Posted Mar 23, 2012 18:46 UTC (Fri) by cheako (guest, #81350)
Parent article: POSIX_FADV_VOLATILE
I believe that the kernel should reclaim the entire range, however I also think that a big application would have several levels of cache and in the case of tabs each tab might have a cache.
I don't know what effect having many small POSIX_FADV_VOLATILE ranges would be, but an application could clearly make decisions that would pool small caches into one bigger in order to keep the total number of POSIX_FADV_VOLATILE ranges reasonable.
Perhaps exposing the average lifetime over the past five minuets of any POSIX_FADV_VOLATILE range would help an application decide if it should even bother. There can be other performance counters that would help an application determine the number and target size of