Re: Current challenges in Network Management

[Fred Baker <[e-mail]>]

Ramakrishna:

The truth is that there is not a real good definition of network
management.  That has been discussed, on this list and elsewhere, for
some time, and (to my knowledge) is about as well decided as the
definition of Artificial Intelligence.

There is real good definition of two approaches to monitoring devices
in the field (the use of periodic get-next sweeps, and the use of
asynchronous alarms), both of which are implemented in SNMP and have
their supporting and detracting schools of thought.

There is fair definition of the predictive aspects of Network
Management, in terms of building a database (using get-next sweeps at
well defined intervals, and potentially with some summarization of data
by the agent), using normal statistical techniques to determine
trends, and from them to make recommendations to bandwidth managers.
I see some challeges in getting accurate information from devices,
and in making sure that our MIB definitions are the best.

There are some real problems in the realm of reliable private secure
parameterization of systems.  When somebody installs a default route
sucking all my traffic to his system, how does the system REALLY know
that it's the GoodGuyNET operator as opposed to DeadlyEnemy Inc?  When
someone changes the IP Route Mask for some destination to a new value,
how do I know whether perhaps two different managers are trying to do
similar things and overwriting each other?  There's a seperate list to
handle part of that question, and you might be interested to get on
it.  Start out by getting the internet drafts:

	draft-ietf-snmpsec-admin-00.ps
	draft-ietf-snmpsec-admin-00.txt
	draft-ietf-snmpsec-mib-00.txt
	draft-ietf-snmpsec-protocols-00.ps
	draft-ietf-snmpsec-protocols-00.txt

There is also considerable lack of science in user interface
methodologies.  Having a map of the network on the screen, painted
different colors depending on current status, is a real good approach
for immediate problem detection.  Graphing displays are good for current
state counter/gauge monitoring.  Fancy editors can help with trouble
ticket systems.  But I don't know of any NMS vendors who have the gall
to say that their solution is at the pinnacle of usability.

Beyond that, I think that we as an industry are still groping to know
what the right questions are, and answers are further away.

Fred