Is there data around this about any possible changes in performance of
either file writes/reads/deletes, or general memory utilization during
writes/reads/deletes?
Specifically, it takes n long to delete x number of files, or takes n
long to create x number of directories, or takes n long to read x number
of files or x number of megs into memory, etc?
The reason I ask is that news and mail servers seem to be the most
common file-corruption witnesses, could this help them?
On Tue, 2002-03-26 at 10:17, Steve Lord wrote:
> This gets the xfs inode down to 400 bytes, and as a consequence we can
> fit 10 into a page now. So we reduce the slabcache memory consumption
> of xfs.
>
> Date: Tue Mar 26 08:19:25 PST 2002
> Workarea: jen.americas.sgi.com:/src/lord/xfs-ishrink
>
> The following file(s) were checked into:
> bonnie.engr.sgi.com:/isms/slinx/2.4.x-xfs
>
>
> Modid: 2.4.x-xfs:slinx:114887a
> linux/fs/xfs/xfsidbg.c - 1.177
> - cleanup inode debug functions to remove fields
>
> linux/fs/xfs/xfs_rw.h - 1.62
> - prototype change for xfs_inval_cached_pages
>
> linux/fs/xfs/xfs_rw.c - 1.353
> - make xfs_inval_cached_pages take a parameter indicating the
> lock state of the iolock.
>
> linux/fs/xfs/xfs_vnodeops.c - 1.520
> - remove usage of i_dev
>
> linux/fs/xfs/xfs_itable.c - 1.102
> - remove i_dev
>
> linux/fs/xfs/xfs_dmapi.c - 1.50
> - i_dev replaced with i_mount->m_dev
>
> linux/fs/xfs/xfs_inode_item.c - 1.97
> - access i_pincount via function now, it is an atomic_t
>
> linux/fs/xfs/xfs_iocore.c - 1.28
> - make io_lock and io_iolock debug only
>
> linux/fs/xfs/xfs_vfsops.c - 1.339
> - access i_pincount via function now, it is an atomic_t
>
> linux/fs/xfs/xfs_dfrag.c - 1.28
> - Add parameter to xfs_inval_cached_pages indicating ilock state
>
> linux/fs/xfs/xfs_iget.c - 1.152
> - replace the sv and spinlock for inode pinning with a waitqueue
> and an
> atomic_t
>
> linux/fs/xfs/xfs_mount.c - 1.276
> - remove i_dev usage
>
> linux/fs/xfs/xfs_inode.c - 1.329
> - i_dev replaced with i_mount->m_dev, remove assert references
> to
> io_queued_bufs and use lighter weight implementation for inode
> pinning.
>
> linux/fs/xfs/xfs_inode.h - 1.157
> - Remove unused fields in the xfs_inode and xfs_iocore
> structures
>
> linux/fs/xfs/linux/xfs_lrw.c - 1.129
> - Add parameter to xfs_inval_cached_pages indicating ilock state
>
> linux/include/linux/xfs_support/mrlock.h - 1.3
> - shrink mrlock 4 more bytes
>
--
Austin Gonyou
Systems Architect, CCNA
Coremetrics, Inc.
Phone: 512-698-7250
email: austin@xxxxxxxxxxxxxxx
"It is the part of a good shepherd to shear his flock, not to skin it."
Latin Proverb
|