On 04/24/17 04:42, Christoph Borsbach wrote:
Hello everyone,
first off: I know that the topic of "load" has been discussed numerous
times, and been a topic on undeadly [1]. I know that this number is not
that important.
However:
After upgrading 3 of my systems to 6.1 (from 6.0) I noticed the load
average (15min value) has gone up by roughly 1.0, both in the output of
daily(8) over some days now and when checking manually with w, top, or
uptime.
The systems in question differ a bit:
- amd64 MP (KVM-Guest, dmesg [2], load-example [3])
- amd64 SP (VMware Guest, dmesg and examples not handy right now)
- i386 SP (Alix, dmesg [4], load examples [5])
All were upgraded last week with bsd.rd to 6.1-RELEASE. The systems
perform as well as ever and nothing was changed aside from upgrading
system and packages. I'm just interested what could change the behavior.
A quick check of src/sys/uvm/uvm_meter.c does not show me any changes
recently.
Has anybody observed this as well and has an explanation for this?
Thanks,
Christoph
Christoph,
What has changed 6.0 - 6.1 is the entire operating system. uvm_meter.c
may not have changed but the other sub-systems have, which effects
the way things works. It's the same with playing mp3's and you get
stutter (or not) when disk I/O or other things are in play.
Any OS is a city; largely invisible to us, interactions go on that can
have ripple effects in how things work. The concept of a load average
is nebulous at best. You can spike the system averages any number of
ways so using it to determine how busy the system is at any point in
time is not great. Better to see how fast the system delivers web pages
or files, or ...
Perhaps the uptime / w documentation should explicitly say that
comparing load avs on different versions is a bit like comparing apples
to spark plugs.
--STeve Andre'
