On Fri, 5 Apr 2002, Nev wrote:
I do this already just using log files a perl script and a 5 min cronjob
just my 2 cents :)
That gives you information after the fact. And relies on two other
separate infrastrucutres; perl and cron - to run reliably and timely when
the system is in dire need.
The true power of build in SNMP instrumentation is detailed information on
what is happening -now- to see load surges and trouble *before* it is an
operational problem - and have people alterred before you break your SLA.
Ideally your agent is deeply tied into either the app or the kernel of the
system - has pre allocated all resources (so that a low on memory or file
descriptors is not going to stop it from having reliable info). As a bonus
- the whole network design of SNMP is reasonable robust and quite likely
to get through a congested network.
And then there is the little matter of push - i.e. active notification -
and the ability to reconfigure live; i.e. do an SNMP set to for example
enable/disable a costly feature - lock out an abusing IP address, scale
logging up or down and so.
All in all I would not want to mix logging-after-the-fact for audit
purposes and historic information with real-time information and real-time
management interaction.
Dw.
