Hi,
In the past I have mainly used the performance monitoring events provided by
the CPU's performance monitoring units. It is not uncommon for particular
events monitored by the hardware to be inaccurate (for example BT243
Performance Monitor counters May Produce Incorrect Results in the Intel Xeon
Processor E5 Family Spec Update.
http://www.intel.com/content/www/us/en/processors/xeon/xeon-e5-family-spec-update.html).
The hardware may over count or undercount, or in some cases not count events
at all. Thus, blindly trusting the metrics provided by the performance hardware
is unwise. When I started using some of of the performance metric provided by
PCP I had the same question. How, are the PCP metrics checked to see if they
are accurate/reasonable? For some of the PAPI (http://icl.cs.utk.edu/papi/)
testsuite there are cross checks to see that values agree with the expected
values.
For estimating network utilization using PCP the network.interface.baudrate and
network.interface.speed metrics should provide some indication of the interface
speed. However, looking at the numbers produced by pmval below they don't seem
to be reasonable for 1Gbps ethernet connection of em1.
[wcohen@santana Downloads]$ pmval network.interface.baudrate
metric: network.interface.baudrate
host: santana
semantics: discrete instantaneous value
units: byte / sec
samples: all
wlp3s0 lo virbr0-nic virbr0 em1
? ? 316871376 ? 1622366528
? ? 316871376 ? 1622366528
^C
[wcohen@santana Downloads]$ pmval network.interface.speed
metric: network.interface.speed
host: santana
semantics: discrete instantaneous value
units: Mbyte / sec
samples: all
wlp3s0 lo virbr0-nic virbr0 em1
? ? 7.813E+04 ? 7.813E+06
? ? 7.813E+04 ? 7.813E+06
^C
Are there some PCP tests that check that some of the metrics are sane and
reasonable?
-Will
|