On Wed, Feb 10, 2016 at 4:44 AM, Noah Misch <n...@leadboat.com> wrote: > Killing them was the right decision. Thank you. Like all my cfarm batch > testing, the processes had setpriority(PRIO_MAX). Perhaps I/O load remained > high enough to ruin things.
Unfortunately, process priority is absolutely worthless on machines with hyperthreading. The other day I witnessed someone using all 64 virtual cores of gcc110 (the POWER7 machine), and it was godawfully slow to work with. My rule of thumb is to leave at least one physical core free in order to not disturb other users. So if the machine is 8x8 core (8 physical, 8x HT -- I don't know the real topologies of these POWER machines), I strive to use at most 7 cores. _______________________________________________ Gcc-cfarm-users mailing list Gcc-cfarm-users@gna.org https://mail.gna.org/listinfo/gcc-cfarm-users
