site_archiver@lists.apple.com Delivered-To: darwin-dev@lists.apple.com On Saturday, October 2, 2004, at 10:58 pm, Dylan McNamee wrote: A simple experiment should validate this hypothesis: try it, and drive some paging! ok thanks, ben. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.appl... Ben Dougall asked: with mmap once you've read in some of a file and made some modifications is there anyway to avoid those modifications being swapped out to the backing store? what would be preferable is in the event of the system needing more memory, rather than swapping out those modifications, save them directly to the original file -- is that what happens now, by default maybe, or is there anyway of doing that? While I haven't examined Darwin's code lately, I am pretty familiar with the Mach's VM code from years past, and my memory says that what you want to have happen is supported. (At pageout, "dirty" pages are written back to the file backing the mapped region, which is a file-backed Memory Object). Revisiting the mmap man page reminds me that this might only be the case if the region is mmapped including the MAP_SHARED flag. right, it'd certainly make sense if that is an option as otherwise, if the modified data is definitely for saving, it could easily get saved to the drive, then loaded again, unnecessarily, then saved to the file. rather than just straight to the file. This email sent to site_archiver@lists.apple.com