On Thu 28 Jun 2012, Michal Hocko wrote: > On Fri 22-06-12 13:26:14, Paul Slootman wrote: > > Perhaps I'm triggering something that exists since before 3.0, but > > anyway: > > > > After some time, all swap space gets gradually used up, without a clear > > indication what's using it (at least, I haven't managed to find out). > > > > System is running debian testing, and most usage is a lot of rxvt > > processes mostly ssh'ed out to other systems, and google chrome. > > I suspect google chrome may be the cause of the problem. > > Root is btrfs, /home is NFS. > > > > The system earlier had 4GB RAM and swap is currently 5 x 2GB LVM > > partitions. With that config I needed to reboot after about a week, as > > the system ended up thrashing the swap. I've added 8GB RAM, and now the > > uptime is 42 days, system still usable. > > > > Stopping google-chrome at such a point in time usually does not help. > > > > At every reboot I upgrade to the latest kernel :) Currently running > > 3.4.0-rc6, but I saw the same behaviour with all 3.x kernels I tried. Memory was full again yesterday, at which point I tried 3.5.0-rc4. Unfortunately something there (or something I may have changed in the config) prevents my google chrome from starting all of my open tabs; about 1/3 remain blank with a loading spinner running. Opening a new tab and entering one of those URLs gives "window not responding" error after some time. Wierd. > > I would have thought that with almost 10GB memory free (w/o cache) such > > a swapoff should succeed. I also wonder why that 9GB cached memory is > > being held; it's not released after echo 3 > drop_caches . > > Shmem: 9181016 kB <<< > Because the most of the memory is anonymous and shmem. > ipcs -pm should tell you about the current segments and pids behind. OK, I'll do that the next time, thanks. I hadn't noticed the Shmem line (I didn't really know where to begin looking :-) I find it a bit unexpected that this is shown by "free" as cached memory. I did notice however, that after restarting the X server the memory apparently _was_ released (stopping all the windows didn't seem to help). Paul -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>