This adds yet another mount option for XFS, kiocluster, this indicates
use kiobufs for all I/O requests, and use write clustering on converting
delalloc writes to real disk blocks, the kio flag now means use kiobufs
but do not do write clustering. Currently write clustering is giving
worse performance turned on rather than off - probably more to do with
when it is triggered rather than anything else.
This also turns off all the compiled in debug printk code in pagebuf
and the pagebuf tracing code being compiled in. These can be turned
back on for debugging via editing page_buf.h
Date: Wed Aug 16 11:07:38 PDT 2000
Workarea: jen.americas.sgi.com:/src/lord/xfs-linux.2.4.0-test5
The following file(s) were checked into:
bonnie.engr.sgi.com:/isms/slinx/2.4.0-test1-xfs
Modid: 2.4.0-test1-xfs:slinx:72194a
linux/include/linux/fs.h - 1.56
- Add MS_KIOCLUSTER super block flag.
linux/fs/xfs/linux/xfs_super.c - 1.82
- Add kiocluster mount option to control used of clustered writes for
delalloc page cleaning.
linux/include/linux/page_buf.h - 1.62
- Remove some prototypes. Also turn off pagebuf tracing code compilation
and debugging printks by default. If you want these built into the
kernel
you NEED to edit page_buf.h now.
linux/fs/xfs/linux/xfs_mount_opt.c - 1.11
- Add kiocluster mount option to control used of clustered writes for
delalloc page cleaning.
linux/fs/pagebuf/page_buf.c - 1.24
- Remove some dead code, add options to allow us to play with page
locking more.
linux/fs/pagebuf/page_buf_io.c - 1.23
- Separate out the decision to use I/O clustering on delalloc page
cleaning
from the decision to use kiobufs for I/O. Also do not allow write
requests
for clustered delalloc requests to linger in the request queue for too
long with all the pages locked down.
|