Posted May 3, 2012 9:07 UTC (Thu) by Klavs (subscriber, #10563)
Parent article: O_HOT and O_COLD
IMHO it will never work to let developers designate which files are HOT/COLD - as the article also suggests :)
This should be something the OS should handle - and there could be tools for the sysadmin.
I'd very much like it if the this feature could also support f.ex. distributing data, so if I f.ex. build 1 logical volume, consisting of 1 types of storage: SLOW (a few 2TB disks) and FAST (several smaller disks) - that way the HOT data could be kept on the FAST part, and the "bulk data" part - which often sits unused, can be put on the SLOW part - without having to shuffle data around (as we do in SANS today) and "intermix" mount points etc.
If the logistics for designating HOT/COLD data got good enough, one could even imagine a case, where the sysadmin could get an overview of hot "filled" the HOT part was - so he knows when to add more small/fast disks, and when to just add more slow large diske.
It would definetely help a lot of usecases (for server use obviously :)