On 11/18/2016 06:30 AM, Robert Nichols wrote: > On 11/18/2016 07:39 AM, Frédéric Bron wrote: >>> While tmpfs does utilize RAM it will actually move contents to swap >>> in the event RAM is >>> actually needed. >> >> So that's normal that my RAM looks full (in top) and the swap and /tmp >> are empty. This means that I still have about 8Gb free RAM available, >> right? I am happy then. > > You have all 16GB available. That "8GB" for /tmp is just a limit. The > actual space used grows and shrinks as you add and delete files there. > Think of it like the ulimit for process size. Just because processes > might be allowed to grow to 4GB doesn't mean that all processes _use_ 4GB. > > There really isn't a lot of difference between /tmp in RAM and /tmp on > disk. > > With /tmp on disk, files are written initially to the buffer cache, and > get flushed out to disk when something else needs the RAM or, > eventually, by the automatic push of dirty buffer pages. For the latter > case, a copy of the data remains in the buffer until something else > needs that RAM. Short-lived files might spend their entire lives in the > buffer and never get pushed out to disk. > > With /tmp in RAM, files get written to the RAM pages and get pushed out > to swap when something else needs the RAM. > > The only difference is whether longer-lived files always get written to > disk regardless of the memory load. > > As for RAM appearing full, see <http://www.linuxatemyram.com/>. The main problem for using a RAMdisk-based /tmp is that a hell of a lot of utilities still write temp files to /tmp and with its limited size, you often get "filesystem full" errors. Before you start on me, I know the new guidelines say for "large" temp files the program is supposed to use /var/tmp, but the VAST majority do NOT do that and probably never will. I live in the real world, solving real problems and the first thing I do is mask tmp.mount on any installation. I'm somewhat surprised that they actually gave us a relatively easy way to get around this stupid idea. I wish it were as easy to sh*tcan systemd and journalctl (two other really moronic, overly complicated and badly implemented ideas). ---------------------------------------------------------------------- - Rick Stevens, Systems Engineer, AllDigital ricks@xxxxxxxxxxxxxx - - AIM/Skype: therps2 ICQ: 226437340 Yahoo: origrps2 - - - - When all else fails, try reading the instructions. - ---------------------------------------------------------------------- _______________________________________________ users mailing list -- users@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to users-leave@xxxxxxxxxxxxxxxxxxxxxxx
