Johansen gave a good reminding that the RMA possibly was not a root cause for the application running slowly even if it has high RMA, and provided detail information for why some memory would be allocated on remote node. Thanks.
>From my point, numatop had better still focus on within the scope of memory access and memory allocation, whereas in some cases, they are not the root cause for low performance. For what does the numatop phase 2 look like, it's still not a clear picture for me yet. I should dig into the application tuning by using numatop prototype, dtrace and other tools to find the clues and filter out the common parts. During the tuning, I will go ahead with the suggestions from Eric and Johansen. It will start after the numatop prototype is ready and has stable quality. Thanks for the votes from Eric and Johansen. :) Jin Yao -- This message posted from opensolaris.org _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
