Speaker was Kenneth Vastola, Department of Electrical, Computer and Systems Engineering
Rensselaer Polytechnic Institute.
|Questions should be addressed to firstname.lastname@example.org.
Main topics are network sensing and network management.
The goal is to tackle problems such as routing, congestion control, Medium Access Layer (MAC) protocol through network management.
Why is this hard?
Speaker proceeded to explain several Management Information Base (MIB) variables in detail.
It was emphasized that the SNMP primitives allowed for no processing, only collection of data.
Hence, network management is really network monitoring with human interaction required.
Speaker suggested that mobile agents could/should do network sensing and management
The reason for this is to figure out jumping characteristics such as:
One approach is pinging by agent system ("Do-it-yourself"). Issues with this approach are big data loads, waste of processing resources if narrowly focussed on agents, and scalability.
An alternative approach is to use the MIB data ("Generic"). The idea is to use a Generalized Likelihood Ratio estimator to gauge the state of the network and anticipatefailures and degradation. This approach has been proven to work well empirically.
Discussion (in chronological order):
Q: Couldnít TCP/IP round trip estimation be used for the purpose of network sensing?
A: The writer missed the answer to this question.
Q: What are the type I and type II errors of your estimation models?
A: 1% and 40%, respectively.
Q: Will IPv6 or ATM obviate need for Network Management?
A: On the contrary, need is greater than ever, because Quality of Service (QoS) demands it.
C: Speaker seems to have neglected body of research on mobile agent network sensing: Management by delegation and network management by mobile agents.
Q: What do mobile agent people want to know about the network?
A: From Node A, you would like to know the bandwidth and latency between nodes B and C. Additionally, being able to check on service availability would be desirable.
C: Hewlett Packard Labs use federated distributed network monitoring successfully.