19

The information expressed here is the personal opinions of the presenters and the comments here should not be taken personally During the course of this

  • View
    213

  • Download
    0

Embed Size (px)

Citation preview

The information expressed here is the personal opinions of the presenters and the comments here should not be taken personally

During the course of this discussion, we were forced to quote a few paragraphs from various publications so bear with us when we read them out

Err… did I calculate the values correctly?

Authors: Hitesh Ballani, Paul Francis

Presentation By: Rahul Potharaju, Rejaie Johnson

Advisor: Aleksander Kuzmanovic

One of the Motivations: Enterprise Networking and Computing: The Need for Configuration Management

Published: January 2004 4 years back

Cited by: 100502?0!

Anyways, what is the other 20%?

Which IT industries were surveyed?

62% Configuration errors?

Hypothesis: A message expressing an opinion based on incomplete evidence

Developing the system without knowing its interface?

… we agree with their basic hypothesis and in this paper assume the presence of a management channel.

• Cannot dictate where the management channel operates or does not operate of the same physical links. Why? Would it be more beneficial if the management channel takes more control of this.

• How can this be done?

• What makes this better than the other research areas?

• Gaps can be filled in with “other” work. Why can’t it be your work?

However, there is the vast body of literature that does deal with protocol implementation, i.e. through abstraction...

• Does not specify as to what the data plane does.

• Only talks about network management, but rarely talks about how things are going to be forwarded or routed to one network to the other.

• What about cryptography? How will this work?

• Not a flaw but is your script inspired from a particular language?

• If this is going to be adopted, will other networks be able to adapt to this?

• Paper does forget to mention about ISP’s.

How are these abilities monitored? Local or Global Algorithms?

Introduction of unnecessary(or necessary but again, adding extra problems) tussles?

There are so many issues to be addressed. For instance:

If this is extended to global usage, hot potato routing becomes dominant!

Modules and pipes report on their performance with thesemetrics. They can also advertise the ability to offer performance trade-offs in terms of these metrics.

… there are scenarios where the NM will have to deal withprotocol-specific details.

Ok… lets accept it. But how many will NM deal with? With a protocol outburst, NM would soon fade away.

Take Skype for instance – Would they give out the protocol specifications?How would you deal with it then?

Did the authors miss that out or was it just unnecessary to deal with it?

• VPN – Single Enterprise Domain – All is well… Looks like a “frog in a well” scenario. How would you account for Multiple Domain managements etc..?

• Again, VPN is being presented because it is well suited for their problem? VPN doesn’t involve much of the headaches like various parameter tunings etc…

Configure connectivity between the customer-facing interfaces <ETH,A,a> and <ETH,C,f> for traffic between C1-S1 and C1-S2.

Drop connectivity between the customer-facing interfaces <ETH,A,a> and <ETH,C,f> for traffic between C1-S1 and C1-S2.

Maybe it works only in VPN scenarios?

Let’s say you managed to get CONMan deployed – Now, if there is a change in one of the low-level protocols, how do you plan to propagate the changes?

Would NM be a central point of failure?

CONMan should be highly flexible/adaptable/secure?

To incorporate the never ending list of low-level

protocolsTo suit various enterprises Enterprises rely on

CONMan!