Hullo
many computer measurements assume a fixed performance cpu (eg top, ps,
sar). As far as I can see, not only are the domain models of these
tools being updated, but I cannot even find out what their existing
measurements mean in the face of a varying cpufreq.
Can anyone point me at anything that will help me understand how
variable cpufreq impacts the metrics, other than reverse engineering
the source code and experimenting with the binaries?
A simple example, if I'm trying to work out how much spare capacity
I've got on a compute server, I need to know how much I'm using now.
If I start with, say, top, I can see that process x is consuming 25%
of the cpu and that 30% of the time the server's in a wait state.
What's not obvious is what these are percentages of. It's not even all
that clear that they are consistent between the total values and the
individual values for each process. To make a robust process, I'm
going to want to use something that collects continuous samples, but
I'm not sure that sar collects the relevant cpufreq for each sample
period (nor whether such a cpufrequency can be measured or estimated).
virtualisation complicates the picture further, as cpufreq can be
considered more of a continuous attribute, rather than discrete.
Any thoughts on where I can get some baseline information on how these
different perspectives hang together?
tia
Tim Coote
--
To unsubscribe from this list: send the line "unsubscribe cpufreq" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
