Latest: 10.10.4 uses Sequential Trim "Warning: 10.10.4 might use Queued Trim instead of Sequential!"

Discussion in 'Mac Pro' started by haralds, Jul 1, 2015.

  1. haralds, Jul 1, 2015
    Last edited: Jul 12, 2015

    haralds macrumors 6502

    haralds

    Joined:
    Jan 3, 2014
    Location:
    Silicon Valley, CA
    #1
    Note: this has been resolved not to be an issue!

    I just lost my 5TB Fusion drive based on a Corsair M500 (CT960) that had been working great until the 10.10.4 upgrade. While moving blocks from/to SSD, the system totally corrupted.

    My hypothesis is that while earlier versions were using Sequential Trim, Apple might have switched. Some Corsair and I believe Samsung drive fail with Queued Trim.

    Of rouse, this is independent of whether you use Trim Enabler of the new native command.

    Caveat emptor! I am now staring at a 30plus hours TimeMachine restore....
     
  2. crjackson2134 macrumors 68020

    crjackson2134

    Joined:
    Mar 6, 2013
    Location:
    Charlotte, NC
    #2
    I'll see if I can find the article again, but OS X is using sequential trim. Only linux has the option of queued/sequential. There are many reasons data can get hosed other than trim.
     
  3. crjackson2134 macrumors 68020

    crjackson2134

    Joined:
    Mar 6, 2013
    Location:
    Charlotte, NC
    #3
    This isn't what I was looking for but it may be helpful to you, I'll keep searching.

     
  4. haralds thread starter macrumors 6502

    haralds

    Joined:
    Jan 3, 2014
    Location:
    Silicon Valley, CA
    #4
    I am waiting for an assured test on the latest 10.10.4 drivers. The failure behavior absolutely looks like a trim failure.

    The fact the El Capitan does not have it in the current beta build does not mean the drivers for 10.10.4 release were not changed. BTW, I ran the beta builds of 10.10.4 without problems.

    Caveat emptor!
     
  5. crjackson2134, Jul 1, 2015
    Last edited: Jul 1, 2015

    crjackson2134 macrumors 68020

    crjackson2134

    Joined:
    Mar 6, 2013
    Location:
    Charlotte, NC
    #5
    There is a script to test for that already, but I hesitate to post it since it's not mine, and it's not intended for use by novice users. Create an account at clindori's website and you can find the script. 10.10.4 has already been tested and passed the muster, it uses sequential trim. It works fine for me and everyone else I know. If you don't want to use it that's fine too, you likely won't see much difference for quite some time anyway.

    BTW, the script is intended for use on an empty drive...
     
  6. mikeboss macrumors 65816

    mikeboss

    Joined:
    Aug 13, 2009
    Location:
    switzerland
    #6
    I don't think that we're doing anything wrong by posting the script here... of course I'll delete it if the author of the script (not me) has anything against this... so, here it is.
     

    Attached Files:

  7. haralds thread starter macrumors 6502

    haralds

    Joined:
    Jan 3, 2014
    Location:
    Silicon Valley, CA
    #7
    Here is why this was the most likely hypothesis:
    - No problem with drive temps and SMART. Drive health has been great.
    - No console errors, the OS thought everything was fine
    - Rapidly expanding drive "file permission" errors starting with frequently used files in ~/Library as they were moved.

    To the system everything looked fine. But as it regrouped the Fusion content, the Corsair started mucking up sector assignments - permanently.

    Still could be something else, but given the timing and the known history, this looked like a logical hypothesis.

    We will know in 30+ hours, whether there is another issue.

    The system has been error and crash free for months.
     
  8. crjackson2134 macrumors 68020

    crjackson2134

    Joined:
    Mar 6, 2013
    Location:
    Charlotte, NC
    #8
    Not necessarily, I've had spinners (obviously not fusion drives) present similar problems when moving data, but there is no trim to blame there. Reformatting and refreshing everything seems to fix the problem forever. It's possible the problem was related to the fusion implementation rather than trim. Also, since there is a rotational drive in the mix, it's that much harder to pin down. It could be related to trim, but even if it is, that doesn't mean that OS X is using queued trim commands instead of sequential.

    I can see that you are a hard sell, so I won't bother you with this anymore.

    This seems to be a logic Cause and Effect fallacy -
    assuming that the effect is related to a cause because the events occur together.
    1. Example: When the rooster crows, the sun rises. Therefore, the rooster causes the sun to rise.
    2. Example: When the fuel light goes on in my car, I soon run out of gas. Therefore, the fuel light causes my car to run out of gas.
     
  9. haralds thread starter macrumors 6502

    haralds

    Joined:
    Jan 3, 2014
    Location:
    Silicon Valley, CA
    #9
    Thanks. That is why I called it a hypothesis. It has not advanced yet to a full theory
    The system has worked in this configuration for well over a year up and through 10.10.4 beta (which did not have the trimforce command, I checked.)
    "Correlation is no causation" is another name for this and one of my pet peeves, too.
     
  10. ZMacintosh macrumors 65816

    ZMacintosh

    Joined:
    Nov 13, 2008
    #10
    how was your Fusion drive constructed? was it natively created when installed OS X or through a command/hack? im wondering if it was through a command/hack and trimforce enabled it jolted that.

    if you have an SSD+HDD OS X will auto create a fusion drive for you, i'd TRIM The SSD first then have OSX recreate your fusion, then reinstall and restore and see if that goes smoother
     
  11. filmak, Jul 2, 2015
    Last edited: Jul 2, 2015

    filmak macrumors 6502a

    Joined:
    Jun 21, 2012
    Location:
    between earth and heaven
    #11
    I also think that something probably went wrong with your fusion setup.
    Maybe the trim change just triggered the fault.
    Also perhaps there is something with the specific restrictions of the MacPro's SATA2 bandwidth and your fusion drive.

    It seems that they still use the sequential method for TRIM.

    We have enabled trim in a Mac Pro 3,1 under 10.10.4 with two SSDs, a Crucial M500 and a Samsung evo 840 and everything seems fine, till now.
     
  12. haralds thread starter macrumors 6502

    haralds

    Joined:
    Jan 3, 2014
    Location:
    Silicon Valley, CA
    #12
    Cool. Keep us posted. I hope, I am wrong...
     
  13. ZMacintosh macrumors 65816

    ZMacintosh

    Joined:
    Nov 13, 2008
    #13
    sounds like you need to enable trim, then recreate your fusion drive.
     
  14. haralds thread starter macrumors 6502

    haralds

    Joined:
    Jan 3, 2014
    Location:
    Silicon Valley, CA
    #14
    Actually, the sequence of events does not matter, since they are unrelated.

    But I recreated my system and backed it up once more. Based on the lack of evidence that the OS X trim mode changed, I enabled it again after running some benchmarks. I will keep you posted on whether I will run into issues again.

    BTW, there is no "hack" involved in creating a Fusion Drive. Apple simply has not (yet) provided a high level tool. It simply involves diskutil to create a coreStorage drive out of two or more drives with the SSD first and then creating a logical jhfs+ volume on this logical drive. That's it!

    BTW, I really could not see a difference in performance, but testing it just was too much of an itch...
     
  15. haralds thread starter macrumors 6502

    haralds

    Joined:
    Jan 3, 2014
    Location:
    Silicon Valley, CA
    #15
    I just
    I just was pointed to what looks like a knowledgable post here:
    http://forums.macrumors.com/threads...ved-performance.1891936/page-10#post-21469307

    I have since used the native command to enable trim, all seems to be fine so far.
     
  16. lowendlinux Contributor

    lowendlinux

    Joined:
    Sep 24, 2014
    Location:
    North Country (way upstate NY)
    #16
    I was going to say the same thing but in a more curt way, I'm glad you responded.
     
  17. haralds thread starter macrumors 6502

    haralds

    Joined:
    Jan 3, 2014
    Location:
    Silicon Valley, CA
    #17
    Luckily, this seems to be a non-issue. MacCNN did some extensive testing and all looks good so far:
    http://www.macnn.com/articles/15/07...noted.in.testing.the.popular.ssd.line.129472/
     
  18. TonyK macrumors 6502a

    TonyK

    Joined:
    May 24, 2009
    #18
    My M500 Crucial on 10.10.4 so far has had no issues for several weeks. What makes you think Apple changed from sequential to queued for their trim commands? Everything I've read has indicated only some Linux distributions enabled queued trim.
     
  19. DPUser macrumors 6502a

    Joined:
    Jan 17, 2012
    #19
    He doesn't think that anymore… :)
     

Share This Page