On Wed, Oct 10, 2018 at 01:03:50AM +0000, Rik van Riel wrote: > On Tue, 2018-10-09 at 14:47 -0400, Johannes Weiner wrote: > > > These workloads also deal with tens of thousands of open files and > > use > > /proc for introspection, which ends up growing the proc_inode_cache > > to > > absurdly large sizes - again at the cost of valuable cache space, > > which isn't a reasonable trade-off, given that proc inodes can be > > re-created without involving the disk. > > > > This patch implements a "zero-seek" setting for shrinkers that > > results > > in a target ratio of 0:1 between their objects and IO-backed > > caches. This allows such virtual caches to grow when memory is > > available (they do cache/avoid CPU work after all), but effectively > > disables them as soon as IO-backed objects are under pressure. > > > > It then switches the shrinkers for procfs and sysfs metadata, as well > > as excess page cache shadow nodes, to the new zero-seek setting. > > This patch looks like a great step in the right > direction, though I do not know whether it is > aggressive enough. > > Given that internal slab fragmentation will > prevent the slab cache from returning a slab to > the VM if just one object in that slab is still > in use, there may well be workloads where we > should just put a hard cap on the number of > freeable items these slabs, and reclaim them > preemptively. > > However, I do not know for sure, and this patch > seems like a big improvement over what we had > before, so ... Fully agreed, fragmentation is still a concern. I'm still working on that part, but artificial caps and pro-active reclaim are trickier to get right than prioritization, and since these patches here are useful on their own I didn't want to hold them back. > > Reported-by: Domas Mituzas <dmituzas@xxxxxx> > > Signed-off-by: Johannes Weiner <hannes@xxxxxxxxxxx> > > Reviewed-by: Rik van Riel <riel@xxxxxxxxxxx> Thanks!