![]() | This is an archive of past discussions. Do not edit the contents of this page. If you wish to start a new discussion or revive an old one, please do so on the current talk page. |
Archive 1 | ← | Archive 4 | Archive 5 | Archive 6 | Archive 7 |
The column "minimum number of drives" is just plain incorrect. First of all, it appears to refer to a nondegenerate minimum number of drives which is completely different from the actual minimum number of drives (for example, a 3-disk RAID-1 can be considered a degenerate 3-disk RAID-6; this is important for implementation since it may be useful for reshaping), but second of all, requiring 4 drives for RAID-5 is flat-out incorrect. Hpa ( talk) 02:39, 3 January 2012 (UTC)
Is there a standard for triple-mirroring? This is basically the same thing as a standard two-drive mirror configuration, except the data is written to three drives at once, and can be read from any of the three. Any two drives can fail and no data is lost.
This would be an acceptable alternative to RAID-6 when you have huge 300+ gig hard drives, but only say 100 gig of server data. No parity engine is needed, just write duplication across independent parallel controller channels for maximum performance.
+N Mirror basically means data could be mirrored across however many drives you want, up to the controller's interface limit. Eight redundant drives? Sure, why not.
DMahalko ( talk) 03:27, 24 February 2012 (UTC)
The Array Failure Rate, as defined in the table, lists the probability of the data loss in the given period under assumption that no drives are replaced during this period. This metric, while in itself valid, is not what one would typically expect as the "failre rate". Leaving the dead drive without replacement is against any maintenance practice and/or recommendations. I suppose we should at least add another footnote clarifying this? Alexey V. Gubin ( talk) 19:44, 12 March 2012 (UTC)
According to Patterson and Hennessy 2012 [1], RAID 2 is Error Detecting and Correcting Code, and RAID 3 is Bit-Interleaved Parity. This is different from what is in the article. Espertus ( talk) 00:40, 29 March 2012 (UTC)
If the section concludes, "In short, the choice between RAID 5 and RAID 10 involves a complicated mixture of factors. There is no one-size-fits-all solution, as the choice of one over the other must be dictated by everything from the I/O characteristics of the database, to business risk, to worst case degraded-state throughput, to the number and type of drives present in the array itself. Over the course of the life of a database, one may even see situations where RAID 5 is initially favored, but RAID 10 slowly becomes the better choice, and vice versa.", I don't see how the neutrality of the section can be called into question. — Preceding unsigned comment added by 68.104.214.16 ( talk) 15:39, 30 April 2012 (UTC)
Isn't saying "RAID Array" just like saying "ATM Machine"? Why stop at array? Why not call it "RAID Array of Independent Redundant Discs"?
173.13.189.193 ( talk) 22:15, 17 September 2012 (UTC)
In the history section the claim is made that: "In October 1986, the IBM S/38 announced "checksum" - an operating system software level implementation of what became RAID-5" which is one year before Patterson defined the term RAID. I have done a pretty thorough search for literature evidence for this and can find none- does anyone have a reference. Otherwise I will remove it in the future. 121.45.215.68 ( talk) —Preceding undated comment added 04:35, 2 December 2012 (UTC)
The section "RAID 10 versus RAID 5 in relational databases" has several problems- the main one is that RAID5 is no longer recommended or used for any large data set, presumably including most databases. Could this section be generalised to compare parity versus non-parity (mirrored) RAID schemes? Also, there is a lack of references. 121.45.215.68 ( talk) —Preceding undated comment added 22:44, 3 December 2012 (UTC)
The sentence "a single drive failure results in reduced performance of the entire array until the failed drive has been replaced and the associated data rebuilt" appears twice in the description of RAID 5. Somebody should fix it. 173.34.179.126 ( talk) 14:19, 22 December 2012 (UTC)
I've just made a few edits to improve clarity.
I'm looking for constructive criticism regarding the following ones:
(and feel free to make them for me.
CONFUSING:
"Background scrubbing can be used to detect and recover from UREs (which are latent and invisibly compensated for dynamically by the RAID array) as a background process, by reconstruction from the redundant RAID data and then re-writing and re-mapping to a new sector; and so reduce the risk of double-failures to the RAID system[42][43] (see Data scrubbing above)."
Idea:
Add a subheading to this section and the one above what's CONFUSING
and
REWORD THUS:
Double....
or perhaps
Section on Atomicity: Wow, this is confusingly written. Perhaps it should kick off with the simple statement, like "Databases are designed to maintain data consistency despite the non-atomicity of the write process of the disk drives on which they normally store data, even when there is unexpected power loss." If I'm not mistaken, either disk drives can't perform atomic write processes, period, or ones that can are rare enough to merit little more than a parenthetical.
Write case reliability:
P(aragraph)3 doesn't mesh with P1 and P2. In fact all 3 are about write-back, but only P3 calls it that. Merge 'em.
Hardware Labeling:
Some systems have a feature intended to reduce operator error and facilitate drive failure management by having an LED by a failed drive indicate that it is a drive that has failed.
History:
REMOVE:With S/38 checksum, when a disk failed, the system stopped and was powered off. Under maintenance, the bad disk was replaced and the new disk was fully recovered using RAID parity bits. While checksum had 10%-30% overhead and was not concurrent recovery, non-concurrent recovery was still a far better solution than a reload of the entire system. With 30% overhead and the then high expense of extra disk, few customers implemented checksum.[citation needed]
doesn't make sense.
It's so confusingly written I don't know what it's trying to say or how to fix it.
-- Elvey ( talk) 11:42, 30 December 2012 (UTC)
I have reverted two edits by 145.225.17.100 ( talk · contribs) ( http://en.wikipedia.org/?title=RAID&diff=523846804&oldid=523363466 and http://en.wikipedia.org/?title=RAID&diff=523846913&oldid=523846804 )due to the fact they removed the entity of the Standard Raids section, removed part of the cite in the nested/hybrid section, and added no content. I'm going to assume this is either vandalism or accidental, though the IP does have other changes elsewhere that look less suspicious. Anything needed other than undoing the edit? Nazzy ( talk) 15:58, 19 November 2012 (UTC)
The article claims RAID 3 is a 'theoretical' RAID level and 'not used in practice.' This is incorrect, EMC Clariion and VNX support a RAID 3 option. — Preceding unsigned comment added by 98.229.192.30 ( talk) 21:28, 14 December 2012 (UTC)
121.45.193.118 ( talk) 12:09, 20 December 2012 (UTC)
You can see the RAID 3 reference, as well as 0, 1, 1/0, 5, and 5 in the current (1/7/2013) datasheet for EMC VNX on their website. http://www.emc.com/collateral/hardware/data-sheets/h8520-vnx-family-ds.pdf — Preceding unsigned comment added by 168.159.213.60 ( talk) 20:02, 7 January 2013 (UTC)
The section ==Mirroring versus parity RAID levels in relational databases== is very problematic- e.g. the claim that "Given the rare nature of drive failures in general, and the exceedingly low probability of multiple concurrent drive failures occurring within the same RAID, the choice of RAID 5 over RAID 10 often comes down to the preference of the storage administrator" is incorrect-- UREs are extremely likely to happen with large disks and thus RAID 5 is not recommended for large databases by the major storage manufacturers (as noted in the main article). The references given are to a non-authoritative web site and do not support the claims in any case. A well-referenced section on this topic could be useful, but the current section is misleading to users of RAID. The current text is copied below for improvement before possibly returning to the main article. 121.45.213.224 ( talk) 06:54, 28 January 2013 (UTC)
It seems that RAID 10 and RAID 1 should have the same space efficiency formula, namely 50% since they both only use mirroring to tolerate single drive failures. — Preceding unsigned comment added by 99.245.3.15 ( talk) 19:40, 1 February 2013 (UTC)
What you’re referring to as Linux, is in fact, GNU/Linux, or as I’ve recently taken to calling it, GNU plus Linux. Linux is not an operating system unto itself, but rather another free component of a fully functioning GNU system made useful by the GNU corelibs, shell utilities and vital system components comprising a full OS as defined by POSIX.
Many computer users run a modified version of the GNU system every day, without realizing it. Through a peculiar turn of events, the version of GNU which is widely used today is often called “Linux”, and many of its users are not aware that it is basically the GNU system, developed by the GNU Project. There really is a Linux, and these people are using it, but it is just a part of the system they use.
Linux is the kernel: the program in the system that allocates the machine’s resources to the other programs that you run. The kernel is an essential part of an operating system, but useless by itself; it can only function in the context of a complete operating system. Linux is normally used in combination with the GNU operating system: the whole system is basically GNU with Linux added, or GNU/Linux. All the so-called “Linux” distributions are really distributions of GNU/Linux. — Preceding unsigned comment added by 175.100.80.20 ( talk) 10:33, 4 February 2013 (UTC)
Hello! I'm concerned the table might be slightly wrong. Specifically, under column "Read performance" for level RAID1, it is stated "nX", which supposes the performance will scale as read is split among the drives (same process as RAID0). I believe this is not accurate.
The article itself states, under the RAID1 section: "the read request is serviced by either of the two drives containing the requested data".
So which one is it? Does RAID1 allow split reading on numerous drives, or can only a single drive in the array be read? From what I know, and from what I found searching, this seems to be entirely dependent on the RAID controller being used. Findings in this blog is that RAID1 read speeds are either slightly lower or equal to the read speeds of a single drive (that was either both a Windows Software RAID1 and an Intel ICH5R integrated Motherboard (aka Fake RAID) controller). I wasn't able to find actual proof that some controllers do in fact support split reading, but some people do mention that certain OS are able to do it. Quantos88 ( talk) 16:07, 29 March 2013 (UTC)
The figures used in the "Comparison" table (currently section 2.9 as I write this) are very useful, but only to people who already understand the technology. For anybody else -- say, somebody trying to use this page as if it were out of an encyclopedia -- the lack of explanatory keys makes the right-hand column much less helpful than it could have been.
Specifically, we need a brief explanation of what "A1", "B2", etc, mean in this context. Yes, somebody familiar with RAID knows that these represent storage blocks (or other sizes depending on the RAID level), but there's no obvious "redundancy" in the images. Even for, say, RAID 5, it's not obvious that any of that data has in fact been stored more than once.
The reason I didn't try to add such a sentence is that I wouldn't know how to write it. I had pointed a colleague at this page when they asked for a breakdown of RAID levels, and they came back more confused than they had been at the start. 99.102.75.166 ( talk) 14:43, 15 April 2013 (UTC)
In the main table formulae are given of array failure rates but no citations are given to confirm them, and indeed they have been changed several times such that I have no confidence that they are correct. Explicit references for each of them is needed- I added a "citation needed" previously but this has been removed. 74.73.169.211 ( talk) 16:43, 21 April 2013 (UTC)
Under the table in 'Comparison' it is stated that
**** Raid 10 can only lose 1 drive per span up to the max of 2/n drives
Should this be n/2 drives? Michaelcochez ( talk) 09:02, 15 July 2013 (UTC)
IEEE.org as of 4-6 Dec 1991 and updated as of Aug 2002 cited Inexpensive, [1]
IEEE.org as of June 29 2009-July 2 2009 and updated as of 29 September 2009 cited Independent, [2]
NetApp -Independent [3]
EMC - Independent [4]
Western Digital Independent [5]
A paper from Duke University Dept. of Computer Science in 1993 - Inexpensive [6]
in the section "reliability terms" it says: "System failure is defined as loss of data and its rate will depend on the type of RAID. For RAID 0 this is equal to the logical failure rate, as there is no redundancy." however wouldn't the faliure rate be higher, asuming each drive in the array has a chance of faliure and the data is spread across more than one drive? ~patrick
A common opinion (and one which serves to illustrate the dynamics of proper RAID deployment) is that RAID 10 (a non-parity, mirrored RAID) is inherently better for relational databases than RAID 5, because RAID 5 requires the recalculation and redistribution of parity data on a per-write basis. [1]
There are, however, other considerations which must be taken into account other than simply those regarding performance. RAID 5 and other non-mirror-based arrays offer a lower degree of resiliency than RAID 10 by virtue of RAID 10's mirroring strategy. In a RAID 10, I/O can continue even in spite of multiple drive failures. By comparison, in a RAID 5 array, any failure involving more than one drive renders the array itself unusable by virtue of parity recalculation being impossible to perform. Thus, RAID 10 is frequently favored because it provides the lowest level of risk. [2]
Modern SAN design largely masks any performance hit while a RAID is in a degraded state, by virtue of being able to perform rebuild operations both in-band or out-of-band with respect to existing I/O traffic. Given the rare nature of drive failures in general, and the exceedingly low probability of multiple concurrent drive failures occurring within the same RAID, the choice of RAID 5 over RAID 10 often comes down to the preference of the storage administrator, particularly when weighed against other factors such as cost, throughput requirements, and physical spindle availability. [2] failed verification
The formulas for the Array Failure Rates in the Comparison Table were changed recently (April 2013). Does anyone have a reference for the calculations? Mobileseth ( talk) 03:23, 24 August 2013 (UTC) mobileseth
Biassed in putting SAS/enterprise drives in italics. Reduces the error handling to nothing more than setting TLER. ("How to not let a drive with bad blocks drop out so you need to replace it. Instead keep running with it! Yay!") — Preceding unsigned comment added by 2001:A60:11FA:6701:993E:921B:42E4:8DFE ( talk) 19:41, 2 December 2013 (UTC)
It doesn't even mention stuff like 520byte vs 512byte formatting or that normally drives in a hw raid get updated firmware that was tested for reliability. I guess next week someone will update it showing that only ZFS knows how to "check" a disk. — Preceding unsigned comment added by 2001:A60:11FA:6701:993E:921B:42E4:8DFE ( talk) 19:08, 2 December 2013 (UTC)
![]() | This
edit request has been answered. Set the |answered= or |ans= parameter to no to reactivate your request. |
Please change 2/n to 1/2 in (1) the "Space efficiency" column of the Raid 10; (2) "Raid10 can only lose one drive per span up to the max of 2/n drives" Oyang2010 ( talk) 19:11, 3 April 2014 (UTC)
The table claims that the write performance for RAID levels 4,5, and 6 are (n-1)x, (n-1)x, and (n-2)x, respectively. presumably n is the number of disks in the set and x is the write performance in IOPS of one such disk.
These claims are contrary to industry norms, which are x, nx/4, nx/6, respectively in the above list.
The article should provide a reference for the claims or they should be removed. — Preceding unsigned comment added by 168.159.213.208 ( talk) 17:31, 16 April 2014 (UTC)
I undid Dsimic's last edit, where several instances of "data is" were again changed to "data are". Numerous editors keep editing this article to change these back to use the singular verb form, in what can only be described as edit warring. Unlike traditional warring, though, I've observed Dsimic versus the world instead of just one editor.
While I agree with Dsimic that the plural Latin nouns usually end in "a" and single Latin nouns in "um", Wiktionary offers a very insightful a third possibility. Basically, "data" can also be used as an uncountable noun—which is the norm in computing contexts like RAID. Surely that's because individual bits number in the trillions in today's machines and thus counting becomes a vain exercise. Similarly, water is made up of individual molecules but "a glass of waters" is poor grammar. I believe this to be why the vernacular supports the singular form, and why we should stick with it here. – voidxor ( talk | contrib) 05:40, 14 June 2014 (UTC)
A recent edit by Snori describes super-expensive SLED devices as “super-reliable” and the smaller, less expensive disks as “comparability unreliable”. That's not actually true, and I think it skews the description of the basic idea behind RAID.
The leading motiviation to develop RAID was the higher performance which could be realized; associated benefits include larger and more flexible storage capacity, and reductions in equipment cost, size, and power consumption. Single disk reliability wasn't a big factor. Unician ∇ 07:52, 1 September 2014 (UTC)
@ Dsimic:The 3330, actually the 3830 storage control used a Fire code appended to the end of each block for error correction. It was not a Hamming code as used in RAID-2. More significantly, the Hamming code works in real time while the appended Fire Code works after a block has been buffered; so the reference to the 3330 should be deleted, regardless.
I am very certain that the 353 did use a Hamming code so it that sense it was similar to RAID-2. It was actually a Redundant Array of "Independent" Heads, 40 heads transferred in parallel, 32 data, 7 Hamming code and 1 parity. That's why the term "similar approach" is used in the article. I'd have to dig a bit to find a reference if this is disputed, but I am sure i can find one. Tom94022 ( talk) 01:55, 15 January 2015 (UTC)
e.g., see
page 157 of IBM 7030 Data Processing System Reference Manual; it discloses 32 data bits plus 7 ECC bits in the "high speed disk storage unit" which I am pretty sure is the IBM 353.
Tom94022 (
talk)
18:27, 15 January 2015 (UTC)
Dsimic nice find with the Katz quote. I believe I heard him say the same thing at some point but didn't bother to look for it because Strebe mis-states what the article says and then mis-applied WP:SYNTH. The RAID paper defines RAID-1 as disk mirroring. The literature of the far earlier DEC product discloses disk mirroring. It is no violation to then say disk mirroring existed before the term RAID-1 was defined. Ditto for each of the other RAID levels and for the introductory sentence. But that is moot now, thanks to yr great work. Tom94022 ( talk) 07:46, 18 January 2015 (UTC)
I propose that Standard RAID levels be merged into RAID. I think that the content in the Standard RAID levels article can easily be explained in the context of RAID, if it is not already. I cannot justify the duplication of effort when it comes to listing RAID levels. One example is the recent addition of a table of levels to the Standard RAID levels article, when a much more established table of differences already exists in the RAID article. Listing basic RAID levels in two places is redundant, so I removed it. I felt bad when doing so because an editor worked hard on the new table, having not seen the existing one. It's easy to see how one would land on Standard RAID levels and assume they are on Wikipedia's one-and-only RAID article. – voidxor ( talk | contrib) 06:49, 23 January 2014 (UTC)
Hello, 67.6.185.65! As I've noted in my edit summary, RAID 10 is a nested (hybrid) level, not one of the standard RAID levels. Thus, it is wrong to add RAID 10 into the RAID § Standard levels section; please see RAID § Nested (hybrid) RAID section for further information. — Dsimic ( talk | contribs) 20:01, 14 April 2015 (UTC)
In the summary of RAID 4, it is contrasted with RAID 2 and 3. A quote "As a result, more I/O operations can be executed in parallel, improving the performance of small transfers.[2]" This is only true for reads. It is untrue for writes. RAID 4 serializes writes and can perform only 1 at a time. This is because all writes, by definition, must update their respective parity data and in RAID 4 this parity resides on a single spindle. RAID 4 does one write at a time, whether small write or full stripe, exactly like RAID 2 and 3. Worse it does so without their bandwidth advantages.
Further the reference to NetApp's RAID-DP is misleading. It's not RAID-4. Its closer to RAID 6 since each stripe has two parity blocks and one is diagonally formed.
Next, the key advantage of RAID 5 is that its parity is distributed over all the members of the raid string. Unlike RAID 4, a write is not serialized by a single spindle. On a RAID 5 array the number of simultaneous writes can be as a high as 1/3 of the number of members in the string, e.g. a 15 drive string could have 5 simultaneous writes occurring. A RAID 4 string can never have more than 1 regardless of its size. The article leads you to believe RAID 4 has some advantage, when it's all the other way.
Lastly, all of the negative points you make about RAID 5 are true for RAIDs 1, 3, and 4 as well. A read failure during rebuild or degraded operation (one drive failed) can happen for any of them and would be equally unrecoverable. You don't point any of that out.
The comments on RAID 5 have a 'sky is falling' tone and lack the above perspectives. — Preceding unsigned comment added by 73.16.141.73 ( talk) 00:29, 7 December 2015 (UTC)
Could add something about assumptions: who is expected to benefit from this? No mention of Windows 7 Pro. User trying to do mirroring (RAID 1) having come across it in the create and format hard disk partitions section of control panel. Well the "system crash" section in this Wiki may be a warning to the newcomer, but Dsimic has labelled my point about power interruption as pretty much a nonsense. Is Dsimic suggesting the assumption that a UPS is what everyone would already have before thinking about RAID? Where would they have come across that knowledge? On a non-mirrored system a power failure is overcome easily at the next boot. It does not require a message sent to Microsoft &c, the way a "crash" may. Soundhill ( talk) 12:00, 19 January 2016 (UTC) Soundhill
The article says:
> The array will continue to operate so long as at least one member drive is operational.
There is no description of the failure scenario.
For people that are new to RAID, we could use a more complete description of what happens when a drive fails in RAID 1. For instance, is there a light or alarm, etc? Just another line or two would complete the description. Consider that a RAID 1 enclosure for two drives is the least expensive and most marketed RAID enclosure, so the Wikipedia description of RAID 1 is going to bring a lot of views. Thanks 75.110.98.103 ( talk) 23:38, 4 February 2016 (UTC)
Hello, A876! Regarding your edit comment, there are three things:
|website=
parameters (where applicable, of course).With all that in mind, I've restored the references and improved the wording a bit further. Hope you agree. — Dsimic ( talk | contribs) 22:30, 20 April 2015 (UTC)
Any chance of mentioning the 'old' name for RAID; which was "Redundant Array of Inexpensive Disks"?
203.214.22.116 (
talk)
01:13, 14 March 2017 (UTC)
Oops! I just found it... Sorry. 01:15, 14 March 2017 (UTC) — Preceding
unsigned comment added by
203.214.22.116 (
talk)
I just dumped the citation needed for "Needs 4 discs". There's a good explanation on the Standard_RAID_levels page, but the summary of a 3 disc system would be: Disc 1: Data Disc 2: XOR of all the data blocks. With one disc that's just a copy of the data. Disc 3: Clever correction based on the data. That need be no more than a copy. (then rotates around the discs for the other data blocks) That's just a 3-way mirror. — Preceding unsigned comment added by Number774 ( talk • contribs) 14:38, 5 July 2017 (UTC)
The article states RAID 1 offers no parity, but a mirror is an even parity. I think it adds no value to explain this to the reader, but maybe it can be corrected? — Preceding unsigned comment added by 145.53.72.151 ( talk) 18:54, 19 January 2016 (UTC)
Actually the above comment is wrong on several counts. First a RAID 1 array could have more than 2 members. HP OPENVMS provides this, up to 8 in fact in what they call a shadowset. With more than 2 disks you can catch inconsistent copies. Further it implies that parity systems check parity on reads, they don't. This would require reading the entire stripe, which is too much overhead. With RAID-5 this would be useless as you wouldn't know which version was correct. With RAID-6 some implementations scrub parity in the background, since there are three sources for each block, the block itself, first parity, and second parity. This allows for a voting mechanism. Uncommon RAID systems such as RAID-3 might actually check parity on each read but it is an academic point, there are very few if any commercial versions of them.
As well the article claims RAID-1 read performance is "slower than the fastest drive". In virtually all implementations a read is queued to the disk member with the shortest operations queue. In practice RAID-1 sets give superior read response times and total read rates as compared to single drives. — Preceding unsigned comment added by Sciencia61 ( talk • contribs) 12:35, 19 June 2018 (UTC)
You can see the original paper here: http://www.cs.cmu.edu/~garth/RAIDpaper/Patterson88.pdf RAID was originally a cost reduction method, to allow for the use of cheaper disks than were sold by the then large vendors (IBM and DEC). Parity was added to make up for the perceived unreliability of the cheaper disks. Later when RAID was commercialized, and made expensive by the likes of EMC and NETAPP, marketing efforts to justify the now high prices required the change from inexpensive to independent. You can read one of the original authors (Katz) commenting on the very issue in the last paragraphs of this paper: https://web.eecs.umich.edu/~michjc/eecs584/Papers/katz-2010.pdf — Preceding unsigned comment added by Sciencia61 ( talk • contribs) 12:53, 19 June 2018 (UTC)
I added a further reference from the SNIA document explaining that while write caches had a power fail vulnerability, good implementations design for this and add methods to ensure no data is lost. In practice this is almost always a battery backup system, usually with redundant batteries. — Preceding unsigned comment added by Sciencia61 ( talk • contribs) 14:07, 19 June 2018 (UTC)
Footnote 82 is 404 (broken link) — Preceding unsigned comment added by 2607:FEA8:560:3F6:C534:53BD:159A:817C ( talk) 14:47, 18 January 2018 (UTC)
This write-up is completely missing the transition from disk controller level RAID to computer system level RAID in which computer systems are virtualized as disks and entire computer system failure data was masked via fail-over.
SeaChange International invented and patented a system level RAID in 1995. Patent number #5,862,312.
By 1999 Hewlett Packard and Larry Elison's nCuBE start-up successfully challanged the SeaChange patent's "network switch" claim in Delaware superior court, which led to wide use of the patent in computer cluster systems which is often called RAIN for Redundant Array of Independent Nodes. For instance, this is the original design idea of the first Isilon Systems product launched in 2001 accdoring to Wikipedia.
The Wiki articles on RAIN, RAIS (systems) etc. also do not document the history of this virtualization from disks to anything replicatable that needs redundancy. Today we even have RAIN overload: Redundant Array of Independent NAND which you can find inside every SSD ! — Preceding unsigned comment added by SeaChanger ( talk • contribs) 20:38, 19 January 2019 (UTC)
Shouldn't the following: "Consequently, using RAID for consumer-marketed drives can be risky, and so-called "enterprise class" drives limit this error recovery time to reduce risk...."
be changed to:
"Consequently, using consumer-marketed drives for RAID can be risky, and so-called "enterprise class" drives limit this error recovery time to reduce risk...."
?
-- 176.20.208.76 ( talk) 13:40, 22 February 2019 (UTC)
![]() | This is an archive of past discussions. Do not edit the contents of this page. If you wish to start a new discussion or revive an old one, please do so on the current talk page. |
Archive 1 | ← | Archive 4 | Archive 5 | Archive 6 | Archive 7 |
The column "minimum number of drives" is just plain incorrect. First of all, it appears to refer to a nondegenerate minimum number of drives which is completely different from the actual minimum number of drives (for example, a 3-disk RAID-1 can be considered a degenerate 3-disk RAID-6; this is important for implementation since it may be useful for reshaping), but second of all, requiring 4 drives for RAID-5 is flat-out incorrect. Hpa ( talk) 02:39, 3 January 2012 (UTC)
Is there a standard for triple-mirroring? This is basically the same thing as a standard two-drive mirror configuration, except the data is written to three drives at once, and can be read from any of the three. Any two drives can fail and no data is lost.
This would be an acceptable alternative to RAID-6 when you have huge 300+ gig hard drives, but only say 100 gig of server data. No parity engine is needed, just write duplication across independent parallel controller channels for maximum performance.
+N Mirror basically means data could be mirrored across however many drives you want, up to the controller's interface limit. Eight redundant drives? Sure, why not.
DMahalko ( talk) 03:27, 24 February 2012 (UTC)
The Array Failure Rate, as defined in the table, lists the probability of the data loss in the given period under assumption that no drives are replaced during this period. This metric, while in itself valid, is not what one would typically expect as the "failre rate". Leaving the dead drive without replacement is against any maintenance practice and/or recommendations. I suppose we should at least add another footnote clarifying this? Alexey V. Gubin ( talk) 19:44, 12 March 2012 (UTC)
According to Patterson and Hennessy 2012 [1], RAID 2 is Error Detecting and Correcting Code, and RAID 3 is Bit-Interleaved Parity. This is different from what is in the article. Espertus ( talk) 00:40, 29 March 2012 (UTC)
If the section concludes, "In short, the choice between RAID 5 and RAID 10 involves a complicated mixture of factors. There is no one-size-fits-all solution, as the choice of one over the other must be dictated by everything from the I/O characteristics of the database, to business risk, to worst case degraded-state throughput, to the number and type of drives present in the array itself. Over the course of the life of a database, one may even see situations where RAID 5 is initially favored, but RAID 10 slowly becomes the better choice, and vice versa.", I don't see how the neutrality of the section can be called into question. — Preceding unsigned comment added by 68.104.214.16 ( talk) 15:39, 30 April 2012 (UTC)
Isn't saying "RAID Array" just like saying "ATM Machine"? Why stop at array? Why not call it "RAID Array of Independent Redundant Discs"?
173.13.189.193 ( talk) 22:15, 17 September 2012 (UTC)
In the history section the claim is made that: "In October 1986, the IBM S/38 announced "checksum" - an operating system software level implementation of what became RAID-5" which is one year before Patterson defined the term RAID. I have done a pretty thorough search for literature evidence for this and can find none- does anyone have a reference. Otherwise I will remove it in the future. 121.45.215.68 ( talk) —Preceding undated comment added 04:35, 2 December 2012 (UTC)
The section "RAID 10 versus RAID 5 in relational databases" has several problems- the main one is that RAID5 is no longer recommended or used for any large data set, presumably including most databases. Could this section be generalised to compare parity versus non-parity (mirrored) RAID schemes? Also, there is a lack of references. 121.45.215.68 ( talk) —Preceding undated comment added 22:44, 3 December 2012 (UTC)
The sentence "a single drive failure results in reduced performance of the entire array until the failed drive has been replaced and the associated data rebuilt" appears twice in the description of RAID 5. Somebody should fix it. 173.34.179.126 ( talk) 14:19, 22 December 2012 (UTC)
I've just made a few edits to improve clarity.
I'm looking for constructive criticism regarding the following ones:
(and feel free to make them for me.
CONFUSING:
"Background scrubbing can be used to detect and recover from UREs (which are latent and invisibly compensated for dynamically by the RAID array) as a background process, by reconstruction from the redundant RAID data and then re-writing and re-mapping to a new sector; and so reduce the risk of double-failures to the RAID system[42][43] (see Data scrubbing above)."
Idea:
Add a subheading to this section and the one above what's CONFUSING
and
REWORD THUS:
Double....
or perhaps
Section on Atomicity: Wow, this is confusingly written. Perhaps it should kick off with the simple statement, like "Databases are designed to maintain data consistency despite the non-atomicity of the write process of the disk drives on which they normally store data, even when there is unexpected power loss." If I'm not mistaken, either disk drives can't perform atomic write processes, period, or ones that can are rare enough to merit little more than a parenthetical.
Write case reliability:
P(aragraph)3 doesn't mesh with P1 and P2. In fact all 3 are about write-back, but only P3 calls it that. Merge 'em.
Hardware Labeling:
Some systems have a feature intended to reduce operator error and facilitate drive failure management by having an LED by a failed drive indicate that it is a drive that has failed.
History:
REMOVE:With S/38 checksum, when a disk failed, the system stopped and was powered off. Under maintenance, the bad disk was replaced and the new disk was fully recovered using RAID parity bits. While checksum had 10%-30% overhead and was not concurrent recovery, non-concurrent recovery was still a far better solution than a reload of the entire system. With 30% overhead and the then high expense of extra disk, few customers implemented checksum.[citation needed]
doesn't make sense.
It's so confusingly written I don't know what it's trying to say or how to fix it.
-- Elvey ( talk) 11:42, 30 December 2012 (UTC)
I have reverted two edits by 145.225.17.100 ( talk · contribs) ( http://en.wikipedia.org/?title=RAID&diff=523846804&oldid=523363466 and http://en.wikipedia.org/?title=RAID&diff=523846913&oldid=523846804 )due to the fact they removed the entity of the Standard Raids section, removed part of the cite in the nested/hybrid section, and added no content. I'm going to assume this is either vandalism or accidental, though the IP does have other changes elsewhere that look less suspicious. Anything needed other than undoing the edit? Nazzy ( talk) 15:58, 19 November 2012 (UTC)
The article claims RAID 3 is a 'theoretical' RAID level and 'not used in practice.' This is incorrect, EMC Clariion and VNX support a RAID 3 option. — Preceding unsigned comment added by 98.229.192.30 ( talk) 21:28, 14 December 2012 (UTC)
121.45.193.118 ( talk) 12:09, 20 December 2012 (UTC)
You can see the RAID 3 reference, as well as 0, 1, 1/0, 5, and 5 in the current (1/7/2013) datasheet for EMC VNX on their website. http://www.emc.com/collateral/hardware/data-sheets/h8520-vnx-family-ds.pdf — Preceding unsigned comment added by 168.159.213.60 ( talk) 20:02, 7 January 2013 (UTC)
The section ==Mirroring versus parity RAID levels in relational databases== is very problematic- e.g. the claim that "Given the rare nature of drive failures in general, and the exceedingly low probability of multiple concurrent drive failures occurring within the same RAID, the choice of RAID 5 over RAID 10 often comes down to the preference of the storage administrator" is incorrect-- UREs are extremely likely to happen with large disks and thus RAID 5 is not recommended for large databases by the major storage manufacturers (as noted in the main article). The references given are to a non-authoritative web site and do not support the claims in any case. A well-referenced section on this topic could be useful, but the current section is misleading to users of RAID. The current text is copied below for improvement before possibly returning to the main article. 121.45.213.224 ( talk) 06:54, 28 January 2013 (UTC)
It seems that RAID 10 and RAID 1 should have the same space efficiency formula, namely 50% since they both only use mirroring to tolerate single drive failures. — Preceding unsigned comment added by 99.245.3.15 ( talk) 19:40, 1 February 2013 (UTC)
What you’re referring to as Linux, is in fact, GNU/Linux, or as I’ve recently taken to calling it, GNU plus Linux. Linux is not an operating system unto itself, but rather another free component of a fully functioning GNU system made useful by the GNU corelibs, shell utilities and vital system components comprising a full OS as defined by POSIX.
Many computer users run a modified version of the GNU system every day, without realizing it. Through a peculiar turn of events, the version of GNU which is widely used today is often called “Linux”, and many of its users are not aware that it is basically the GNU system, developed by the GNU Project. There really is a Linux, and these people are using it, but it is just a part of the system they use.
Linux is the kernel: the program in the system that allocates the machine’s resources to the other programs that you run. The kernel is an essential part of an operating system, but useless by itself; it can only function in the context of a complete operating system. Linux is normally used in combination with the GNU operating system: the whole system is basically GNU with Linux added, or GNU/Linux. All the so-called “Linux” distributions are really distributions of GNU/Linux. — Preceding unsigned comment added by 175.100.80.20 ( talk) 10:33, 4 February 2013 (UTC)
Hello! I'm concerned the table might be slightly wrong. Specifically, under column "Read performance" for level RAID1, it is stated "nX", which supposes the performance will scale as read is split among the drives (same process as RAID0). I believe this is not accurate.
The article itself states, under the RAID1 section: "the read request is serviced by either of the two drives containing the requested data".
So which one is it? Does RAID1 allow split reading on numerous drives, or can only a single drive in the array be read? From what I know, and from what I found searching, this seems to be entirely dependent on the RAID controller being used. Findings in this blog is that RAID1 read speeds are either slightly lower or equal to the read speeds of a single drive (that was either both a Windows Software RAID1 and an Intel ICH5R integrated Motherboard (aka Fake RAID) controller). I wasn't able to find actual proof that some controllers do in fact support split reading, but some people do mention that certain OS are able to do it. Quantos88 ( talk) 16:07, 29 March 2013 (UTC)
The figures used in the "Comparison" table (currently section 2.9 as I write this) are very useful, but only to people who already understand the technology. For anybody else -- say, somebody trying to use this page as if it were out of an encyclopedia -- the lack of explanatory keys makes the right-hand column much less helpful than it could have been.
Specifically, we need a brief explanation of what "A1", "B2", etc, mean in this context. Yes, somebody familiar with RAID knows that these represent storage blocks (or other sizes depending on the RAID level), but there's no obvious "redundancy" in the images. Even for, say, RAID 5, it's not obvious that any of that data has in fact been stored more than once.
The reason I didn't try to add such a sentence is that I wouldn't know how to write it. I had pointed a colleague at this page when they asked for a breakdown of RAID levels, and they came back more confused than they had been at the start. 99.102.75.166 ( talk) 14:43, 15 April 2013 (UTC)
In the main table formulae are given of array failure rates but no citations are given to confirm them, and indeed they have been changed several times such that I have no confidence that they are correct. Explicit references for each of them is needed- I added a "citation needed" previously but this has been removed. 74.73.169.211 ( talk) 16:43, 21 April 2013 (UTC)
Under the table in 'Comparison' it is stated that
**** Raid 10 can only lose 1 drive per span up to the max of 2/n drives
Should this be n/2 drives? Michaelcochez ( talk) 09:02, 15 July 2013 (UTC)
IEEE.org as of 4-6 Dec 1991 and updated as of Aug 2002 cited Inexpensive, [1]
IEEE.org as of June 29 2009-July 2 2009 and updated as of 29 September 2009 cited Independent, [2]
NetApp -Independent [3]
EMC - Independent [4]
Western Digital Independent [5]
A paper from Duke University Dept. of Computer Science in 1993 - Inexpensive [6]
in the section "reliability terms" it says: "System failure is defined as loss of data and its rate will depend on the type of RAID. For RAID 0 this is equal to the logical failure rate, as there is no redundancy." however wouldn't the faliure rate be higher, asuming each drive in the array has a chance of faliure and the data is spread across more than one drive? ~patrick
A common opinion (and one which serves to illustrate the dynamics of proper RAID deployment) is that RAID 10 (a non-parity, mirrored RAID) is inherently better for relational databases than RAID 5, because RAID 5 requires the recalculation and redistribution of parity data on a per-write basis. [1]
There are, however, other considerations which must be taken into account other than simply those regarding performance. RAID 5 and other non-mirror-based arrays offer a lower degree of resiliency than RAID 10 by virtue of RAID 10's mirroring strategy. In a RAID 10, I/O can continue even in spite of multiple drive failures. By comparison, in a RAID 5 array, any failure involving more than one drive renders the array itself unusable by virtue of parity recalculation being impossible to perform. Thus, RAID 10 is frequently favored because it provides the lowest level of risk. [2]
Modern SAN design largely masks any performance hit while a RAID is in a degraded state, by virtue of being able to perform rebuild operations both in-band or out-of-band with respect to existing I/O traffic. Given the rare nature of drive failures in general, and the exceedingly low probability of multiple concurrent drive failures occurring within the same RAID, the choice of RAID 5 over RAID 10 often comes down to the preference of the storage administrator, particularly when weighed against other factors such as cost, throughput requirements, and physical spindle availability. [2] failed verification
The formulas for the Array Failure Rates in the Comparison Table were changed recently (April 2013). Does anyone have a reference for the calculations? Mobileseth ( talk) 03:23, 24 August 2013 (UTC) mobileseth
Biassed in putting SAS/enterprise drives in italics. Reduces the error handling to nothing more than setting TLER. ("How to not let a drive with bad blocks drop out so you need to replace it. Instead keep running with it! Yay!") — Preceding unsigned comment added by 2001:A60:11FA:6701:993E:921B:42E4:8DFE ( talk) 19:41, 2 December 2013 (UTC)
It doesn't even mention stuff like 520byte vs 512byte formatting or that normally drives in a hw raid get updated firmware that was tested for reliability. I guess next week someone will update it showing that only ZFS knows how to "check" a disk. — Preceding unsigned comment added by 2001:A60:11FA:6701:993E:921B:42E4:8DFE ( talk) 19:08, 2 December 2013 (UTC)
![]() | This
edit request has been answered. Set the |answered= or |ans= parameter to no to reactivate your request. |
Please change 2/n to 1/2 in (1) the "Space efficiency" column of the Raid 10; (2) "Raid10 can only lose one drive per span up to the max of 2/n drives" Oyang2010 ( talk) 19:11, 3 April 2014 (UTC)
The table claims that the write performance for RAID levels 4,5, and 6 are (n-1)x, (n-1)x, and (n-2)x, respectively. presumably n is the number of disks in the set and x is the write performance in IOPS of one such disk.
These claims are contrary to industry norms, which are x, nx/4, nx/6, respectively in the above list.
The article should provide a reference for the claims or they should be removed. — Preceding unsigned comment added by 168.159.213.208 ( talk) 17:31, 16 April 2014 (UTC)
I undid Dsimic's last edit, where several instances of "data is" were again changed to "data are". Numerous editors keep editing this article to change these back to use the singular verb form, in what can only be described as edit warring. Unlike traditional warring, though, I've observed Dsimic versus the world instead of just one editor.
While I agree with Dsimic that the plural Latin nouns usually end in "a" and single Latin nouns in "um", Wiktionary offers a very insightful a third possibility. Basically, "data" can also be used as an uncountable noun—which is the norm in computing contexts like RAID. Surely that's because individual bits number in the trillions in today's machines and thus counting becomes a vain exercise. Similarly, water is made up of individual molecules but "a glass of waters" is poor grammar. I believe this to be why the vernacular supports the singular form, and why we should stick with it here. – voidxor ( talk | contrib) 05:40, 14 June 2014 (UTC)
A recent edit by Snori describes super-expensive SLED devices as “super-reliable” and the smaller, less expensive disks as “comparability unreliable”. That's not actually true, and I think it skews the description of the basic idea behind RAID.
The leading motiviation to develop RAID was the higher performance which could be realized; associated benefits include larger and more flexible storage capacity, and reductions in equipment cost, size, and power consumption. Single disk reliability wasn't a big factor. Unician ∇ 07:52, 1 September 2014 (UTC)
@ Dsimic:The 3330, actually the 3830 storage control used a Fire code appended to the end of each block for error correction. It was not a Hamming code as used in RAID-2. More significantly, the Hamming code works in real time while the appended Fire Code works after a block has been buffered; so the reference to the 3330 should be deleted, regardless.
I am very certain that the 353 did use a Hamming code so it that sense it was similar to RAID-2. It was actually a Redundant Array of "Independent" Heads, 40 heads transferred in parallel, 32 data, 7 Hamming code and 1 parity. That's why the term "similar approach" is used in the article. I'd have to dig a bit to find a reference if this is disputed, but I am sure i can find one. Tom94022 ( talk) 01:55, 15 January 2015 (UTC)
e.g., see
page 157 of IBM 7030 Data Processing System Reference Manual; it discloses 32 data bits plus 7 ECC bits in the "high speed disk storage unit" which I am pretty sure is the IBM 353.
Tom94022 (
talk)
18:27, 15 January 2015 (UTC)
Dsimic nice find with the Katz quote. I believe I heard him say the same thing at some point but didn't bother to look for it because Strebe mis-states what the article says and then mis-applied WP:SYNTH. The RAID paper defines RAID-1 as disk mirroring. The literature of the far earlier DEC product discloses disk mirroring. It is no violation to then say disk mirroring existed before the term RAID-1 was defined. Ditto for each of the other RAID levels and for the introductory sentence. But that is moot now, thanks to yr great work. Tom94022 ( talk) 07:46, 18 January 2015 (UTC)
I propose that Standard RAID levels be merged into RAID. I think that the content in the Standard RAID levels article can easily be explained in the context of RAID, if it is not already. I cannot justify the duplication of effort when it comes to listing RAID levels. One example is the recent addition of a table of levels to the Standard RAID levels article, when a much more established table of differences already exists in the RAID article. Listing basic RAID levels in two places is redundant, so I removed it. I felt bad when doing so because an editor worked hard on the new table, having not seen the existing one. It's easy to see how one would land on Standard RAID levels and assume they are on Wikipedia's one-and-only RAID article. – voidxor ( talk | contrib) 06:49, 23 January 2014 (UTC)
Hello, 67.6.185.65! As I've noted in my edit summary, RAID 10 is a nested (hybrid) level, not one of the standard RAID levels. Thus, it is wrong to add RAID 10 into the RAID § Standard levels section; please see RAID § Nested (hybrid) RAID section for further information. — Dsimic ( talk | contribs) 20:01, 14 April 2015 (UTC)
In the summary of RAID 4, it is contrasted with RAID 2 and 3. A quote "As a result, more I/O operations can be executed in parallel, improving the performance of small transfers.[2]" This is only true for reads. It is untrue for writes. RAID 4 serializes writes and can perform only 1 at a time. This is because all writes, by definition, must update their respective parity data and in RAID 4 this parity resides on a single spindle. RAID 4 does one write at a time, whether small write or full stripe, exactly like RAID 2 and 3. Worse it does so without their bandwidth advantages.
Further the reference to NetApp's RAID-DP is misleading. It's not RAID-4. Its closer to RAID 6 since each stripe has two parity blocks and one is diagonally formed.
Next, the key advantage of RAID 5 is that its parity is distributed over all the members of the raid string. Unlike RAID 4, a write is not serialized by a single spindle. On a RAID 5 array the number of simultaneous writes can be as a high as 1/3 of the number of members in the string, e.g. a 15 drive string could have 5 simultaneous writes occurring. A RAID 4 string can never have more than 1 regardless of its size. The article leads you to believe RAID 4 has some advantage, when it's all the other way.
Lastly, all of the negative points you make about RAID 5 are true for RAIDs 1, 3, and 4 as well. A read failure during rebuild or degraded operation (one drive failed) can happen for any of them and would be equally unrecoverable. You don't point any of that out.
The comments on RAID 5 have a 'sky is falling' tone and lack the above perspectives. — Preceding unsigned comment added by 73.16.141.73 ( talk) 00:29, 7 December 2015 (UTC)
Could add something about assumptions: who is expected to benefit from this? No mention of Windows 7 Pro. User trying to do mirroring (RAID 1) having come across it in the create and format hard disk partitions section of control panel. Well the "system crash" section in this Wiki may be a warning to the newcomer, but Dsimic has labelled my point about power interruption as pretty much a nonsense. Is Dsimic suggesting the assumption that a UPS is what everyone would already have before thinking about RAID? Where would they have come across that knowledge? On a non-mirrored system a power failure is overcome easily at the next boot. It does not require a message sent to Microsoft &c, the way a "crash" may. Soundhill ( talk) 12:00, 19 January 2016 (UTC) Soundhill
The article says:
> The array will continue to operate so long as at least one member drive is operational.
There is no description of the failure scenario.
For people that are new to RAID, we could use a more complete description of what happens when a drive fails in RAID 1. For instance, is there a light or alarm, etc? Just another line or two would complete the description. Consider that a RAID 1 enclosure for two drives is the least expensive and most marketed RAID enclosure, so the Wikipedia description of RAID 1 is going to bring a lot of views. Thanks 75.110.98.103 ( talk) 23:38, 4 February 2016 (UTC)
Hello, A876! Regarding your edit comment, there are three things:
|website=
parameters (where applicable, of course).With all that in mind, I've restored the references and improved the wording a bit further. Hope you agree. — Dsimic ( talk | contribs) 22:30, 20 April 2015 (UTC)
Any chance of mentioning the 'old' name for RAID; which was "Redundant Array of Inexpensive Disks"?
203.214.22.116 (
talk)
01:13, 14 March 2017 (UTC)
Oops! I just found it... Sorry. 01:15, 14 March 2017 (UTC) — Preceding
unsigned comment added by
203.214.22.116 (
talk)
I just dumped the citation needed for "Needs 4 discs". There's a good explanation on the Standard_RAID_levels page, but the summary of a 3 disc system would be: Disc 1: Data Disc 2: XOR of all the data blocks. With one disc that's just a copy of the data. Disc 3: Clever correction based on the data. That need be no more than a copy. (then rotates around the discs for the other data blocks) That's just a 3-way mirror. — Preceding unsigned comment added by Number774 ( talk • contribs) 14:38, 5 July 2017 (UTC)
The article states RAID 1 offers no parity, but a mirror is an even parity. I think it adds no value to explain this to the reader, but maybe it can be corrected? — Preceding unsigned comment added by 145.53.72.151 ( talk) 18:54, 19 January 2016 (UTC)
Actually the above comment is wrong on several counts. First a RAID 1 array could have more than 2 members. HP OPENVMS provides this, up to 8 in fact in what they call a shadowset. With more than 2 disks you can catch inconsistent copies. Further it implies that parity systems check parity on reads, they don't. This would require reading the entire stripe, which is too much overhead. With RAID-5 this would be useless as you wouldn't know which version was correct. With RAID-6 some implementations scrub parity in the background, since there are three sources for each block, the block itself, first parity, and second parity. This allows for a voting mechanism. Uncommon RAID systems such as RAID-3 might actually check parity on each read but it is an academic point, there are very few if any commercial versions of them.
As well the article claims RAID-1 read performance is "slower than the fastest drive". In virtually all implementations a read is queued to the disk member with the shortest operations queue. In practice RAID-1 sets give superior read response times and total read rates as compared to single drives. — Preceding unsigned comment added by Sciencia61 ( talk • contribs) 12:35, 19 June 2018 (UTC)
You can see the original paper here: http://www.cs.cmu.edu/~garth/RAIDpaper/Patterson88.pdf RAID was originally a cost reduction method, to allow for the use of cheaper disks than were sold by the then large vendors (IBM and DEC). Parity was added to make up for the perceived unreliability of the cheaper disks. Later when RAID was commercialized, and made expensive by the likes of EMC and NETAPP, marketing efforts to justify the now high prices required the change from inexpensive to independent. You can read one of the original authors (Katz) commenting on the very issue in the last paragraphs of this paper: https://web.eecs.umich.edu/~michjc/eecs584/Papers/katz-2010.pdf — Preceding unsigned comment added by Sciencia61 ( talk • contribs) 12:53, 19 June 2018 (UTC)
I added a further reference from the SNIA document explaining that while write caches had a power fail vulnerability, good implementations design for this and add methods to ensure no data is lost. In practice this is almost always a battery backup system, usually with redundant batteries. — Preceding unsigned comment added by Sciencia61 ( talk • contribs) 14:07, 19 June 2018 (UTC)
Footnote 82 is 404 (broken link) — Preceding unsigned comment added by 2607:FEA8:560:3F6:C534:53BD:159A:817C ( talk) 14:47, 18 January 2018 (UTC)
This write-up is completely missing the transition from disk controller level RAID to computer system level RAID in which computer systems are virtualized as disks and entire computer system failure data was masked via fail-over.
SeaChange International invented and patented a system level RAID in 1995. Patent number #5,862,312.
By 1999 Hewlett Packard and Larry Elison's nCuBE start-up successfully challanged the SeaChange patent's "network switch" claim in Delaware superior court, which led to wide use of the patent in computer cluster systems which is often called RAIN for Redundant Array of Independent Nodes. For instance, this is the original design idea of the first Isilon Systems product launched in 2001 accdoring to Wikipedia.
The Wiki articles on RAIN, RAIS (systems) etc. also do not document the history of this virtualization from disks to anything replicatable that needs redundancy. Today we even have RAIN overload: Redundant Array of Independent NAND which you can find inside every SSD ! — Preceding unsigned comment added by SeaChanger ( talk • contribs) 20:38, 19 January 2019 (UTC)
Shouldn't the following: "Consequently, using RAID for consumer-marketed drives can be risky, and so-called "enterprise class" drives limit this error recovery time to reduce risk...."
be changed to:
"Consequently, using consumer-marketed drives for RAID can be risky, and so-called "enterprise class" drives limit this error recovery time to reduce risk...."
?
-- 176.20.208.76 ( talk) 13:40, 22 February 2019 (UTC)