Sandforce SSD's in RAID0: Does background garbage collection still work?

Discussion in 'Mac Pro' started by wirelessness, Aug 26, 2010.

  1. wirelessness macrumors 6502

    Joined:
    Jun 20, 2010
    #1
    I am thinking about adding a pair of Sandforce SSD drives in RAID0 on my Mac OSX device. Since OSX does not support trim I am dependent entirely on over-provisioning and idle garbage collection to keep SSD drives in good shape long term.

    My question is if you use two drives that support BGC in a software RAID0 configuration does the BGC still work as effectively as with a single drive long term?
     
  2. VirtualRain macrumors 603

    VirtualRain

    Joined:
    Aug 1, 2008
    Location:
    Vancouver, BC
    #2
    Yes, the BGC as you call it is a function of the drive's controller and will work independently of higher level implementation details (RAID or OS).

    Let's hope that Apple get's TRIM implemented soon though, and for RAID arrays of SSD's as well, since that's a common use case for early adopters like us. (BTW, Intel recently released windows drivers to support the TRIM command for logical RAID0 arrays as well)
     
  3. wirelessness thread starter macrumors 6502

    Joined:
    Jun 20, 2010
    #3
    I wonder if it matters for OSX if you use a drive with 28% over-provisioning ala the OWC Mercury Extreme Pro RE or if you can now use an extended SSD drive like the OCZ Vertex 2 (similar to none RE from OWC) w/ 7% over-provisioning?
     
  4. VirtualRain macrumors 603

    VirtualRain

    Joined:
    Aug 1, 2008
    Location:
    Vancouver, BC
    #4
    If you are using the drive in a very write heavy situation (database transaction server, scratch drive, etc.) then you want more over-provisioning, but for a typical desktop drive which is 90% reads, then the OCZ's 7% over provisioning is plenty.
     
  5. deconstruct60 macrumors 604

    Joined:
    Mar 10, 2009
    #5
    Given the same write rates with or without RAID-0 it should, if anything, work more effectively. Minimally it should stay the same.

    From the drive's perpsective it has no idea what it is connected to. It merely accepts a series of read/write requests. Whether that comes directly for the host, a RAID card, or two hamsters on a treadmill, it has no clue.

    Since RAID-0 splits up the workload, the amount of requests per device goes down (again in the context of a fixed workload thrown at both a single SSD or multiple ones. ). For a fixed workload, the amount of idle time going down is hard pressed to happen when the incoming requests have also gone down. What exactly is consuming the idle time in that context ?

    Pragmatically, you may see an increases in requests because the workload thrown at the two contexts are different. Because the RAID-0 set up is faster throw more work at it. That's seems doubtful in most cases because with SATA II bottlenecks not going to get huge speed ups. That is more likely to happen in case of single versus RAID-0 spinning hard drives because there is a much bigger gulf in performance between the two.

    Similarly, if throwing extra money at storage to implement RAID ( minimally 2 * disk costs ) then likely not going to be doing "normal" number writes either. There are some situations where are a special case "90% read" , but like stated above once have faster performance, normally more folks also leverage it. So would see a net increase of both read and writes in many workloads. That's why there are RE drives. Folks are more likely to attempt to abuse them more than normal in a RAID set up. Users who moderate themselves while stepping to RAID-0 then they don't need them as much.
     

Share This Page