SQL Server Performance

ML350 G4p, Sustained disk IO is poor

Discussion in 'SQL Server 2005 Performance Tuning for Hardware' started by egmiii, Apr 12, 2007.

  1. egmiii New Member

    I recently ran HD Tach on my drives and was surprised at the poor results.

    1 HP Smart Array 6400 controller (2 Channels)
    C:/D: 2x300GB 10,000 RPM (3.5) RAID 1, Channel 1
    E:/F: 4x146GB 10,000 RPM (3.5) RAID 1+0, 2 drives on Channel 1 and 2 drives on Channel 2

    The C/D drive goes from 75 to 40 MB/s as reads progress across the disk.
    This is great. Makes sense for a 2 drive mirror.

    The E/F drive starts at 115 MB/s, holds at that rate till the 235GB point, then starts declining to 80MB/s at the end of the drive. This does not seem right.

    All 6 drives are Ultra320. It seems like there is a bottleneck limiting the peak performance of the E/F combination. I was thinking the system bus, but it's a 64-bit PCI-X, which should be able to handle the throughput.

    Any ideas? Anything I can check in the HP managment tools? This has been bugging me for days.
  2. joechang New Member

    the drop off over the length of the disk is normal,
    see Toms Hardware Guide for typical disk transfer rate curves

    only the 4 drive config seems a little low
    check for alignment on disk part,
    also depends on what HD Tach actually does
  3. egmiii New Member

    Thanks for your response. The C/D drive follows the proper curve as the drive head moves from the outer to inner part. Should all RAID 10 arrays on HP hardware follow a curve, provided they are not flooding the channel or PCI bus?

    What tool would you recommend for checking the sustained read speed of an array? This is a QA server, so I can't blow away the existing database files.

    How do I verify the alignment? I ran a VBS script you submitted to the forum a few months ago, but the data doesn't mean much to me.

    Thanks again...
  4. joechang New Member

    what might be happening is that your db is already occupying the fast portion

    sequential full disk tests can only be done on clean (ie, new) config
    without an exisiting data

    next time you set up a new prod server schedule 1-2 weeks for performance testing before putting live data on it

Share This Page