On 2012年07月02日 15:29, Rik van Riel wrote:
On 06/30/2012 04:23 AM, Nai Xia wrote:
Oh, sorry, I think I forgot few last comments in my last post:
In case you really can take my advice and do comprehensive research,
try to make sure that you compare the result of your fancy sampling algorithm
with this simple logic:
"Blindly select a node and bind the process and move all pages to it."
Stupid it may sound, I highly suspect it can approach the benchmarks
you already did.
If that's really the truth, then all the sampling and weighting stuff can
be cut off.
All the sampling and weighing is there in order to deal with
processes that do not fit in one NUMA node.
This could be either a process that uses more memory than
what fits in one NUMA node, or a process with more threads
than there are processors in a NUMA node, or both.
That is what the complex code is for.
Quote Andrea's words:"fully converge the load into one node
(or as fewer nodes as possible)".
And I think I've said several times this weight is
actually W = x * y form. I think I've made my points
clear.
Well, if you insist, I will keep silence.
--
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/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>