On Mon, Oct 11, 2004 at 11:16:01AM -0500, Chris Evert wrote:
> Nathan Scott wrote:
> >
> >/proc/slabinfo will hold some clues, send it over this way.
Oh, another useful tidbit is /proc/meminfo.
> I bumped /proc/sys/vm/vfs_cache_pressure to 150 and that seemed to work
> til this weekend.
>
> A snapshot of /proc/slabinfo during this time:
Hmmm, well, these are your big slab users...
> slabinfo - version: 2.0
> # name <active_objs> <num_objs> <objsize> <objperslab>
> <pagesperslab> : tunables <batchcount> <limit> <sharedfactor> : slabdata
> <active_slabs> <num_slabs> <sharedavail>
> ext3_inode_cache 6898 17840 512 8 1 : tunables 54 27
> 8 : slabdata 2230 2230 0
> radix_tree_node 35085 46284 276 14 1 : tunables 54 27
> 8 : slabdata 3306 3306 0
> inode_cache 2054 2080 384 10 1 : tunables 54 27
> 8 : slabdata 208 208 0
> dentry_cache 10215 26468 152 26 1 : tunables 120 60
> 8 : slabdata 1018 1018 0
> buffer_head 486022 486525 52 75 1 : tunables 120 60
> 8 : slabdata 6487 6487 0
XFS seems to be playing ball and giving back slab cache memory
when asked to do so - ext3 is holding onto a fair bit, and I
can't tell who's used all those buffer heads - its the VMs
responsibility to get rid of those, with help from individual
filesystems.
cheers.
--
Nathan
|