public inbox for linux-scsi@vger.kernel.org
 help / color / mirror / Atom feed
* Slower write performance in RHEL6
@ 2011-07-26 11:22 Li Liu
  2011-07-26 11:38 ` Ric Wheeler
  0 siblings, 1 reply; 2+ messages in thread
From: Li Liu @ 2011-07-26 11:22 UTC (permalink / raw)
  To: linux-scsi

Dear all,

    In RHEL6(linux-2.6.32-71.el6), I added some codes in
scsi_softirq_done to check the result of execute scsi commands:
WRITE_6, WRITE_10, WRITE_12, WRITE_16 when I executed "dd if=/dev/zero
of=/dev/xda bs=64k count=10000". I found a lot of write error messages
in scsi_softirq_done, the disposition(variable) was ADD_TO_MLQUEUE and
the speed was 78MB/s.
    Then I changed MPT_SCSI_CMD_PER_DEV_HIGH from 64 to 32 in
mptscsih_change_queue_depth, the speed was 140MB/s and none error
messages.
    * /dev/xda is a scsi disk.
    * the scsi host is LSI1064E.
    * MPT_SCSI_CMD_PER_DEV_HIGH will be assigned to blk_queue_tag->max_depth.

    My questions are:
    Is the TCQ depth limited by hard disk, how can I get the limited?
    Why the speed was drop to 78MB/s when MPT_SCSI_CMD_PER_DEV_HIGH was 64?
    Could I change MPT_SCSI_CMD_PER_DEV_HIGH to 32 or 16 or 8, Will it
cause any problem?

    Thanks for any ideas.

Thanks,
Alex Liu

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2011-07-26 11:38 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-07-26 11:22 Slower write performance in RHEL6 Li Liu
2011-07-26 11:38 ` Ric Wheeler

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox