Hi.... I cc: the reply to the new list address. I hope you don't mind.... Now, let's what I can share here.... On Tue, Jan 4, 2011 at 18:49, Sowmya Sridharan <sowmya.sridharan@tcs.com> wrote:
I have seen slab cache increase by even 1GB, over two to three days on a stable system.(The system was sending and receiving bulk amounts of packets).
No kidding? 1 Gig? Wow.... But that's alright, I think that still make sense ... highly frequent sock related cache creation I guess.... "slabtop -s c" would tell you the highest cache size ...
When is slab usually freed?
AFAIK when it can no longer "grow".... same like page cache, it will grow to fill your RAM until it exhausts everything left by anonymous page allocation plus some amount of reserved pages. And AFAIK too, it's tunable, but I couldn't recall which kernel parameters that control it.
Also when I analyzed /proc/slabinfo, I was able to see that the number of active slab objects were increasing mostly for task_struct, dentry_cache and proc_inode_caches. When are these caches allotted objects, in general?
task_struct--> during new process creation dentry_cache --> after reading directory entry metadata from a block devices and cache them proc_inode_cache --> almost similar like above, but this one is strictly for /proc entries. -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com