On Thu, Sep 04, 2014 at 11:29:29AM -0400, Johannes Weiner wrote: > On Tue, Sep 02, 2014 at 04:21:43PM +0100, Mel Gorman wrote: > > On Tue, Sep 02, 2014 at 09:51:20AM -0400, Johannes Weiner wrote: > > > On Mon, Sep 01, 2014 at 01:55:51PM +0100, Mel Gorman wrote: > > > > I cannot find a good reason to incur a performance penalty on all > > > > 64-bit NUMA > > > > machines in case someone throws a brain damanged TV or graphics card in > > > > there. > > > > This patch defaults to node-ordering on 64-bit NUMA machines. I was > > > > tempted > > > > to make it default everywhere but I understand that some embedded > > > > arches may > > > > be using 32-bit NUMA where I cannot predict the consequences. > > > > > > This patch is a step in the right direction, but I'm not too fond of > > > further fragmenting this code and where it applies, while leaving all > > > the complexity from the heuristics and the zonelist building in, just > > > on spec. Could we at least remove the heuristics too? If anybody is > > > affected by this, they can always override the default on the cmdline. > > > > I see no problem with deleting the heuristics. Default node for 64-bit > > and default zone for 32-bit sound ok to you? > > Is there a strong reason against defaulting both to node order? Zone > ordering, if anything, is a niche application. We might even be able > to remove it in the future. We still have the backup of allowing the > user to explicitely request zone ordering on the commandline, should > someone depend on it unexpectedly.
Low memory depletion is the reason to default to zone order on 32-bit NUMA. If processes on node 0 deplete the Normal zone from normal activity then other nodes must keep reclaiming from Normal for all kernel allocations. The problem is worse if CONFIG_HIGHPTE is not set. A default of node-ordering on 32-bit NUMA increases low memory pressure leading to increased reclaim and potentially easier to trigger OOM. I expect this problem was worse in the past when the normal zone could be filled with dirty pages under writeback. However low memory pressure is still enough of a concern that I'm wary of changing the default of 32-bit NUMA without knowing who even cares about 32-bit NUMA. -- Mel Gorman SUSE Labs -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/