Timothy Murphy wrote:
I've been looking (not very hard) at RRD (round-robin database),
and I notice that several examples displaying things like memory usage
use snmp (more precisely snmpwalk) to gather the information.
I'm just wondering if this still makes sense.
Most of the information seems to be available in /proc ,
and I wondered if it is just conservatism
that leads people to keep on with snmp ?
You are making the assumption that RRD/MRTG or any other tool doing the
monitoring would be on the same box you are monitoring. This might make
sense in some instances but in general you would not want to spend the
resources needed to do full-fledged monitoring on a box that is supposed
to be running some calculation, or database or dedicated to some
user/useful function.
Along with that you would not be able to (easily) do event correlation
or comparisons of the data from different systems. Generally you would
put the network management functions on system that are dedicated to
that function. Consider a system that becomes unreachable; is it the
network interface of the system or some other device? How do you handle
all the events that are generated if you did not centralize the network
management function?
I've even seen infrastructures that have a separate network entirely for
out-of-band network management. This provides (1) a dedicated
environment so that the NMS (network management system) traffic does not
interfere with production traffic and (2) a way in to the system should
there be a catastrophe on the production network. Although I'm not sure
how common this is any more.
The issues of memory and disk space are just part of the NMS problem,
sometimes even secondary. SNMP provides a standard network based
interface to gather data from disparate systems, too. Think of a Windows
server that does not have /proc. All you need is the MIB and you can
monitor it just as you would your Linux box. Same would go for special
purpose systems like routers, switches, and firewalls.
Hope this helps.
bk
Timothy Murphy wrote:
I've been looking (not very hard) at RRD (round-robin database),
and I notice that several examples displaying things like memory usage
use snmp (more precisely snmpwalk) to gather the information.
I'm just wondering if this still makes sense.
Most of the information seems to be available in /proc ,
and I wondered if it is just conservatism
that leads people to keep on with snmp ?
