NUMA scheduling progress
NUMA scheduling progress
Posted Oct 1, 2013 23:27 UTC (Tue) by Karellen (subscriber, #67644)Parent article: NUMA scheduling progress
For example:
If a process calls linux clone() with CLONE_VM (e.g. starts a new thread) then the new process/thread is initially put in the same NUMA group as its parent.
If a process calls clone() without CLONE_VM (e.g. fork()s) then NUMA grouping becomes a "maybe" and could change in future based on the number of copy-on-write page faults it takes - which were going to happen anyway.
If a process calls exec(), it is taken out of it's current NUMA group and starts a new one.
If a process calls mmap() with MAP_SHARED, check for other mappings of the same memory block, and, based on the size of the memory block mapped and the amount of other memory already allocated by the process, at that point possibly migrate the process to an existing NUMA group containing other processes with the same memory mapped. (And consider removing a process from a NUMA group on munmap())
Why wouldn't/doesn't that work? (Or does it just not work *well enough*?)
The LWN site is currently under high scraper load, so comment display has been suppressed for anonymous users. If you are a human, you may read the comments by clicking the button below:
Note: you can avoid this step in the future by logging into your LWN account.
