On Fri, 2011-12-02 at 19:36 +0800, Eric Dumazet wrote: > Le vendredi 02 décembre 2011 à 16:23 +0800, Alex Shi a écrit : > > From: Alex Shi <alexs@xxxxxxxxx> > > > > Times performance regression were due to slub add to node partial head > > or tail. That inspired me to do tunning on the node partial adding, to > > set a criteria for head or tail position selection when do partial > > adding. > > My experiment show, when used objects is less than 1/4 total objects > > of slub performance will get about 1.5% improvement on netperf loopback > > testing with 2048 clients, wherever on our 4 or 2 sockets platforms, > > includes sandbridge or core2. > > > > Signed-off-by: Alex Shi <alex.shi@xxxxxxxxx> > > --- > > mm/slub.c | 18 ++++++++---------- > > 1 files changed, 8 insertions(+), 10 deletions(-) > > > > netperf (loopback or ethernet) is a known stress test for slub, and your > patch removes code that might hurt netperf, but benefit real workload. > > Have you tried instead this far less intrusive solution ? > > if (tail == DEACTIVATE_TO_TAIL || > page->inuse > page->objects / 4) > list_add_tail(&page->lru, &n->partial); > else > list_add(&page->lru, &n->partial); For loopback netperf, it has no clear performance change on all platforms. For hackbench testing, it has a bit worse on 2P NHM 0.5~1%, but it is helpful to increase about 2% on 4P(8cores * 2SMT) NHM machine. I was thought no much cache effect on hot or cold after per cpu partial adding. but seems for hackbench, node partial still has much effect. > > > -- 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/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>