Re: [PATCH RFC 1/5] sched,numa: build table of node hop distance

[Rik van Riel]

On 10/14/2014 02:47 AM, Peter Zijlstra wrote:
> On Sun, Oct 12, 2014 at 09:28:04AM -0400, Rik van Riel wrote:
> > On 10/12/2014 09:17 AM, Peter Zijlstra wrote:
> > > On Wed, Oct 08, 2014 at 03:37:26PM -0400, riel@xxxxxxxxxx wrote:
> > > > +	sched_domains_numa_hops = kzalloc(sizeof(int) * nr_node_ids * nr_node_ids, GFP_KERNEL);
> > > > +	if (!sched_domains_numa_hops)
> > > > +		return;
> > >
> > > That's potentially a _BIG_ table (1M for a 512 node system).
> > > The node_distance has magic allocations and is of u8 size, is there any
> > > way we can re-use node_distance and avoid a second O(n^2) allocation?
> >
> > You are right, this should be a u8 at the least.
> >
> > Beyond that, I am not convinced that merging things into
> > the same array is worthwhile, since (IIRC) nr_node_ids
> > should be set to the actual number of nodes on the system
> > by then.
>
> The thing is, it looks like all you do is compare hop distance, and the
> order of the hop distances is the exact same order as the regular numa
> distance. I could not find a place where you use the actual hop value.

I use the actual hop distances when doing the scoring
for glueless mesh topologies, in patch 4/5.

> So if all you're interested in is the relative ordering, that should be
> the same for both.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/