If anyone else is out there (there is a bug w/ mailman at the moment,
so I can't see if anyone else has subscribed), probably the first
thing is to decide what metrics would be useful to report.
As a starting point for discussion, here's what I was thinking:
CPU Table: (per cpu)
cpu id (key)
minor faults
major faults
cross calls
interrupts
context switches
involuntary context switches
# of syscalls
% idle
% user
% sys
Filesystem Table: (per filesystem)
Mountpoint (key)
FS Type
Mount time
# of open(2) operations
# of close(2) operations
# of read(2) operations
number of bytes read
# of write(2) operations
number of bytes written
# of fsync operations
# of lookups
# of file creation calls
# of file deletion calls
# attribute retrevial calls
# of attribute modification calls
# of security attribute retrieval calls
# of security attribute modification calls
Since a lot of SNMP utilities support the notions of counters, rather
than doing a lot of computation or aggregration, my thinking is
present the data, and let those utilities (such as rrdtool) do the
work of figuring out rates or aggregating data. However, there might
be a few cases where it makes sense to present an pre-computed value.
