Talk:Solid-state drive/Archive 2

Archive 1

Archive 2

Archive 3

Archive 4

Archive 5

RAM-based SSD's become obsolete

Let's consider. 1. Read speed of both RAM and "RAIDed" flash drives might be as fast as the interface allows (at least two vendors are now ready to supply SATA 3 flash SSD with 500-600 MB/sec read speed. 2. Write speed is a bit more difficult to manage, but Micron already samples flash SSD with 250 MB/sec write speed, May also rise the write speed to 400-500 MB/sec with some price tag added (interleaving needs a bit more complex approach. 3. In principle, it is now possible to produce direct PCIe (even direct to memory channel - see Intel Turbo Memory) attached flash subsystems with multiple levels of interleaving and stripping working on PCIe 3 speed with the price no greater than $20 per GB.

The "wear-out" issue is also becoming less significant - mothern drives may work up to 3-5 years as Vista OS drives.

Compare this price with prices on cheapest RAM-based solutions and you'll see that (might be exept for very rare ultra-write-intensive SQL applications) flash SSDs are now much more price/performance attractive than RAM-based (with the same functionality).Stasdm (talk) 07:04, 13 October 2008 (UTC)

This isn't my field, but

...this story on Engadget today (http://www.engadget.com/2008/07/01/ocz-reveals-core-series-sata-ii-2-5-ssds-128gb-for-479/) looks pretty significant: "these drives check in at $169 (32GB), $259 (64GB) and $479 (128GB), which -- as you undoubtedly recognize -- are amazing price points. Each unit utilizes NAND flash technology, possesses a 1.5-million hour mean time before failure and delivers 120 - 143Mbps read / 80 - 93Mbps write speeds. The sub-0.35ms seek times are also worthy of a tip of the hat, and the low power consumption just makes things unnecessarily sweeter." - Dan Dank55 (talk)(mistakes) 20:29, 1 July 2008 (UTC)

Answer: Unfortunately, all this data is not quite right and needs to be looked at with a lot of care. The mtbf does not have any meaning at all, it could as well last 10 minutes. Also consider that those drives use MLC and not SLC chips. The random write is much worse than the stated 80 Mbps (which means Mbit per second and is wrong in this case. It should be MB/s = Megabyte per second). Alltogether: The prices are dropping, herewith the quality. —Preceding unsigned comment added by RunningGer (talk • contribs) 12:12, 3 August 2008 (UTC)

Subject matter/WP:V, WP:OR, WP:RS experts needed re. ATA/ATAPI support for SSD

Greetings. I am currently having great difficulty with editor Ramu50 re the Advanced Technology Attachment article, specifically over whether solid state drives are "really" supported by ATA.

They are by SATA by not by PATA, if the user want to prove otherwise fineAndrewcrawford (talk) 20:25, 17 August 2008 (UTC)

No, the PATA spec includes specific provisions for SSD. Anyway this is long since settled. Jeh (talk) 18:05, 5 September 2008 (UTC)

I claim they are; the ATA docs say so, explicitly and also by inclusion ("any form of storage device may be placed on the interface"). Ramu50 claims that citing the ATA docs is "original research". Accordingly Ramu50 has repeatedly removed the mention of SSDs from the article lede.

I have opened a case at WP:RSN. The specific section at WP:RSN is here. I've included a large number of diffs there.

This dispute is also discussed, if that is the word, in a long thread (actually several) at the AT Attachment article's talk page: talk:Advanced Technology Attachment.

The input of any SSD or ATA subject matter experts and/or Verifiability, Original Research, or Reliable Sources experts would be appreciated. Thank you. --Jeh (talk) 09:26, 17 July 2008 (UTC)

"faraday cage"

I'm going to let someone else check me on this, but I'm almost 100% sure this statement is completely irrelevant. A faraday cage is basically anything conductive surrounding the drive--there is no reason why SSDs do not have a faraday cage, and I'm almost certain all reasonable manufacturers already use a metal case for their SSDs (they're expensive enough already, why would they skimp with plastic?) Anyway, one problem DRAM-based versions might be susceptible to are soft errors (read on wikipedia). Generally, I'd imagine this is pretty easily mitigated with some error correction (which probably is implemented, anyway. . .) Threepointone31 (talk) 06:06, 7 September 2008 (UTC)

"SLC vs MLC"

Lower priced drives usually use multi-level cell (MLC) flash memory, which is slower and less reliable than single-level cell (SLC) flash memory.[13][14]

This statement is only half true. Read speeds between MLC and SLC FLASH do not vary. Only the write speed is slower on MLC. The statement/source should be more clear on this... Source: my experience - Psignosys —Preceding unsigned comment added by Psignosys (talk • contribs) 17:11, 14 April 2009 (UTC)

12 GB minimum size for minilaptop?

Several vendors are selling linux based minilaptops with only 4 GB of storage.

For example:

http://www.dell.com/content/products/productdetails.aspx/laptop-inspiron-9

Include a note as the need to do away with windows bloatware (t.m.) in order to accomplish this for a useful device.

Hcobb (talk) 17:00, 7 September 2008 (UTC)

I assume you were commenting about the line where it said that SSDs range from 12GB to 256GB. Since there was no citation for it, and both of those numbers are dubious at best, even if they were correct when written they aren't likely to stay so for very long, so I removed the sentence. Zodon (talk) 19:37, 9 September 2008 (UTC)

Input on SSD article from Storage Networking Industry Assoc. Solid State Storage Initiative

Hello Wiki-editors. Last week, while speaking to an SSD industry analyst, I discovered your SSD article. It's actually quite good. Recently, the Storage Networking Industry Assoc. (SNIA -- www.snia.org) started a new group within our organization called the Solid State Storage Initiative (SSSI). We would like to help edit and contribute to your SSD article and hopefully related content / articles. Is there a particular Wikipedia editor or editorial group we should dialog with?

Thanks.

72.20.136.100 (talk) 18:53, 11 September 2008 (UTC)Neal Ekker

If you are looking for a particular editor your best bet is to you look on the hardware taskforce first then computing project. If you are not fussed majority of editors here will be glad to help you improve the article :)--Andrewcrawford (talk) 19:13, 11 September 2008 (UTC)

Andrew, can we copy the current content, edit it in a Word doc, provide citations as per Wikipedia instructions, and e-mail this doc to you? Then you or another editor could post the content with all the Wikipedia conventions, scripts, etc. so that we wouldn't fumble the ball with those? Or is that too much work on your side and we should just dive into the Edit This Page text and root around as we will? I want to work with Wikipedia in whatever way is best for Wikipedia. Would you be kind enough to give me some guidance?

Neal —Preceding unsigned comment added by 72.20.136.100 (talk) 21:57, 11 September 2008 (UTC)

I have a lot of work on so that would be asking a lot of me, i am willing to help you update it but that just a little to much for me. If you edit each section to what is more appropriate then add reference for it using <ref></ref> i will edit it later to comply more with wikipedia guidance. --Andrewcrawford (talk) 08:12, 12 September 2008 (UTC)

Andrew-

Sorry for taking so long to respond, but Hurricane Ike came through and provided some distraction here in Houston (we were closed for a week), then we had our first Solid State Storage Initiative (SSSI) elections and productive meetings out at IBM in San Jose. Fun. Now I'm back in the saddle. I don't want to cause you more work; I want to help SSSI contribute to your already impressive SSD content. I've discovered that others in the Storage Networking Industry Assoc. (SNIA) have contributed to Wikipedia previously. Let me find out from them if we have an experienced Wikipedia writer, and if so, maybe we can supply all our own contributions with minimal help from you. Neal

No problem i cant take on full project my self jsut now have my degree to do weirdly it covering SSD but only minorly. If you post a edit toa section or add a new section send mea message on my talk page ill check it out and make improvement to wikipedia standard wehre necessary and put work to yourself :) but please do not do the entire page at once i just abut have a heart attack--Andrewcrawford (talk) 22:41, 26 September 2008 (UTC)

Write cycles and wear leveling

Please stop removing cited material. If you think the information is incorrect, please provide citations for other views on the matter.

The Infoworld article Lucas Mearian (August 27, 2008). "Solid-state disk lackluster for laptops, PCs". Retrieved 2008-09-12. says:

"Corporate-grade SSD uses single-level cell (SLC) NAND memory and multiple channels to increase data throughput and wear-leveling software to ensure data is distributed evenly in the drive rather than wearing out one group of cells over another. And, while some consumer-grade SSD is just now beginning to incorporate the latter features."

"It matters whether the SSD drive uses SLC or MLC memory. SLC generally endures up to 100,000 write cycles or writes per cell, while MLC can endure anywhere from 1,000 to 10,000 writes before it begins to fail," [according to Fujitsu's vice president of business development Joel Hagberg.]

Most of the articles on flash media that I have seen here provide no citations to back up statements about wear leveling being used or not used in various devices. The main citation used in several articles for the number of write cycles seems to be a web page documenting a Linux file system, which gives a number with no basis whatever (and it isn't clear to what extent the source is a WP:RS for that information). The infoworld article cited may be mistaken in these matters, but at least it is a citation in an at least moderately reliable source.

If there are other views on the matter - fine, provide some reliable sources and lets fix it up. (i.e., I make no claims for the correctness of the information in the infoworld article, but by Wikipedia standards, it is better than what was there in that it is verifiable) Zodon (talk) 01:18, 21 September 2008 (UTC)

My edit comments was obviously too brief, so I will have to try to be even more explicit: First, as you seem to imply yourself, the "cited" article may not be very reliable, as it uses a quite sloppy and tendensious tone, even in its own quotations. Furthermore, a more complete version of your quotation above would be:

"Consumer-grade SSD generally uses multilevel cell (MLC) NAND flash memory, which has greater capacity and a lower-price point but suffers from slower I/O and as much as 10 times fewer read/writes over its life span. Corporate-grade SSD uses single-level cell (SLC) NAND memory and multiple channels to increase data throughput and wear-leveling software to ensure data is distributed evenly in the drive rather than wearing out one group of cells over another. And, while some consumer-grade SSD is just now beginning to incorporate the latter features to increase its performance, there will still be a cost/capacity disparity for years to come."

Here, "the latter features" probably aims at "SLC" and "multiple channels", not at "wear-leveling" (as it reads "performance"). The fact that wear-leveling is mentioned last is likely due to careless writing/aiming, ignorance, or the tendensious style that characterizes much of the article. Most importantly, it is certainly not anything from which you could deduce a categorical statement like "wear levelling is not common in consumer level devices"... For that, you would need either an explicit citation from a reliable source or concrete examples of "consumer level devices" without wear levelling.

A longer version of your second quotation reads: "For one thing, it matters whether the SSD drive uses SLC or MLC memory. SLC generally endures up to 100,000 write cycles or writes per cell, while MLC can endure anywhere from 1,000 to 10,000 writes before it begins to fail, according to Fujitsu's Hagberg. For its part, Western Digital's laptop hard-disk drive boasts up to 600,000 write cycles." What about the tone and style here?

Another citation clearly illustrates how the writer guesses: "With software enhancements, MLC can exceed SLC performance, but at its core, it's still MLC memory, which means its life span is greatly reduced because cells store more data more often. In fact, generally speaking, the higher performance in an SSD drive, the longer life it will have because of better drive efficiency.". Anyone with a basic understanding of fundamental "flash memory technology" would see that this is quite confused guesswork.

Here are some other articles and papers about wear levelling at various technical levels; I could not find a single one suggesting that "wear levelling is not common in consumer level devices":

http://www.stec-inc.com/technology/flash_memory_controller.php

http://techreport.com/articles.x/15433

http://www.storagesearch.com/ssdmyths-endurance.html

http://lwn.net/Articles/288657/

http://www.cdfreaks.com/news/14976-Experts-warn-about-SSD-security-risks.html

http://www.engadget.com/2008/02/23/samsung-puts-the-kibosh-on-ssd-reliability-worries/2

http://www.mycom.se/product_info.php?products_id=54602

http://www.e-disk.com/article_misconceptions_ssd_longevity.html

http://www.hardwarecanucks.com/forum/storage/8690-ssd-wear-leveling-partitions.html

http://www.solidkor.com/en/technology/414we.html

http://oss.oracle.com/pipermail/btrfs-devel/2008-February/000513.html

http://www.imation.com/products/pdfs/SSD-Reliability-Lifetime-White-Paper.pdf

http://www.imation.com/products/pdfs/Imation-SSD-Performance-White-Paper.pdf

HenkeB (talk) 12:27, 25 September 2008 (UTC)

My comment about the matter of reliability of the source was purely courtesy. The edit summaries trying to remove the source claimed that the information was incorrect (without giving any citations or basis for that claim). It was and is not obvious that the source is biased or incorrect, I was just trying to invite discussion.

Article interpretation:

1. Since wear leveling is the last feature mentioned, it is reasonable to assume it is one of the "later features," however many other features were intended. I don't see trying to re-interpret it to fit some other idea of what it should say.
2. Since consumer level SSDs do exist and have existed for a while, but the statement was made that they are just starting to incorporate these features, that can reasonable be interpreted as the ones existing prior to now didn't. However I have improved the item by making it closer to what the cited source says. (i.e. that they are starting to incorporate this feature now (2008).

It is not clear what point you are trying to make about the tone and style of the quotation relating to number of write cycles. It seems a straight forward presentation of information. The information given seems to correspond to what I have seen elsewhere as far as flash (I thought the figures for hard disks were interesting, but I haven't seen figures to compare them to.)

I didn't make use of the final section quoted, so I don't see what the point is.

I haven't had a chance to look at the citations you provided yet. I will look at them as time permits, or perhaps you have suggestions for improvements to the item which incorporate these sources which we could discuss? Zodon (talk) 07:57, 26 September 2008 (UTC)

If you really was "trying to invite discussion", then please try consider my arguments.

1. It's not about "some other idea", it's about correctness. Your interpretation would seem reasonable giving the short out of context excerption you prefer to look at, however, a slightly longer version makes that interpretation very dubious (see above).
2. You cannot possible use a single source for your categorical conclusion about "consumer level devices" unless the quality of the source is at the level of a neutral and reliable academic paper. Otherwise, you would need either more than one source or concrete examples to support that kind of bold statement (third time I say this...)
3. Given the fact that wear leveling has been used in various cheap devices (CF, SD, USB-sticks etc) for many years (see references in the wear leveling article, for instance), it would be a little naive to belive that SSDs did not.
4. "I didn't make use of the final section quoted, so I don't see what the point is." Really? If you are that uninterested in technical basics, perhaps you should delegate some decisions on precise formulations to other people (with the knowledge needed to judge whether things makes sense or not).

HenkeB (talk) 14:16, 26 September 2008 (UTC)

I responded to your arguments where they related to the article, agreeing that the wording of the wear leveling part could be improved, and asking for clarification about what you found questionable in the section on number of writes. Beyond that, I requested clarification of what is the point of critiquing the source's style, or the clarity of other sections of the article which weren't used in the Wikipedia article.

The wear leveling article was quite thin on references of the sort you mention when I looked a few days ago, as was the USB flash drive article. I looked at various other articles on flash devices, they tended to mention what wear leveling is, but were short on actual citations backing up claims that they were used in this or that type of device, or measurements of quality/etc.

Why make such a big deal about trying to discredit this one reference? It is a news item, so one does not expect the same level of care/writing/understanding as an article in Nature. It improved the wikipedia article by providing some basis for the statements made there. For the purpose of Wikipedia, verifiability is the quality metric, not "correctness," so lets get more that is verifiable. I suggested at the outset that rather than asserting that something is correct without evidence, that evidence be provided to support that view. I still think that that is a more constructive approach.

If wear leveling is commonly used in consumer grade SSDs - fine, what reference says so?

Does the quality of wear leveling vary much? (i.e., is it a marketing buzzword that everybody claims, but some do better than others).

If MLC drives offer more write cycles than mentioned in the article - how many, and what reference says so? etc. Zodon (talk) 22:06, 26 September 2008 (UTC)

First, what exactly do you mean by "Please cease your personal attacks on me at talk:Solid state drive. If you wish to courteously discuss article improvements, fine. Zodon (talk) 21:53, 26 September 2008 (UTC)" on my talk page? I find that quite discomforting.

I'm trying very hard to discuss article improvements, with very little success unfortunately. Once again, my main point is that you cannot rely on a single arbitrary source for your categorical conclusion about "consumer level devices" unless the quality of the source is at the level of an academic paper. Otherwise, you would need something more, such as examples, to support that kind of bold statement.

Yes, verifiability is important, but verifiability does not mean you can take any source, regadless of quality, and (more or less) loosely "interpret" what it says. You need detailed knowledge of your own (sometimes expertise), to be able to judge which sources can be considered reliable and what information relevant and/or reasonable. Truth and correctness must always be number one for any encyclopedia, that's the reason we have them.

HenkeB (talk) 23:45, 26 September 2008 (UTC)

Update: Answers to your questions:

• "If wear leveling is commonly used in consumer grade SSDs - fine, what reference says so?" As I said above, there is no reason to suspect "consumer grade" SSDs should be exceptions from other flash based devices, such as cheap CF, SD, and USB devices, which have employed wear leveling for several years (see external links in the wear leveling article for instance). The reason it's not written out explicitly may be the same as why computer manufacturers seldom explicitly state that this model has heat sink and fan in order to prevent the CPU from self destruct (if you see my point).
• It varies, but not to the extent that the limited number of block erases easily shines trough, not even in fairly extreme cases. The most pessimistic life expectancy calculations I've seen is 5 years (a hypothetical scenario of continous writing at max bandwith 24/7 for five years); more "typical usage" life expectancies can be anywhere between 23 years and thousands of years (see my links above, for instance, or the wiki linked in the SSD article).
• I think most manufacturers specify 10,000 block erase cycles for multi-level NAND-flash, at least that's the figure in datasheets and papers I have come across. Speculating a little, I suspect most SSDs (as opposed to low cost CF/SD/USB devices) will use single level in the future, as the saving is only a factor of two, and multi-level inherently a little slower.

HenkeB (talk) 17:00, 27 September 2008 (UTC)

I think the article should state the wear leveling algorithms typically differ from drive to drive (often the algorithm is engineered for the specific drive depending on life expectancy, density, and can affect drive performance during heavy wear leveling implementation) Sorry no link to source :( Psignosys (talk) 13:41, 6 January 2009 (UTC)Psignosys

To clarify this, there needs to be studies that test: 1. Wear leveling with sequential writes on an empty drive 2. Wear leveling with sequential writes on an 99% full drive (with all other sectors unmodified) 3. Wear leveling with 4kB random writes on an empty drive 4. Wear leveling with 4kB random writes on an 99% full drive (with all other sectors unmodified)

My hypothesis is that every drive will handle this differently, but that 2 is worse than 1 and 4 is worse than 3. Whether 3 and 4 are worse than 1 and 2 depends on the quality of the implementation. Jimhsu77479 (talk) 16:55, 20 March 2009 (UTC)

'Problems with SSDs on Windows'

The linked article for this section doesn't cover the statements, especially:

"Many Windows users who purchase the MLC disks are baffled as to why the performance of their flash drive is slow." "The final problem is that the NTFS file system isn't well suited for these flash disks." "but the fact remains that MLC and SLC disks both perform exceedingly well in Linux using a journaled filesystem like ext3, attaining write speeds far above manufacturer benchmarks, as high as 126MB/s with sequential writes"

I'm tempted to scrub this section altogether unless someone can provide a source or re-write so that the section more accurately represents what the source says. Macthorpe (talk) 06:46, 22 September 2008 (UTC)

I put in some citation needed tags. Although it could use some polishing of the language, and some of the details need citation or deleting, the basics don't seem to be too unlikely to be true, just needs a bit more citation. I have seen other citations on problems with Windows and flash media (e.g. Windows doesn't handle large write blocks of flash efficiently). Suggest leave it for a while and try to accumulate sources. Zodon (talk) 04:50, 23 September 2008 (UTC)

The claims of this section are specious IMO. The claim about 512 byte sectors is wrong, as the Windows file cache does readahead and writebehind in 64KB buffers minimum. I believe the claim about Windows processes constantly accessing the flash drive is misplaced also, as I don't believe the indexing and similar mechanisms normally pay attention to removable drives. Jeh (talk) 11:30, 12 October 2008 (UTC)

Do you have any citations for that? Various people in the industry say the poor performance under Windows is because of differences between hard disks and SSD, and Vista has been optimized for hard disks.

SSD are not usually treated as removable drives, they look to the machine like hard disks. Zodon (talk) 19:47, 12 October 2008 (UTC)

Yes, various people in the SSD industry say that; file under "yes, they would say that, wouldn't they?" The citation for the 64KB readahead/writebehind in the file cache, page reads, and the modified page writer is of course Russinovich and Solomon, Windows Internals. Also note that the default (and very widely used) allocation cluster size on NTFS is 4 KB. Really, I'm very familiar with the Windows I/O system, file systems, etc., and I just don't see where it is "optimized" for a 512 byte sector size. btw the notion of "running with virtual memory turned off" is pretty laughable. Re. removable drives, sorry, I was thinking of add-on drives, not drives mounted as internal. So things like indexing would still be happening, but I see no reason to think these should burden a SSD proportionally more than they burden a hard drive. Jeh (talk) 00:41, 13 October 2008 (UTC)

Not clear why they would say that about Windows, and not about other operating systems. Note that one of the articles also says that Apple and Sun are working on optimizations for flash. The item started out with some more particulars about how some Linux file systems perform better with flash SSD.

As far as references - I meant to ask about references indicating that either Windows didn't have performance issues with flash based drives, or that the issues were likely due to other causes. The size of the readahead/writebehind cache seemed easy enough to document, wasn't questioning that. But where bottle-necks are can be more subtle (e.g. data misalignment, etc.). Sorry my message above wasn't very clear on that.

Seems like the impact of indexing might depend a lot on how much writing it did and in what pattern. Zodon (talk) 04:22, 13 October 2008 (UTC)

"Tests by an SSD maker show that Windows 2000 is markedly faster than Windows XP, Windows Vista, Mac OS X or Linux, when it is run from an SSD using NAND flash memory.[41] Windows 98 is even faster, but does not support wear leveling, so would cause the drive to wear out quickly.[41]" - This is absolute BS!! Wear leveling is done by the SSD controller. NOT BY THE OS... I read the source and was AMAZED to see it. Unless I am misinterpreting the statement, I strongly believe this is incorrect and should be corrected immediately. The OS (especically today) makes memory accesses at the LBA level. It is up to the drive controller to deteremine which NAND cells to use for that specific LBA. Once a NAND cell becomes worn, it will move that LBA to different physical location on the NAND. The OS will NOT BE INFORMED of this action (nor do they have influence on the matter). Again, please review and change this. —Preceding unsigned comment added by Psignosys (talk • contribs) 16:56, 27 March 2009 (UTC)

Could it be something different about how Win98 accessed the hard disk? (Might also be differences in the structure of the file system - maybe FAT forces a lot of rewriting and interacts poorly with the wear leveling - resulting in faster wear. Not exactly the mechanism indicated in the reference, but could have similar result.) The item does seem to agree with what the reference said, reference indicating alternative view would probably help clarify. Zodon (talk) 03:40, 28 March 2009 (UTC)

But they didn't state the file system used. So you can use FAT for XP (and Windows 2000 and Vista?). Would that make those SSD Friendly OSs misbehave too? I feel that this statement is nonsense... Maybe Win98 doesn't play nice with the wear leveling in the SSD but you can't say that it doesn't support SSD as, to my knowlage, Linux is the only OS with explicit wear leveling when using JFFS, although there may be others. Win2000, XP and Vista are not on that list either... Errolt (talk) 14:11, 30 March 2009 (UTC)

Not used wiki for some time so edit me if its not right.. the wear leveling is done by the SSD in HW on the SSD, pc is not aware its an SSD, the OS is not aware at all that it is running on an SSD at all as far as Any OS with LBA access its an hard disk so makes no matter what OS is used (that is on sale from now Vista or lower). windows 7 (or maybe that JFFS) what will happen with that os there be an command support(or file system + win7 only) so that when free space is free it tells the SSD that that block is free that may sound contrdity but it mean it can wipe that block out so its ready as the current way is that it only wipe the flash block once the SSd has been filled once and it takes alot longer for it to Wipe and Load data into an cell then it does if its all ready been wiped (Intel SSD trys to do it) then all it has to do is Write the data to the cell , pure guess if supports it or not if currant ssds have that support yet, Win7 is doing apart from auto defrag off and optimise page file for minmiale use (good read for every one >> Click me anandtech review , some web site explanes all of this alot better), and Speed problems was All jmicron fault as its drive was optimised for read not write more importantly Small writes that Vista does alot of and xp not so much, even the Dual Jmicron should be compleaty ingored unless you got 2 of them in raid with Write cache turnd on(Vista adv perfoamce on the disk from device manager) Leexgx (talk) 03:04, 4 April 2009 (UTC)

The source is terrible and appears to be second-hand. It appears to be a review of a benchmark, but the magnitude of information that was omitted in the review makes it useless. For example, what performance metric are the quoted percentages referring to, not to mention how are they scaled? Additionally, the person quoted seems to lack the kind of understanding of operating systems that would seem to be necessary to accurately understand and analyze the outputs of a benchmark. For example: Linux "never runs anything in the background". At best, they simply failed to convey their meaning. At worst, it shows a complete lack of familiarity with multitasking operating system concepts. I'm willing to put my faith in the benchmark conductors professional status, but either way, since the original benchmark data or publication is not linked to, I can only conclude that the source is not acceptable. Hence I am removing the entire paraphrase of its content. Could be re-written by considering this review of several different manufacturer's experiences with SSD technology and various Operating Systems, which is much more informative and includes references to the same benchmark amongst its sources. —Preceding unsigned comment added by 164.67.97.133 (talk) 03:51, 8 April 2009 (UTC)

'Problems with SSDs on Windows' The problem is with JMicron JMF602B or JMF602 controller that are used on MLC SSD drives (the JMicron JMF602B X2 just makes it 2x faster then slow when writeing files), ssd re-review should look at that review (Tip press the Print button that is on the page) there is alot of info in there that some may just ingore, allso if you use SSD on RAID or use motherboard chipset that buffers Writes thay hide the Write problems sometimes, the new Indilinx controller for SSD is set to Fix what JMicron broke 1275 Indilinx review i know i am quoteing anandtech alot but thay are an good review web site, the one who posted computerworld site does not match real world problems that is infact the controller on the SSD in the way it handles the MLC flash on the SSD drives is the problem,its not the OS fault that the disk is takeing 300ms to 1-2secs to save files, OS are expecting files to be saved in an timely manner Leexgx (talk) 21:09, 11 April 2009 (UTC)

Manufacturers dropping images here

Alot of the images on the page appear to be added by companies selling these things. While that's not bad in itself, I notice some of them are poorly license (EG. The one I marked just now was listed as CC-BY-SA, however it came straight off a website that was completely copyrighted. 76.117.247.55 (talk) 18:48, 22 August 2009 (UTC)

ExFAT

No credible source actually states that ExFAT is intended for SSDs (SATA connected hard disk-like storage) rather than "flash media". Lots of places hint that it is intended for mobile USB storage-type flash media. IOW; I do not believe what this article says about ExFAT is true. europrobe (talk) 15:35, 17 October 2009 (UTC)

Thanks for taking the time to explain your deletion. You seem to be equating SSDs with DRAM drives, and equating flash media with USB drives. But SSDs and flash media are not mutually exclusive:

• "An SSD is commonly composed of DRAM volatile memory or primarily NAND flash non-volatile memory." – Solid-state drive#Architecture and function

The article's hatnote reiterates this:

• "This article is about flash-based, DRAM-based and other solid-state drives. For other flash-based solid-state storage, see USB flash drive. For software-based secondary storage, see RAM disk."

The terminology is supported by the uses of flash memory described in the article Flash memory#Flash memory as a replacement for hard drives.

In any case, the Microsoft quote which you deleted makes clear that exFAT improves write-boundary alignment, which is important for non-volatile SATA SSDs as well as for those using other interfaces:

• "The exFAT file system driver adds increased compatibility with flash media. This includes the following capabilities: Alignment of file system metadata on optimal write boundaries of the device; Alignment of the cluster heap on optimal write boundaries of the device."[1]

You cite three webpages implying that exFAT is intended for USB flash media. But none of the articles state that exFAT is optimized only for USB media. In fact your third citation explicitly contradicts this:

• "The exFAT file system is a new file format system to address the growing demand and size of mobile storage like USB sticks, PDAs, and solid state hard drives." [emphasis added]

It seems clear that exFAT is implemented to take account of SSD characteristics. I have therefore reverted your deletion and added this last quote as a supporting reference.

Hope that clarifies and helps us avoid conflict. Richardguk (talk) 04:46, 18 October 2009 (UTC

Thanks for your detailed response. I realize that I have run out of arguments in this matter. Based on what we have discussed, I would say that none of us can really say for sure whether ExFAT is optimized for (flash based) SSD:s or (flash based) USB storage. It would seem that NTFS (or ext4 or similar) would be a much better choice for a desktop or server file system. ExFAT still lacks support for several things which users have come to expect from an advanced OS. It only makes sense to use ExFAT for mobile storage. Time will tell, I guess. europrobe (talk) 20:11, 18 October 2009 (UTC)

SLC versus MLC

This section is either missing information, or contains an irrelevant example. The paragraph opens with a statement that MLC is slower than SLC, then notes MLC slowness can be mitigated by a variety of mechanisms. This is all good. The problem: The "For Instance" followup explains how a (faster) SLC drive beat (4) (slower) MLC drives in RAID(0) configuration. If SLC is faster than MLC, an example is not needed of how an SLC drive will beat MLC drives. Some critical content is missing for the example to be meaningful; without that content, the statement needs to be stricken. 68.177.173.34 (talk) 17:18, 29 January 2010 (UTC)

The whole sentence doesn't make any sense they compare apples to oranges, the SLC "hard drive" is in fact a 4x PCI-Express and the MLC is actually a array of 4 MLC drives. In reality it's a whole different matter some MLC ARE better than some SLC drives, and the other way around. The SLC drives may have the potential to be better but it's not just a question of the memory used but also of the electronics around it, put the SLC with some crappy electronics and it will be even worst than a MLC drive.Strumf (talk) 10:19, 22 April 2010 (UTC)

@_@ —Preceding unsigned comment added by 66.229.204.110 (talk) 11:36, 11 June 2010 (UTC)

Tandy ROM - Did not belong

Because this article is focused on rewritable storage, the following comment did not pertain: "Tandy MS-DOS machines were equipped with DOS and DeskMate in ROM, as well" so I have removed it.§ Music Sorter § (talk) 22:19, 13 June 2010 (UTC)

"a certain degree of fragmentation is actually better for reads"

Source ?Rex4 (talk) 09:28, 16 July 2010 (UTC)

Check Anandtech[2]. "Fragmentation" is not the best word to describe; rather, SSDs use a sort of parallel writing scheme in order to boost performance by utilizing the inherent parallelism of multiple channels. The "fragmentation" problem is when one performs many small writes, by repeated cycles of writing-reading-invalidation, the spots that are free to accept data become randomly distributed within any particular flash channel, making write speeds very slow (because the SSD has to do an erase cycle for nearly every write).

So in essence, the statement is correct but should be better worded. —Preceding unsigned comment added by Jimhsu77479 (talk • contribs) 19:51, 28 July 2010 (UTC)

Can we get a rating system in place?

How about a rating system of the top SSD's to date?

I have been scouring the web to find the Solid State Drive for my machine. Find one than a bit later find another. Been continuously changing my mind... A list of the top benchmarked and reviewed drives would be ideal for a new hardware item like this...—Preceding unsigned comment added by 76.24.87.68 (talk) 06:45, 4 August 2010 (UTC)

{Editorial note: the original question posed was a request for feedback on a particular SSD. I removed that content and kept the portion that could contribute to the article per Wikipedia:What Wikipedia is not#FORUM. § Music Sorter § (talk) 17:44, 7 August 2010 (UTC)}

I don't think Wikipedia is the right place for a rating system that would be based on primary research. Unless there is an industry standard for ratings and a controlled method for measurements with no variance possible, there is no way to control the outcome. The encyclopedic nature of Wikipedia is to present what is already published. I think we might be able to talk about a rating system if it existed, but we would be less likely to present the information as a primary source. § Music Sorter § (talk) 17:44, 7 August 2010 (UTC)

exFAT and no evidence of benefits to an SSD

I believe there have been some debates on this article in the past if exFAT did anything for SSDs or not. I searched the MS site for any evidence and found nothing. Certainly they differentiate between SSDs and USB flash devices with Windows 7, but for exFAT they only talk about USB or flash devices. SSDs are different from generic flash devices. The MS source cited below never uses the term SSD which is suspicious because they use it a lot on Windows 7. Also the other article cited is the only article that discusses SSDs with exFAT. Since no other prominent sites support this claim it seems very dubious and should be removed until we can find additional support or evidence that it is somehow beneficial for SSDs.

Microsoft's exFAT file system is optimized for SSDs. According to Microsoft, "The exFAT file system driver adds increased compatibility with flash media. This includes the following capabilities: Alignment of file system metadata on optimal write boundaries of the device; Alignment of the cluster heap on optimal write boundaries of the device." Support for the new file system is included with Vista Service Pack 1 and Windows 7 and is available as an optional update for Windows XP.

§ Music Sorter § (talk) 09:41, 9 August 2010 (UTC)

References

The first citation already refers explicitly to "solid state hard drives". The second citation (which is from Microsoft) refers to "flash media", but the Wikipedia article itself notes: "As of 2010, most SSDs use NAND-based flash memory" (2nd paragraph). In other words, SSDs are a type of flash drive (in a technical if not a marketing sense) and therefore the Microsoft statement encomapasses SSDs. Surely alignment and levelling are beneficial in SSDs just as they are in older types of flash drive? — Richardguk (talk) 20:00, 9 August 2010 (UTC)

The first citation focuses on UBS, CF, and similar small configuration Flash storage. While the article does one time mention "solid state hard drives" as you pointed out, it would not be the first time an author of a short article misunderstood the content in which he was writing.

Although USB Flash sticks and SF cards both use Flash memory, full SSDs are very different in many ways. Look at the performance of a SATA SSD and compare it to the performance of a USB Flash drive. They are orders of magnitude different in performance. This in part comes through the operating system.

As I previously mentioned the second source is from Microsoft (the author of the code) and they never once used the term solid state drive or SSD. No other sources claim exFAT is good for SSDs. If it was good for it, Microsoft would have added it to their own page and others would have written about it. I recommend we continue to hold off returning this material until we find another reputable source to support the claim. Better yet an article from Microsoft making a statement to support the claim. § Music Sorter § (talk) 15:44, 26 August 2010 (UTC)

SSD optimized file systems - Page file?

In the article under "SSD optimized file systems" it says "The page file should also be disabled because the constant updates to the file will cause unnecessary additional wear on the SSD that is not offset by enough performance gain." However, there's been a lot controversy about whether or not this is true. I believe Microsoft and Intel (can't cite sources) have said that there is no better place to put your page file to, than SSD. This should be true, unless you have enough RAM to be able to disable the page file altogether. I'd like to hear some expert thoughts on this matter. Jokkk (talk) 13:52, 15 November 2010 (UTC)

Quality and performance section quality

The entire section titled "Quality and performance" seems terribly written, especially with regards to mentioning time frames and the organization statements. The section is mostly just a spewing of random statistics on SSDs at different dates without making proper comparisons (to HDDs, or newer/older SSDs) nor including the context of the statistic/statement. SSDs are used for a variety of applications, and posting business networking sources along with residential-use sources without distinguishing them is terrible. Frankly I cannot believe such a piece of text could exist for such a long period of time on such an important/popular article.

I recommend that the section be completely re-written without including any of the currently mentioned statistics or statements, and that new ones be found pertaining to information regarding: performance over time changes for SSDs which are properly maintained (as well as improperly maintained), performance of new SSDs vs new HDDs be it in RAID or otherwise (keeping mention of the application it's being used for), and relevant performance flaws that exist in specific SSDs or generations of SSDs.

Lastly, somewhat as the section says already, a good statement possibly make at the start would be an explanation that SSDs performance are changing/increasing quickly over time and the precision of the section may not be adequate due to age. 24.79.135.235 (talk) 08:28, 26 December 2010 (UTC)

SSD vs Compact Flash

Despite the length of the article, seems substantially absent an IMPORTANT point: is the SSD another -marketing name- for Compact Flash? or IN WHAT it's different (well, apart the DRAM SSD) ? The CF uses NAND flash, and also SSD. The CF have wear leveling and read-write cycle problems, like SSD The CF (integrated) controller support ATA/IDE specs, like SSD. (usually is needed a passive CF-IDE adapter for different pinout from IDE) The CF (integrated) controller support SATA, like SSD. MAYBE "SSD" is a superset of CF.. but is pretty unclear. 88.149.244.131 (talk) —Preceding undated comment added 16:44, 3 June 2011 (UTC).

The similarities between a CF (Compact Flash) or any memory card and a solid-state drive relates more to its interface and primary use. The CF cards and other memory card types are primary used for digital cameras and mobile phones with capacities ranging from 1GB to 64GB today, while SSDs are primarily used for primary data storage in computers with capacities ranging from 64GB to 512GB or more today. I think it is a great question and I will work on an addition to both this article and the memory card article covering CF cards. § Music Sorter § (talk) 22:53, 16 June 2011 (UTC)

Suggestion for image: MacBook Air SSD

I remember seeing images of a MacBook Air teardown on some website. The image of a very tiny "motherboard" with all the SSD storage attached directly to the board (instead of a separated "drive" connected to the board) was a very striking one. Will be great to have it included on this page. Would love to do this myself, but I'm not sure about the copyright issues. — Preceding unsigned comment added by 122.167.208.65 (talk) 17:35, 12 December 2011 (UTC)

Really significantly biased.

Not enough treatment of the problems with SSD technology. While firm-ware bugs are mentioned, (and IMHO cavalierly dismissed), whether the failure of the drive is firmware or hardware caused is irrelevant to most users. The technology is poorly covered here. No discussion of the trade-offs between single layer and multiple layers, nor is there any clear exposition on the virtual certainty that only a finite number of write cycles will be tolerated by each bit. I rate this article marginal in its objectivity. Clearly it is NOT acceptable for permanent memory to fail "catastrophically" and yet this issue is very lightly touched upon. And confounding DRAM technology is another seemingly intentional obfuscation. It is very questionalble that the different technological approaches to SSDs should be lumped together in a table compared to HD technology.71.31.146.16 (talk) 20:04, 17 June 2012 (UTC)

Add a bit on 'reliability' please.

Thank you for this article. The question of reliability seems to come up quite often when discussing SSDs. Although you address this somewhat in your article, I've yet to come across the following aspect, which would be useful and reassuring to have for the reader: What happens when the SSD dies? If the drive simply becomes a brick, that would be useful to know. If the drive simply stops accepting write commands, then that too would be useful to know - which is to say, when an SSD dies, can it simply be cloned on to another one and you get on with your day, or is the data simply gone. Granted, it depends upon the nature of the failure, but someone examining this article is usually looking for 'what are these beasties', and thus, typical scenarios. Done but readable matters - an OS can be reloaded (but PITA), but reassembling the particular collection of programs and data, once dead, can be challenging. (Never mind that backups should be made, etc., nor the challenge of live snapshotting drive images.) So, it would be useful if the typical 'death state' of an SSD were mentioned. Bs27975 (talk) 04:28, 9 July 2012 (UTC)