Re: [patch 3/3] x86_64: Node local pda take 2 -- node local pda allocation

[Andi Kleen]

> Do you have an idea of the performance gain we could expect from this node 
> local pda allocation ?

I wouldn't expect very much.

> Say a CPU is on Node 1,  was a change in pda (allocated on Node 0) 
> immediatly mirrored on remote node or not ?

The Opteron caches are write back afaik - this means data only leaves
the L2 cache when other data pushes it out.
But the additional traffic on the interconnect was likely negligible. 

If anything I would expect the reduced latency when a user space program eat up all 
cache and the PDA is needed on the next kernel entry to be helpful.

But it's not very much at least on an Opteron because the NUMA factor
isn't that bad. On Kiran's machines which likely have a higher NUMA 
factor I guess it helps more.

-Andi
-
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/