| |||||||||||||
O_HOT and O_COLDO_HOT and O_COLDPosted Apr 26, 2012 4:39 UTC (Thu) by slashdot (guest, #22014)Parent article: O_HOT and O_COLD
It would seem that a much better approach would be to add a per-inode "block group range" field indicating in which block groups the filesystem should try to allocate file blocks, which would be inherited from the parent directory for new files.
This way, it would be setup by the system administrator depending on his knowledge of the hardware (or by an automated tool with heuristics), it wouldn't pollute the API, and it would be far more flexible.
(Log in to post comments)
O_HOT and O_COLD Posted Apr 26, 2012 7:56 UTC (Thu) by iq-0 (subscriber, #36655) [Link]
Or more generally some sort of "allocation cookie". The way this might be implemented might be, like you say, by allocating from some dedicated blockgroups on Ext-like filesysems.
But you could also envision a more generic "subvolume" like logic for managing this (allocation pools in zfs, certain block device subsets in btrfs, ...)
|
|||||||||||||