1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Storage SSD's in RAID 0 = AWESOME

Discussion in 'Hardware' started by meandmymouth, 10 Sep 2010.

  1. meandmymouth

    meandmymouth Multimodder

    Joined:
    15 Sep 2009
    Posts:
    4,263
    Likes Received:
    314
  2. favst89

    favst89 What's a Dremel?

    Joined:
    23 Jul 2010
    Posts:
    390
    Likes Received:
    13
    The statistics look impressive and I suppose it could be cheaper than buying a single larger/faster drive.
    However, I thought most ssd's lost the important TRIM command when put into a RAID array?
     
  3. meandmymouth

    meandmymouth Multimodder

    Joined:
    15 Sep 2009
    Posts:
    4,263
    Likes Received:
    314
    ah, maybe, I don't know

    Can someone else help here?
     
  4. SlowMotionSuicide

    SlowMotionSuicide Come Hell or High Water

    Joined:
    16 May 2009
    Posts:
    835
    Likes Received:
    20
    They do. TRIM won't get past the RAID controller, unfortunately.
     
  5. TheBlackSwordsMan

    TheBlackSwordsMan Over the Hills and Far Away

    Joined:
    16 Aug 2009
    Posts:
    4,102
    Likes Received:
    534
    If the TRIM is unavailable, it is worthless. Who want to kill his SSD in 6 months ?
     
  6. Sloth

    Sloth #yolo #swag

    Joined:
    29 Nov 2006
    Posts:
    5,634
    Likes Received:
    208
    I suppose you could look at it as: who wants to purge and recreate all of his data every 6 months?

    Makes it a little more viable... maybe... no? Okay nevermind. :D
     
  7. PocketDemon

    PocketDemon Modder

    Joined:
    3 Jul 2010
    Posts:
    2,107
    Likes Received:
    139
    What utter rubbish - why on earth would a lack of trim cause a modern SSD to die after 6 months?

    You're effectively saying that it would be "worthless" to use any SSD using a pre Win7 OS (as neither XP nor Vista have it), the 2.6.33 version of Linux & i'm not 100% whether it's available in OSX yet (if so only the last month or so).

    ...so, thinking this through, are you really trying to argue that every SSD used for 6 months prior to 22nd October 2009 (forgetting technet subscribers) on a Windows based machine, the same prior to Feb 2010 on a Linux machine or (possibly - if OSX still doesn't have it) installed 6 months & a day ago on a Mac is now filling up landfill somewhere having all been RMA'd???

    i didn't think so as the manufacturers are still surviving & offering those 3 year warrantys...


    You have to realise that most/all (i've a sneaking suspicion that some intel ones may not have done - though lots of the earlier ones had it added in f/w updates) of the Gen2 SSDs have some version or other of garbage collection (some better than others) which means that trim, whilst having some advantages on some drives over GC, is not at necessary on them...

    There's also been manual tools that individual manufacturers have created & things like AS-Clean can be used on most drives (not recommended for the SF's apparently though d.t. it's ver of GC) whether individually or in arrays.

    With the SF based SSDs especially (they apparently have some magical version of GC that's not GC), certainly in the beta testing, OCZ found no advantage to having trim turned on or off...


    & even this isn't necessary...

    Well, you will find people who will image, erase (not format as this does nothing to alter speeds) & redo their arrays, but have never had any need at all.



    [edit]

    oh, & yes, running several SSDs in a R0 array is pretty snazzy - did it originally with a pair of 120GB V Turbos & now do it with 4x V2s (on a card that can handle it of course)
     
    Last edited: 10 Sep 2010
  8. DragunovHUN

    DragunovHUN Modder

    Joined:
    30 Oct 2008
    Posts:
    5,149
    Likes Received:
    181
    Anyone remember this?
     
    mrbens likes this.
  9. Rofl_Waffle

    Rofl_Waffle What's a Dremel?

    Joined:
    24 Mar 2010
    Posts:
    504
    Likes Received:
    12
    SSDs can't get damaged with or without trim.

    Trim just reduces junk data on the drives. Junk data ONLY slows write cycles, not read cycles.

    A degraded SSD can be fixed by turning on TRIM to remove junk data or you can have it wiped by writing all its sectors to binary zeros.
     
  10. Syn

    Syn ¯\(°_o)/¯

    Joined:
    4 Aug 2010
    Posts:
    131
    Likes Received:
    2
    I wouldn't bother with RAID0 for SSD's, they're fast enough, and I would want TRIM support.

    Also, I don't really see the need to buy an 80gb for £175 when you can get the 60gb version of the same drive for £110. 60GB is more than enough for a boot drive for me, although I'm sure some of you might disagree :)
     
  11. PocketDemon

    PocketDemon Modder

    Joined:
    3 Jul 2010
    Posts:
    2,107
    Likes Received:
    139
    No - you write 1's on SSDs to wipe them not 0's... if you did the latter you'd make every free cell (other than OP) slow until GC &/or Trim kicked in... Hence ticking FF option with AS clean...

    Though, having said that, doing something like that with a SF drive messes up the GC & wear leveling on them so (again) it's not advisable.

    Well, since it's either the same cost or cheaper to raid them than buy a large one then why not have the extra speed...? You wouldn't deliberately buy a slower SSD would you?

    & (simply someone disagreeing), whilst a fairly basic internet/office installation can happily cope with <40-60GB, once you start adding in a whole set of pro apps for multimedia editing/creation & web design & whatnot a few games & things then 60GB is tiny...
     
    Last edited: 10 Sep 2010
  12. Shaftydude

    Shaftydude Minimodder

    Joined:
    26 Jul 2010
    Posts:
    158
    Likes Received:
    5
    Looking at the PCMark Vantage results, I don't think its worth the performance increase to go raid when the single Corsair SSD gave amazing performance already.
     
  13. thehippoz

    thehippoz What's a Dremel?

    Joined:
    19 Dec 2008
    Posts:
    5,780
    Likes Received:
    174
    :eek:
     
  14. PocketDemon

    PocketDemon Modder

    Joined:
    3 Jul 2010
    Posts:
    2,107
    Likes Received:
    139
    Just a 'quick' (actually very lengthy having finished) addition having actually managed to track down a pre-Win7 link i couldn't find earlier that shows how trim is not necessary...

    ...but, before getting onto that, a quick thing about mis-information about this subject which 'may' be why people have the wrong idea about the need for trim.


    Now, the problem is that some sites have, to a greater or lesser extent depending upon the SSD, made some unjustified claims to the world about stuff -

    for example here they are stating that a f/w with trim is better than one without BUT, are completely ignoring the effect of improvements in GC - attributing anything & everything to trim...

    "The drive was then left for an hour to ensure the TRIM command and any garbage collection algorithms had been completed before being retested using our new suite of benchmarks."

    - (my emphasis added in the passage above) does not equate to what follows -

    "clearly TRIM works almost perfectly in restoring performance to that of the clean drive."

    All that can actually be said based on the reviewer's results is that a combination of adding trim &/or improvements in GC between f/w revisions have had a positive effect such that speeds could be more fully recovered in ~1hr.


    Now, taking the OCZ indilinx drives as an easy example (mainly as i know more about their history & the link i've now recalled also uses them) there's 2 key points to note -

    1. Firstly, the testing method used in the bittech article is artificially extreme, bearing no resemblance to a normal owner's usage - who completely fills & then deletes an OS/apps drive 10x with random data?

    On a more normally 'dirty' SSD with the older 1.3 firmware referred to (when GC was first added but before trim) these are the results.

    Now, obviously i accept that these are clarified within the article as initial testing results but they do clearly show that, even with a first version of GC & a more realistic dirty state, GC will restore speeds independently of the availability of trim...

    ...but 'if' you deliberately slow down with an early GC f/w in the manner used by bitech then 1hr is not enough time to fully recover.


    2. Secondly, the 'aggressiveness' of GC was increased on those OCZ drives from 1.3 onwards - ie it more actively runs in subsequent revisions leading to quicker recovery whether an OS is 'trim enabled' or not or you're running single SSDs or 2+ in a raid array.

    This, therefore, makes it becomes even more foolish to attribute the improvement between the 1.3 & 1.5 f/w versions solely to the inclusion of trim.

    [the facts in both 1 & 2 backup my own r.l. testing results that my OCZ indilinx SSDs with GC worked perfectly long term (~10 months) in a R0 array without slowing down]


    Now, let's assume that trim did actually play a part in regaining speeds in bittech's testing (there's no reason to exclude this of course) - this still does not mean that speeds will suffer long term in a non-trim OS or in a raid array with normal usage...

    ...simply that if you act like a [choose your own expletive] & fill & delete your SSDs 10x (or even once) with either random data (or with 0's as i noticed in another bittech article claiming that this was a 'clean' state to then run dirty comparisons from) just to delete it then you've only got yourself to blame if it takes a longer time to regain speeds than if you use it sensibly.


    Obviously, once again, these 2 points are based around the indilinx OCZ drives &, as i said in an earlier post, some Gen2 SSDs will have better GC than others...

    ...however, since the newer SSDs on the market (ie what anyone on here would sensibly buy now) 'should' all have a form of GC (& all of them again 'should' be better than the early versions), it's clearly not the case that a lack of trim for whatever reason will be disastrous.


    Perhaps the main conclusion would be that, if you're looking to buy a SSD for a non-trim OS (or raiding some) then making sure it/they have a decent method of GC would be advisable - well, the C300 is much slower at regaining speeds (ie has a less aggressive GC) than either a SF or indilinx SSD...

    "Compared to what Indilinx and SandForce have done, Crucial’s implementation just doesn’t cut it. For Windows 7 users running a single drive this shouldn't matter since you have TRIM. But if you're running RAID or another OS, the C300 isn't as desirable."
     
  15. ShakeyJake

    ShakeyJake My name is actually 'Jack'.

    Joined:
    5 May 2009
    Posts:
    915
    Likes Received:
    70
    Really? hdparm showed around 85 mb/s on a read test from me Vertex Turbo, which I how I could tell I hadn't enabled TRIM properly. After going to 2.6.33 it shows 195 mb/s read.
     
  16. Syn

    Syn ¯\(°_o)/¯

    Joined:
    4 Aug 2010
    Posts:
    131
    Likes Received:
    2
    I wouldn't deliberatly by a slower SSD, and I would want the SSD to retain its speed, so personaly I would choose TRIM over RAID0, but thats just preference I guess.

    And I find it hard to understand why everybody feels the need to install the entire Adobe CS5 suite when all the majority of them use is photoshop. I mean yeah some people will be installing a fair few programs, I personally have a lot of games, a few media programs, and some development applications aswell, but installing an entire 'pro multimedia suite' is just pointless for the vast majority of people.
     
  17. Bakes

    Bakes What's a Dremel?

    Joined:
    4 Jun 2010
    Posts:
    886
    Likes Received:
    17
    Well, apart from the debatable loss from not using trim (I'm not arguing, just stating fact - ie that you're debating it here) increasing the number of members in an array increases the chances of future array failure and thus data loss.

    For example, if you have 2 disks with a failure rate of 2% per year, the failure rate is actually about 3.96% - ie you have a 3.96% chance of losing all of your data - significantly bigger than the original 2%.
     
  18. mrbens

    mrbens What's a Dremel?

    Joined:
    15 Aug 2009
    Posts:
    511
    Likes Received:
    4
    That's brilliant! Not seen that before :)
     
  19. PocketDemon

    PocketDemon Modder

    Joined:
    3 Jul 2010
    Posts:
    2,107
    Likes Received:
    139
    No problem at all about having a debate - well, i've fully accepted that trim is more beneficial for some drives than others (pointing fingers at the C300)... it just makes me cross when, supposedly respected sites, make random claims whilst ignoring what's actually happened...

    Okay, when GC first came in then it was treated with suspicion in some quarters, but... Well, i remember Anandtech (who's normally really good) being one of those sites that was 'a little' circumspect with the premise when Samsung first introduced it but, whilst the guy still sees it as being a workaround to a problem that should have been foreseen, accepts that it does work - & the position changed (thanks to actual results & improved/more aggressive GC in later f/ws) prior to things like the foolish bittech article.


    Anyway, to look at your actual point - i don't dispute at all that a R0 array will effectively multiply the chances of failure by the number of drives, though the chances of failure within a 3 year warranty period are obviously low enough for the SSD manufacturers to not expect to have to replace a significant %age of drives of a particular model in that time frame.

    Of course, if a particular drive proved esp prone to failure & it couldn't be corrected with a f/w update (although, the v0002 f/w, which was supposed to correct a whole set of performance & reliability issues, was a really shonky early C300 one that actually caused significant numbers of people to end up with bricked drives - just an example of how things can fall apart), then obviously this would cause both any users & the manufacturer major difficulties...

    ...however that's the chance you take with buying anything - not just SSDs.


    Okay, so it comes down to personal choice, but whether you're using single drives or arrays of any kind (not just R0 ones), that you *should* be maintaining a decent backup regime of critical data - &, for convenience, why not backup the entire OS/apps/etc installation as well for speed of recovery 'if' there were a catastrophic failure...


    As such, you're taking what, in a normal situation would be a minute risk (of single drive failure) &, whilst increasing it, not suddenly making it incredibly likely to happen by using a R0 array - &, without backups, a failure would leave you losing stuff in either case.


    Oh, & it's just crossed my mind that i also pointed out in a previous thread that at least the faster SSDs have to use a form of internal raid anyway in order to get the speeds they do...

    it 'could' therefore be argued that 'if' the use of raid is of such major concern d.t. the increased risk of failure and/or the extra speed from R0 is completely unjustifiable to someone (not your particular argument but it's been mentioned), they should therefore seek out a much slower SSD that both stays within its ONFi spec & doesn't exceed the speed for the specific nand as the only way to (possibly) avoid 'the evils of R0'...

    ...which i think rules out all of the SSDs that we ever discuss as potential purchases on here...
     
    Last edited: 11 Sep 2010
  20. Syn

    Syn ¯\(°_o)/¯

    Joined:
    4 Aug 2010
    Posts:
    131
    Likes Received:
    2
    aint seen that in a while, i was a little disapointed to watch them defrag those drives though D:
     

Share This Page