> Hi, I find an very strange bug, And it is reproduce-able.
>
> I have an machine with Adaptec 7899p card, And two scsi harddisk on one
> scsi ID. It means I need use CONFIG_SCSI_MULTI_LUN options on
> kernel. My kernel is SGI_XFS_1.0.1, I just enable the options, And
> recompile it using RedHat 7.1 default gcc-2.9.6-85.
>
> My scsi harddisk shows to be sda sdb
>
> When I run this script:
> mkfs.xfs /dev/sdb1
> time mount -t xfs /dev/sdb1 /mnt
>
> I got this output
>
> real 0m9.340s
> user 0m0.000s
> sys 0m0.000s
>
> The mount speed is real slow.
> Using ext2xfs
>
> mkfs /dev/sdb1
> time mount /dev/sdb1 /mnt
> output is :
>
> real 0m0.392s
> user 0m0.000s
> sys 0m0.000s
>
> Using resiserfs also very fast.
>
> lvol1 can be from 500M - 10 G , And the result is same.
>
> In the same time I run
>
> mkfs.xfs /dev/sda5
> time mount -t xfs /dev/sda5 /mnt
>
> result is:
>
> real 0m0.523s
> user 0m0.000s
> sys 0m0.000s
>
> sda5 size is same with sdb1
>
> So the question is very clear, mount an XFS on the MULTI_LUN scsi
> device is very slow. Who know the reason?
>
> Dan
>
Well, first you have to explain what the MULTI_LUN thing does, because I
have no idea. However, from XFS's point of view, a device is a device
is a device - unless it is LVM. There are some differences there in
terms of how we do disk I/O to the log.
I would ask that you try a later kernel, there have been changes in
all sorts of things since the 1.0.1 release, there is every chance it
is fixed by some other kernel change. If not, there is probably not a
lot we can do about it, the disk I/O involved in mount is pretty
fixed (a binary chop scan of the log blocks looking for the head and
tail of the log).
Steve
|