currently Linux user/system/nice/idle CPU usage
percentage is calculated using values kept since the
system boot.
which is wrong because with long uptime relative
changes to the values make minimal difference to the
percentage value.
this patch introduces regular sampling of values with
1-minute sliding average (similar to the approach used
for diskio).
this way current CPU load can be effectively monitored
via one snmpget.
the patch
Logged In: YES
user_id=88893
You're right that the current implementation of CPU percentages
under Linux doesn't make much sense. But this is the wrong
way of fixing things.
The whole idea of sliding percentages is contrary to the
SNMP way of working. All of the cpu "percentage" style objects
are currently deprecated, and probably should be withdrawn.
The "raw" counters provide a mechanism for calculating
percentage usage over *any* required period - not just 1m.
Providing a single-GET interface to these would probably
be better handled via the DisMan Expression MIB.