-t test result depends on file system!?

Jan Wagner
2007-12-13
2013-04-27
  • Jan Wagner
    Jan Wagner
    2007-12-13

    Hi,

    I've a 12 SATA disk RAID0 on /dev/md0 and it is formatted for example with vfat. If /dev/md0 is _not_ mounted, I get

    ----------------------
    oper@abidal:/$ sudo hdparm -tT /dev/md0
    /dev/md0:
    Timing cached reads:   2090 MB in  2.00 seconds = 1045.65 MB/sec
    Timing buffered disk reads:  1790 MB in  3.00 seconds = 596.24 MB/sec
    ----------------------

    Howver, if /dev/md0 is mounted to for example /raid, I get much slower throughput:

    ----------------------
    oper@abidal:/$ sudo hdparm -tT /dev/md0
    /dev/md0:
    Timing cached reads:   2084 MB in  2.00 seconds = 1042.09 MB/sec
    Timing buffered disk reads:  544 MB in  3.05 seconds = 178.57 MB/sec
    ----------------------

    The hdparm man page says about -t: "This measurement is an indication of how fast the drive can sustain sequential data reads under Linux, without any filesystem overhead." In a way, the results above contradict this. Is it a bug in hdparm? Or is hdparm just not intended for md block devices? What other tool could be used then for read benchmarks?

     
    • Jan Wagner
      Jan Wagner
      2007-12-13

      Forgot to add, the read thoughputs differ from the above vfat result if the mounted file system is for example xfs, jfs, or ext2. The mounted filesystems are idle while 'hdparm -t' is running. So I wonder where do the differences in the reported read throughput come from?