Re: [PATCH RFC 5/5] sched,numa: find the preferred nid with complex NUMA topology

[Rik van Riel <riel@redhat.com>]

On 10/12/2014 10:56 AM, Peter Zijlstra wrote:
> On Wed, Oct 08, 2014 at 03:37:30PM -0400, riel@redhat.com wrote:
>> +static int preferred_group_nid(struct task_struct *p, int nid)
>> +{
>> +	nodemask_t nodes;
>> +	int hops;
>> +
>> +	/* Direct connections between all NUMA nodes. */
>> +	if (sched_numa_topology_type == NUMA_DIRECT)
>> +		return nid;
>> +
>> +	/*
>> +	 * On a system with glueless mesh NUMA topology, group_weight
>> +	 * scores nodes according to the number of NUMA hinting faults on
>> +	 * both the node itself, and on nearby nodes.
>> +	 */
>> +	if (sched_numa_topology_type == NUMA_GLUELESS_MESH) {
>> +		unsigned long score, max_score = 0;
>> +		int node, max_node = nid;
>> +
>> +		hops = sched_domains_numa_levels;
>> +
>> +		for_each_online_node(node) {
>> +			score = group_weight(p, node, hops);
>> +			if (score > max_score) {
>> +				max_score = score;
>> +				max_node = node;
>> +			}
>> +		}
>> +		return max_node;
>> +	}
>
> This too is O(nr_nodes^2), right?
>
It is, but I suspect the glueless mesh topologies are
never larger than on the order of a dozen nodes or so.

Would you prefer me to make the optimization I proposed
in the other email, or should I just add in a comment
stating that that optimization could be made if it turns
out to be necessary?