Hi, following up on recent statements regarding loadavg() calculation: I currently have a machine compiling a make release, using SCHED_ULE. I use WORLDARGS='-j4' and have observed the following: * During compilation loads gets up to 15 and stays there, but I feel this has been "set in motion" by top(1). The buildprocess already was an hour or 2 on it's way (set it up remotely) and the first time I ran top, the load was three. I quit top, and monitored slowly by calling uptime(1) about every minute manually (to avoid sleep()), but the load still went up. * changing top's refresh frequency to 5, had no impact whatsoever - load still goes up, finally changed it to 1 and kept monitoring. * during 'install' stage load went down to appr. 0.62 and kept lurking between 0.62 and 1.20, still with many make(10 and sh(1) processes, but no compilers. * during kernel build stage load went quickly up to 3 and 5, with less compilers this time. * it's now building ports and balancing between 1.10 and 3.50. I've used this machine more often lately to build dists, but never saw the load come above 6, using SCHED_4BSD. Last time was about 1,5 weeks ago. Kernel config and dmesg attached. Full command is: nohup make release BUILDNAME=5.2-idgnl-CURRENT \ CHROOTDIR=/webdocs/scratch/CURRENT \ CVSROOT=/home/ncvs DOC_LANG=en_US.ISO8859-1 -DMAKEISOS \ RELEASEDIRFILES=/usr/ports/distfiles WORLD_FLAGS=-j4 \ KERNELS='DBSERVER SERVER3' -DNOPORTREADMES & -- Melvyn ======================================================= FreeBSD sarevok.idg.nl 5.2-CURRENT FreeBSD 5.2-CURRENT #0: Wed Jan 28 18:01:18 CET 2004 root_at_sarevok.lan.webteckies.org:/usr/obj/usr/src/sys/SAREVOK_NOAPM_NODEBUG i386 =======================================================
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:41 UTC