overheating or failing, it doesn't slow down, it completely gives up and crashes.
Fragmentation slowing down a flash memory chip? Yes the electrons take a LOT longer to move from different areas of the chip just like the armature of a hard disk drive.
Dont be an idiot Kurian while the moving parts have been replaced with lightning fast controllers, fragmentation makes that 1ms seek time turn into something much bigger when you start scattering stuff on the disk, its all based on how a file allocation table works. while SSD's often use multiple chips at once similar to dual channel memory, the fragments are still looked up on a single chip where the FAT is stored. fragmentation of a file into maybe 150 fragments is ok, but when you get into 10,000 fragments, say goodbye to performance.
It doesn't work this way at all. Please do your homework before publishing a blanket statement like "flash will either work or not work", there is a lot of gray area in between.
Flash has a maximum number of write cycles before it goes bad for that reason the drive has advanced algorithms that purposely spread data all over the flash arrays. As areas of the drive get written to more and more the algorithms changes where new data is written too in order to ware less used areas. This is all done behind the scenes in the drive and is not seen to the device accessing the data (aka the sata controller in the case). This is also done one SD, compact flash, and other flash devices.
If these algorithms are not created very carefully you can get into a situation where the drive is trying to write data into a smaller and smaller number of less used flash blocks across the drive hence creating a massive fragmentation and slowing performance.
Kurian, It has absolutely nothing to do with time it takes electrons to move in the drive, haha. That makes absolutely no scene at all.
"like a processor... if overheating or failing, it doesn't slow down, it completely gives up and crashes."
This is not true. Processors that overheat will generally reduce their speed in order to generate less heat and move back within their ideal operating range.
I personalty think that this is still new technology and think will tell if Intel or PC Perspective is right De fragmenting a SSD Is NOT recommended by any one I know of so fragmented Data on the Drive will slow performance over time.
"The reviewers were largely just copying files around and installing/uninstalling applications, which sounds fairly realistic to us" That's what I do most of the day move large files all over the place and install and uninstall stuff, as I am reviewing or testing software. Thought I don't yet have an SSD, this makes me a little leary of getting one now... So I just bought a much slower sata 1.5 Tera Byte for 105 from Dell with a coupon code from dealnews. Vista gives it a performance rating of 5.9 I can live with it... Just curious what performance rating does Vista give the SSD's? I tried to run a seek test on ubuntu, but my drive was giving me unrealistically high numbers.
The X-25M on my ThinkPad X200 reports 5.9 subscore on Windows Experience Index (Vista Business 64-bit).
Exactly. Who are they to tell you what "normal" usage is?
Hey Daniel, my SLC Samsung gets 6.1 on Windows 7 (which gives lower ratings than Vista for disk subsystem). I paid less for it too, but it is a smaller drive. No deterioration either with these SLC drives. I'm glad I didn't fall for the hype of this dressed up MLC.
Ah....the "It's not happening because we couldn't replicate the problem" response. Seems like the tech and IT industry like to fall back on that response too often. Perhaps the disturbing thing is that Intel wasn't *capable* of duplicating the problem. One would think that a company with their resources should be able to duplicate *anything* a group of magazine weenies can do...
It makes them look even more incompetent for sure. It means they did not even test the product as much as these reviewers before trying to sell it at $600+
Funny. They argument with synthetic workload, real world use, blablabla, what do we learn? They know that such a failure is existent they just don't know a solution for it because there won't be a solution for this failure on these drives. Funny that other SSD's don't seem to have this failure (ok, they have other failures). So something has to be different with the Intel drives, and Intel knows exactly why, but they can't solve it without releasing a different model. Else they could tell us why the synthetic benchmarks get worse over the time.
"the benchmarks they used to evaluate performance do not represent what a PC user experiences."
And which benchmarks did Intel use when they were testing the speed of these drives for the marketing department, eh?
It shouldn't matter whether or not the Intel techs can reproduce the results, if the consumers have problems then it's Intel's problem too. It'll be interesting to see if anyone else can reproduce the drive issues.
Im having a bad day..so you know what...F*** Intel and they squirmy PR
I can't help myself pointing out one major flaw about this photo is that the surface of the SSD reflects the camera it is shooting at. Observed the Canon reflection on the SSD.
Anyone looking for a decently priced SSD, The new OCZ Vertex drives (not the old ones) with the Indilinux controller is nearly as fast as the Intel in read speed and blows it away in write performance.. The OCZ Summit drives are similar, and even the dual-JMicron based Apex is decent..
shit, stupid comment system!
@Frank, Sorry to burst your bubble or what not..in your effort to be clever, you failed to read any background on this issue. The Intel X25M is one of the most expensive SSD drives out there for a few reasons. If you go and do some research you'll see that Intel is one of the few companies that make SLC Flash SSD drives as opposed to using the cheaper MLC Flash SSD drives. This does raise the cost somewhat but provides substantially better performance. In addition to this, the "difference" you refer to is clearly stated in some of the background articles. Intel created a special set of algorithms to handle "wear-leveling" on these drives in order to evenly use the memory on the drive. This needs to be done since Flash memory has a limited number of times it can be written to. Wear-leveling has the drive write more evenly across the memory on the drive so that one part of the drive doesnt get "consumed" faster than another and wear out.
This algorithm has been clearly stated to be the cause of this issue as it introduces the possibility to fragment data on the drive over time since data is moved around and written over different areas of the drive. Intel is probably stalling while they figure out how to retain the initial performance boost (which is quite large actually) while mitigating the longer term degradation. The main problem is that you can't defrag the drives without wiping them clean (because a disk defragmenter currently operates at a level above this algorithm which ends up making the problem worse).
I find the whole issue funny though. On standard platter HD's you write to the first available sector in order to provide speed. This leads to fragmentation over time. On these SSD's it's better to write in continous blocks but you wear down the life-time of the product if you don't use wear-leveling. It's just a bit surprising that Intel thought that they'd be able to avoid fragmentation slowdown by scattering data.
But the SSD this article is talking about, the X25-M, is MLC. Only the "E" series is SLC.
Agh, My bad, The X-25M still has the same wear-leveling algorithm or at least a very similar one.
Ok, first I have to burst your bubbles :) The Intel X25-M is a consumer oriented MLC SSD and the price is, you're right, high for the SSD. It's not as cheap as the not working OCZ Core SSD's but also not as expensive as better SLC SSD's, still expensive enough for a not MLC SSD. But again, it's a MLC SSD. (The Intel X25-E is SLC, but not the here mentioned X25-M)
Second, every SSD has wear leveling. Intel just uses a different, probably not really working, algorithm.
Third, for the price the Intel SSD costs I expect a fully working SSD, which also works for a few years without problems. And I think Intel can't expect that their users wipe the drive clean every few months. So I think everyone can bash Intel if they really produced a product which was nice on the paper, but now, after using it, shows that it's only slightly better than the much less expensive MLC competition. If I had bought such an Intel SSD and experienced such a slowdown, I would want my money back now.
I just think that the whole SSD's gets too much hyped. The expensive SLC SSD's from MTron, ... may work, but not so the less expensive, sadly. And luckily I'm not the idiot who pays for their beta products.
Well, You reiterated a point made by Patriks...
Not all wear-leveling was created equal. Intel's algorithm does indeed show this flaw, however it does show substantially better performance out of the gate...not just on paper.. in actual testing.
And yes while it does suck that this does take a performance hit, it isn't as bad as others have mentioned and yes it does slow down over time...but what platter based HD doesn't?
Personally, I have not jumped onto the SSD bandwagon yet...while I'm pretty sure they will become dominant soon...I'm not 100% convinced just yet...
Remember ZIP Disks?
The guy from PC Perspective is personal friends with people at OCZ, who also happens to be the only site to do a couple of previews of two of OCZ's new drives (no one else has looked at them yet)...
I would not really take to much of what he has to say on this as he is just "pimping" OCZ products and trying to make Intel look bad. I have had both manufactures latest offerings and the Intel is hand down a better product.
BTW: I don't work for or have any interest in either company - period...
Actually the "author" doesn't know ANYONE as OCZ, though as the site owner, I do. I also have friends at Intel, the very ones replying to CNET above.
Thanks for your pointless comment though.
Intel: oh, hey guys, we have this firmware update... it's just for fun you know, no real updates or anything... ;)
With an SSD fragmentation is irrelevant. Fragmentation slows a drive due to adding a lot os seeks to file retrevials, with an SSD there is no head to seek and seek times are 0. The worst part about fragmentation would probably be increased slack space but with the huge sector sizes these days there is a lot of slack space anyways.
SSD fragmentation is not irrelevant...it's just not AS relevant as platter based drives. Fragmentation does slow the drive down as it is required to perform more seeks. Seek times on flash drives are NOT 0...they are near 0...big distinction there and these can add up over time.
it'd be interesting to see how the long term test would have gone with the installation of MS's SteadyState
Ryan,
Then I stand corrected on that small fact. But I still have REAL questions as to how your site got hold of 2 unreleased drives BEFORE anyone else to pre/re view unless you had some interest. Matter of fact, many of the personalities (D111, Tony, Wendy, and of course RyderOCZ) on the OCZ forum are in some cases actually moderators / owners of other sites / forums as well. Makes me really wonder if there is not some $$$ interest from OCZ (kind of like viral marketing) going on here... I don't blame OCZ for wanting market share, I blame them for not telling the whole story, plus I have questions as to the quality of the products.
Heck, do they even (outside of the PSU's) actually even make anything?
Give me a break if the Intel controller is so flawed, I guarantee that whatever OCZ is using is even worse. They are using a new generation Jmicron and we all knew that it would not be up to the level of the Intel controller.
The real solution for now is sticking with SLC.
Pure BS. The testers are working with Intel to learn of and hopefully rectify these problems with a new firmware. I'm so glad I didn't fall for that hype and went SLC instead.
Whatever, what is a "real PC usage"? This is BS. So if buys this drive and ends up installing and uninstalling a lot of programs over the course of time they should have a drive that is permanently crippled (okay unless you do a secure erase every now and then)? This is just a poor product and we will see it go to $200 where such crap should be priced.
The "Founding Fathers" of the USA were very concerned about the possibility of one individual consolidating power into his / her own hands. They set up this whole nation in a delicately balanced system designed to avoid that sort of power-concentration. It's a shame, though, that they never thought about how businesses would do the exact same thing.
I think there is some merit to what Intel is saying. These guys were testing speed by imaging the OS onto the drive. I don't know about you, but most of us never image an OS onto our drive.
The key to this SSD Intel says is that it is adaptive. That is, if you use files often, they get stored in ways that make them quicker to access. However, when you image an OS onto a drive, each file has been touched (written) exactly once. How is the system to know which are the most commonly used files? If you use it a while and in a more "organic" fashion it is possible more of the real-world performance will be preserved.
Most of the comments ignore the fact that fragmentation takes places at many layers, and SSDs still suffer file fragmentation in the same way as HDs, and this can have a performance impact as the time taken for a file access to acquire all of the fragments grows with the number of fragments, and read access to an SSD is not zero (particularly as there is a data structure beneath the file level to undertake wear levelling. I have seen and had to fix file fragmentation on EeePCs that were suffering performance hits of almost 100x compared to nominal, from file fragmentation of the order of 10,000 fragments for 30MB files. The only fix was to defragment with the slightly unsuitable tools available today which are not optimised to undertake a pure file defragment. I do not recommend continuous defragmentation but a periodic triggered defrag is required until we have specialised tools to do this properely. Note data fragmentation on flash devices has been known for a very long time - embedded devices have had to deal with this for years as it has had deterimental effects on flash used for STB storage for example.
or, is INTEL trying to tell us something with the subtle curve?
tomo