On Mon, 22 Sep 2008 09:40:43 +0900 "MinChan Kim" <minchan.kim@xxxxxxxxx> wrote: > On Mon, Sep 22, 2008 at 8:52 AM, Rik van Riel <riel@xxxxxxxxxxx> > wrote: > > On Mon, 22 Sep 2008 07:54:01 +0900 > > MinChan Kim <minchan.kim@xxxxxxxxx> wrote: > > > >> Do You mean current page reclaim mechanism have a problem ? > >> As you know, kernel already have reclaimed fs-specific metadata. > >> What is a problem? > > > > The problem is that that buffer heads are only reclaimed when > > the pages in that memory zone are scanned. > > > > This does not help when you have 63GB of highmem and only > > 1GB of lowmem - low memory will be full long before the VM > > starts scanning high memory. > > Thanks for explaining, Rik. > > Just out of curiosity, I think when memory manager get the pressure, > it will start scanning then it can reclaim metadata in low memory. > What's wrong? It does not scan the metadata directly. It only runs into the buffer heads when it scans the pages those buffer heads are attached to. It will not scan those pages because of lowmem pressure, because those pages are in highmem (which still has lots of free memory). -- All rights reversed. -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
