Question For Nanofrog : raid 6 vs. 60

Discussion in 'Mac Pro' started by JulianBoolean, Oct 8, 2010.

  1. JulianBoolean macrumors regular

    Joined:
    Aug 14, 2010
    #1
    Okay, so... To make a raid 60, you make a raid 6 first then stripe that. If you started with a 4 disk raid 6 array and increased your number of disks to 8 when you stripe it, I have no trouble visualizing performance and capacity advantages.

    However, I'm having trouble visualizing trade offs, speed or capacity differnces if you compared a 8 drive raid 6 vs an 8 drive raid 60.

    If you are not Nanofrog, but you did stay at a holiday inn last night, you are welcome to reply as well.

    Thanks!
     
  2. goMac macrumors 603

    Joined:
    Apr 15, 2004
    #2
    According to Wikipedia, the reason you'd do this is less downtime in case of a failure. If you have a drive fail, you only need to regenerate a 4 drive RAID instead of an 8 drive RAID.

    I'd assume it's faster because there is less overhead in managing a 4 disk RAID as opposed to an 8 disk RAID.
     
  3. JulianBoolean thread starter macrumors regular

    Joined:
    Aug 14, 2010
    #3
    I think you misunderstand my question a bit? In the scenario I'm inquiring about both arrays (RAID 6, and RAID 60) have 8 disks.
     
  4. Honumaui macrumors 6502a

    Joined:
    Apr 18, 2008
    #4
    not him but pretty good on this stuff ;)

    the raid 60 would be slower and less capacity of course than the 6 given the same number of discs the way I learned raid 60 and have used something like it in the past is with multiple controllers :)

    I think if you are going to do raid 6 its a 8 disc min thing ! or maybe 7 min !

    if I was to do 60 it would be to bring two controllers together

    we had a 3 node cluster that used to bring a bunch of raids together using something like this in a form of 2 raid cards on each one and all together for 6 cards but this was for a database that was getting hit big time


    so raid 60 can be good to bring two cards together and spread the I/O and load basically over them both and/or at times to combine two machines together also again one can go down and the other can keep working

    I would never bother with a raid 60 on the same card myself ;)

    so I guess to me the tradeoff is really for super high I/O things like large database etc.. and also to bring together larger clusters to handle the extra load


    I used to be the head guy for a company we did work for and we had a 3 node database server with a 2 node part to handle the main input

    the 3 node was setup kinda like a raid 60 in that the cards were used to share the load and redundancy and the room was needed as we had a million members of resumes and about 500 companies accessing it all the time

    you could bring a machine down and work on it while keeping the other ones going so think of it almost like a huge raid array but instead of pulling discs you could pull complete raid sets !
    pretty cool stuff :)
     
  5. deconstruct60 macrumors 604

    Joined:
    Mar 10, 2009
    #5
    ...needs rewrite.. long day.

    ...
    Generally the same effect in speed up from RAID 6 to 60 as there is in making the RAID 6 stripe set wider.

    RAID 6 with 8 disks uses a stripe set of 6 and two parity.

    RAID 60 with 8 disks uses 2 strips sets of 2 and four parity disks.
    ( have striped over two RAID 6 with four disks : stripe set of 2 and two parity disks )

    You are sinking more into parity drive storage space with the RAID 60 set up.
    While the usable capacity is 6 * drive size versus 2 * ( 2 * drive size )

    Single drive controller isn't a speed hit if it can juggle 6 concurrent READ/WRITE requests at a time. Either getting reading 3 writing 1 blocks concurrent striped blocks along with two parity blocks in first case. In second case could be writing 1 and two parity blocks in each of the 'halves' concurrently.

    If it can't get that much concurrency out of a single controller then there is a natural controller split if can use one each to drive the "separate" halfs of the RAID 60 array ( driver of the two would need to know how to split workload. )

    depends how heavy write workload is.

    Rebuild on the 6 way RAID 6 is going to be longer because more disks to read to construct with parity. The RAID 60 has fewer disks in a stripe set (smaller number to read) and can also rebuild in parallel if had problems in both halves and not at all in the other half it didn't also have a problem .
     
  6. JulianBoolean thread starter macrumors regular

    Joined:
    Aug 14, 2010
    #6
    Another bit of misunderstanding here. I think I do mean stripe not mirror.

    Here is how Nano explains the nested levels in previous thread :

    [..]...............

    50 = Create 2x level 5 arrays, then stripe both of those together. This results in a 5 + 0 configuration, aka 50.

    60 = Same process, but you create 2x level 6 arrays, then stripe them together.

    10 is also the same. Start with 2x level 1 arrays, then stripe the pair.

    http://forums.macrumors.com/showthread.php?t=1026605

    [..]...............

    I think folks are reading my opening post a bit too quickly, or perhaps I'm not being clear? Let's try this another way.

    1. Bob has eight brand new disks from owc and an external pm box to put them in. Bob makes a raid 60. (Bob makes two sets of raid 6 then stripes them together.

    2. Joe also has eight brand new disks from owc and a brand new external pm to put them in. Joe makes a raid 6 out of those eight.

    3. Joe and Bob started with, and end with eight drives each.

    4. Joe and Bob both save an identical 8GB photoshop file to their respective arrays at the same time.

    Question : Who is done saving first?

    5. Joe and Bob both close and reopen the same photoshop file.

    Question : Who will have the file opened faster?

    Hope this helps maybe understand my question a bit better.

    A Mirror of RAID 6 is easy for me to visualize, A Stripe is what I'm having a problem with just doing a thought experiment in my head. It's kinda cluttered in there.

    :)

    Thanks!
     
  7. alphaod macrumors Core

    alphaod

    Joined:
    Feb 9, 2008
    Location:
    NYC
    #7
    RAID 60 will be slightly faster than RAID 6, but it also requires twice as many redundant drives. RAID 6 is n-2 drives capacity, so RAID 60 is n-4 drives capacity. In your case, if you're going with only 8 drives, and you are considering RAID 60, you might as well just go for a RAID 10; that way you don't have to account to parity calculations and the issues that come with that. Of course RAID 6 and RAID 60 both support online expansion.

    Performance wise, the throughput will increase as you add drives to the array. RAID 60 will just increase more, with just the initial penalty of requiring more drives to initialize the array.

    For both read and write, RAID 60 will be faster than RAID 6. RAID 6 does not mirror at all. It's essentially striping just with added parity which is why there is a performance hit (more data to write).

    RAID 6: Think of parts A,B,C,D,E,F for 8 drives and X,Y be parity. A1 (A subscript 1) will write to disk 1, B1 to disk 2, etc. and Parity X and Y will write to the last 2 disks. When you write the next data, all the data will shift one disk, so A2 will write to disk 2, B2 to disk 3, etc.

    And your question is phrased rather incoherently, which is probably why people are having a hard time answering it; and if you don't want anyone but nanofrog to give you a response, you are better served sending a PM.
     
  8. deconstruct60 macrumors 604

    Joined:
    Mar 10, 2009
    #8
    Sorry I didn't clean that up before you replied.


    As pointed out this is really a question of how much parallel throughput can put throw a controller and/or whether can lash multiple controllers to drive more concurrent channels.

    Conceptually, if you can drive both the RAID 60's "RAID 6 halves" at full speed it will finish first. Since more parity disks sets with different data you can write to them faster if can drive them in parallel.

    If writing to four disks at the same time is too much then won't be as much.

    This is a myopic test because nobody spends 90% of their time saving photoshop files.



    Again if haven't choked the controller(s) ; the 6 way RAID 6. Larger stripe sets generally get you faster reads.
    If choked and the RAID60 controller starts prefetching farther out in sequence of logical blocks manages to trigger earlier invocation of other controller path then closer between the two. But fewer disks means competing requests so can't invoke as much parallelism.
     
  9. Honumaui macrumors 6502a

    Joined:
    Apr 18, 2008
    #9
    this is part of what I wrote :)


    the raid 60 would be slower and less capacity of course than the 6 given the same number of discs IN YOUR SITUATION especially


    you are thinking cause its got a 0 in it its going to be faster ? :) or at least I might think you think that :) hehehehe

    lets look at this simple raid 60 is not for small amounts of discs like 8 ? think 16 discs :)

    so you are writing (distributing data) to 4 discs in the raid 60 with 8 discs vs writing to 6 discs in the single Raid 6

    even though you are splitting that and writing one half to one and the other to the other ? well again that single card can only do so much with the info

    in real world 8 discs I bet the raid 60 will be %10 up or down slower ? cause again that controller can split the info but since its splitting it to a lower number of discs its not going to be as fast so that gain of the 0 is lost cause of the amount of discs
    the writes are usually more penalized than the reads ? just to note what I have seen ? :)
    also note !!! in some things like 16 discs say your reads can be a bit quicker %10 up or down (more down I think) and things can be about equal in the writes ! because you are adding more discs so that 0 advantage can start to show but this is also with beefy cards and more memory on them


    again :) the way I learned raid 60 is for more than one controller ? cause again the controller can only do so much
    so unless you can get more discs its a backwards thing :)

    OR I could/should have maybe said earlier when you are dealing with a controller that has 16 or more channels that can open up raid 50 or 60 scenarios for specific things that might need it ?

    I might disagree with this info ? again I could be wrong but I have worked with a lot of raids and what I wrote I think is more correct ? :)
    UNLESS you mean raid 60 is faster with the same amount of discs being used ? meaning a 10 disc raid 60 vs a 8 disc raid 6 ? but since this was not the OP question not sure where you are at :) so if you mean increase the number in 60 then yes I agree but the same number of discs ? not always the case and again writes more equal as discs get over about 12-16 in the array you will see them equal or take a small advantage in read but writes wont usually go up above

    your post made me decide to go get some benchmarks to double check my thoughts :) and to let the OP dig through them

    http://arecaraid.cineraid.com/forum/viewtopic.php?f=7&t=5&sid=ee7d848b3da5045d79eb6b9a2f2b93a3
    as a example and go back and see other benchmarks ? and compare the raid 50 60 and 5 and 6 etc..
     
  10. deconstruct60 macrumors 604

    Joined:
    Mar 10, 2009
    #10
    Depending on your read/write ratio.

    The parity data reads and writes are what kills you, not the calculations.
    Reading and writing is relatively slow...... that is why you are likely using RAID-something in the first place.

    n -4 is still better than n / 2 (of RAID 10) from a capacity perspective. It will be faster because there is one less drive writing to ( two instead of three ) for each block and no funky "double write hole". [presuming not have overly complicated second parity function equation and some decent CPU horsepower in controller, not some grossly underclocked, anemic one. ]


    For the same number of disks the stripe set of RAID 6 can't be the same as RAID 60. If the stripe sets are different why is the read speed the same or better on RAID 60 when it's stripe is smaller ?

    More read/write speed means adding more spindles that store data.

    RAID 60 will fair better in the context of heavy amount of random writes (because more parity spindles). However, for reading it is exactly the opposite effect ( because more data spindles. ) .

    Be careful of benchmarks where they just keep the stripe size approximately constant between RAID 6 and RAID 60 runs ( capacity is tweaked to match them up as close as possible on stripe size. ) those will likely show RAID 60 is faster.
     
  11. Honumaui macrumors 6502a

    Joined:
    Apr 18, 2008
    #11
    just want to make sure the OP also sees this again :) this pretty much boils it down in a nutshell :)
     
  12. deconstruct60 macrumors 604

    Joined:
    Mar 10, 2009
    #12
    I tweaked it after you posted but before I saw this. Only reason dropped the "not parity" is because RAID 6 warps the world view of what is data in the write context because there is so much parity data to get out. Twice as much as the real user data want to store. Put aside RAID 6 and yes forget parity look at the real data.


    Sometimes the parity is the data you need to get out to disk. But generally yes if want speed.... look at the data part and throw spindles (or more modernly stated .... multiple storage units with lowest latency you can afford at them. SSDs too even if they don't spin. ;) )
     
  13. nanofrog macrumors G4

    Joined:
    May 6, 2008
    #13
    JulianBoolean:
    There's plenty of good information already, so I'll keep it as simple as possible.

    Assuming there's no issues (choking), 60 would be faster as a general rule. But under your specific inquiry, 8 disks, it would actually be better to stick with RAID 6 (faster writes than 60, but 60 would be faster at reads). As I presume you'll be writting more than reading (and reads are playback as completed files = not that much throughput is required; ie. watching a movie you just generated), go for faster write throughputs.

    Getting to the specifics as far as choking, the Areca and ATTO models listed in the other thread will be fine. They're the fastest cards out right now.

    Another little note, 60 is more common with relational databases, not large sequential files, which is what you'll be generating as output. So again, 6 is a better implementation for your usage IMO.

    But as you're in front of the system all the time, I'd actually opt for a level 5 (still better capacity and performance, at the cost of a reduced level of redundancy). If you were away from the system, such as it being an off-site location, I'd say go for 6 as a minimum (buys you time to get a human there to replace the disk/s that have gone south).

    The Areca also happens to have an Ethernet port on the back, and you can configure it to send email messages if there's a fault (keep the email open), and you'll have quick notice if there's a failure (it won't just flash a message across your screen), so you use email, or keep a browser open to the IP address: port the card uses. Those email alerts are very handy. ATTO doesn't do this, as they have their own software utility that accesses the card, but it will also give you the information you need. But the Areca cards are cheaper (better price/performance ratio), so it's a bit easier on the wallet.

    Worst case (you could learn a lot about your system), is to get the card and 8+ disks. Then create the various implementations (5/6/50/60), and test them out (real world, so use your software and some test projects). It takes time, but it will show you what your specific usage will be best served by more than any post anyone could make. ;)

    This is all presuming you do go with a RAID card of course. If not, go with level 10 and deal with it when you've run out of capacity, or it ceases to be fast enough (it will slow down as you fill the disks; happens with any mechanical disk implementation, so that's one of the reasons to go for larger capacity as it takes longer to get to this point).

    It's a lot to consider, and I'd advise you take your time, as mistakes tend to be expensive.

    Good luck. :)
     
  14. Honumaui macrumors 6502a

    Joined:
    Apr 18, 2008
    #14
    yeah but I knew what ya meant ;)

    I think sometimes things get to specific :) hehehehe



    this is also good and what I do with everything !! test my own situations my own things and usually try to state that when giving my thoughts to others ;)
    and what has lead me to find the best sweet spot in performance and price

    just like the SSD testing ?

    I do say though its not a bad idea to start from a point of others ? but always test it for yourself ;)

    I have found a few times what I heard and what I tested end up being two dif things ;)
     
  15. nanofrog macrumors G4

    Joined:
    May 6, 2008
    #15
    The smallest details matter with RAID. :eek: :p
     
  16. Honumaui macrumors 6502a

    Joined:
    Apr 18, 2008
    #16
    yes I know and I think you know I know that :) I meant I read what he was saying silly :) heheheheh

    everyone has small slips but generally people can read through them
     
  17. nanofrog macrumors G4

    Joined:
    May 6, 2008
    #17
    It's really easy to make a slip-up when posting on RAID, even for the most experienced professional.

    So my comment directed at the line you posted, and meant to be more fun than anything else, as the information in it is already known. :)
     
  18. Honumaui macrumors 6502a

    Joined:
    Apr 18, 2008
    #18
    I know :) why the silly and hehehehe in my line ;) writing never gets the humor across :) heehheheeh
     
  19. nanofrog macrumors G4

    Joined:
    May 6, 2008
    #19
    Of course. I just didn't want to get left out... :eek: :D :p
     
  20. JulianBoolean thread starter macrumors regular

    Joined:
    Aug 14, 2010
    #20
    Apologies are in order here. :) I directed the question to Nanofrog because I thought nobody else would be interested to answer such a raid specific question. I didn't send a PM because I thought that others might benefit from the response. I've completely overlooked two factors.

    1. I'm discovering that RAID related conversations are never uncomplicated, and that precise language is in order.

    2. The level of generosity, and the width and depth of the knowledge base found here at the MR forum.

    Big thanks to... goMac, Honomaui, alphahod, deconstruct60, and nanofrog. you guys are awesome!
     
  21. JulianBoolean thread starter macrumors regular

    Joined:
    Aug 14, 2010
  22. JulianBoolean thread starter macrumors regular

    Joined:
    Aug 14, 2010
    #22
    That's an awesome link, thanks! :)

    Unless somebody can offer additional info, critique, comparing real world vs benchmark testing I think that the benchmark tests look pretty definitive with regards to my question.

    JB
     
  23. Honumaui macrumors 6502a

    Joined:
    Apr 18, 2008
    #23
    yeah its a good link with good info on all the Areca cards to see what they do etc.. :)


    AND TAG your it !
     
  24. JulianBoolean thread starter macrumors regular

    Joined:
    Aug 14, 2010
    #24
    Okay ... :):):):):):) , and :eek::eek::eek::eek::eek: another thing is :D:D:D

    I think this thread might break the world record for emoticons. :p
     
  25. Honumaui macrumors 6502a

    Joined:
    Apr 18, 2008
    #25
    hehehe :) I use them all the time :) but I am a happy guy :) and dont believe in grammar so they take place of proper sentences for me :)

    except tomorrow got to go do a shoot and want to stay home and catch up on other things ! plus I pulled my back really bad :eek: so super sore !!! not going to be a fun day tomorrow !!!!:(
     

Share This Page