RAID 10, Crystal Disk Mark Weird score?

  • Manufacturer replied..

    When i created the vols for the RAID 10s (1 with 6 disks and 1 with 8 disks) and RAID with the default max vol size.

    They have asked that i delete all the volumes and recreate them all the same size. Timing it to show how the RAID 10 8 disk is quicker than the 6 disk which is quicker than RAID 1. And that my Tests on the Vols will be better...

    Is this normal? should you really see a improvement in speed with a smaller vol compared to a larger one???

  • stebennettsjb (8/1/2011)


    Manufacturer replied..

    When i created the vols for the RAID 10s (1 with 6 disks and 1 with 8 disks) and RAID with the default max vol size.

    They have asked that i delete all the volumes and recreate them all the same size. Timing it to show how the RAID 10 8 disk is quicker than the 6 disk which is quicker than RAID 1. And that my Tests on the Vols will be better...

    Is this normal? should you really see a improvement in speed with a smaller vol compared to a larger one???

    Depending on how much of the disk surface (and which portions of it) you're using in your tests, you may be doing "short-stroking"; the practice of using only the faster portion of the disk. On a spinning circle with a fixed RPM rate, simple geometry tells us that the farther from the center you are, the faster the linear speed under a fixed point (we'll call this the "drive head" for now). The faster the linear speed under the drive head, the faster the sequential data transfer can be.

    With SQLIO, at least, you can generate your test files, use something like Piriform's Defraggler or another defragmentation tool with a graphical interface to see where the files are, and potentially run two tests; one at the "start" of the disk and one at the "end" of the disk (Move large files to the end is an option in Defraggler), to compare the difference.

  • Nadrek,

    simple but genius!

    thanks..

Viewing 3 posts - 16 through 17 (of 17 total)

You must be logged in to reply to this topic. Login to reply