differences in top's bookkeeping between 10.2 and 11

Erich Dollansky erichsfreebsdlist at alogt.com
Sat Dec 26 03:06:42 UTC 2015


Hi,

I do not know if this is of any interest. I have a test program which
starts 1000 threads which use a tiny amount of CPU time every second.
This all works fine, only top surprises me a bit.

FreeBSD raspberry0.alogt.com 10.2-STABLE FreeBSD 10.2-STABLE #0
r291083: Fri Nov 20 09:20:23 UTC 2015
root at releng1.nyi.freebsd.org:/usr/obj/arm.armv6/usr/src/sys/RPI-B  arm

reports:

load averages: 119.74, 76.30, 53.35

and

 1046 root       1001  40    0   165M 33664K umtxn    3:38   0.05% test

and idle shows values between 30 and 50%.

while

FreeBSD raspberry2.alogt.com 11.0-CURRENT FreeBSD 11.0-CURRENT #0
r291495: Tue Dec  1 09:13:20 UTC 2015
root at releng2.nyi.freebsd.org:/usr/obj/arm.armv6/usr/src/sys/RPI2  arm


reports:

load averages:  4.63,  7.64,  4.79

and

1360 root       1001  38    0   199M 47928K uwait   1   2:24  35.89%
test

and idle shows values around 80 to 90%.

The program reaches the hardware limits on 10.2 but top claims that it
takes less than 1% of CPU time. The 30 to 40% as claimed under 11
sounds much more reasonable to me.

I have read of top's bookkeeping habits but never saw it that extreme.

Erich


More information about the freebsd-arm mailing list