Hello, I apologize for not replying to this earlier but I wanted to wait till the documentation that I had written was available. Roman Lapacz wrote:
-------- Original Message -------- Subject: Re: [Nmc-wg] Netconf in pS Date: Thu, 21 Jan 2010 09:39:45 -0500 From: Jason Zurawski <zurawski@internet2.edu> Reply-To: zurawski@internet2.edu Organization: Internet2 To: Roman Lapacz <romradz@man.poznan.pl> CC: nmc-wg@ogf.org <nmc-wg@ogf.org> References: <4B5701F4.3060607@man.poznan.pl>
Hi Roman;
Thanks for providing this, pretty neat. Also for reference the NETCONF RFC is here:
http://tools.ietf.org/html/rfc4741
I'm sending you some info about the investigation work (by Arne Oslebo) on Netconf in pS which is being done in Geant3. After a while I will add a detailed description from one of project documents but it's still under review process so it must wait a bit.
An obvious issue that the author has pointed out to not use NM-WG/NMC-WG is the lack of documentation both at a protocol and service level, which is a failing for the pS consortium in general. Hopefully this will be rectified soon through this group.
There are actually three issues with NM-WG/NMC-WG that we point out: * Lack of proper separation between information model and communication model. * Lack of generic information model for MA and MP. * Lack of proper validation mechanisms. For more details please see the document that Roman sent to the list last week.
To comment on the subject matter itself, I am having a hard time justifying in my mind why trying to use a protocol designed for a similar (but not identical) task is a good idea. In NETCONF's own words it:
provides mechanisms to install, manipulate, and delete the configuration of network devices
This touches a gray area in perfSONAR that has been the topic of DICE meetings in the past, namely managing the service configuration vs the measurement configuration. I think there may be a place in perfSONAR for this protocol, but I don't believe that exchanging/storing the actual data with the NETCONF primitives makes sense.
I know that the topic of what NETCONF can and can not be used for is a controversial topic, even within IETF. Things are however starting to change. For NETCONF to be used as a generic network management protocol it needs proper access control. I attended IETF this week and a private draft proposing an access control model for NETCONF was very well received in the NETCONF working group and it was decided that this is something that the working group will start to work on. So while IETF still has not made any official statement, people are now openly talking about NETCONF becoming the next preferred network management protocol instead of SNMP. This is part of the motivation behind our work. We believe that perfSONAR basically is distributed network management in a multi domain setting. Looking at the data that is stored in the various MA and MP implementations we don't see any fundamental difference between this information and what you typically collect in normal network management. So if NETCONF can be used for normal network management, why can it not be used in perfSONAR? Our simple prototype shows that it can be used.
I will end by noting a reply that Martin has been known to give in the past: "its all just XML!".
I agree completely but would like to turn this argument around. Since there already exists XML protocols for transporting data from A to B, NETCONF being one of them, why spend so much time and effort defining yet another protocol? After all "its all just XML".
Maybe we can see a demonstration at the next developers meeting to see how valuable this could be. We had I demo site up and running, but I see that it is now giving error messages. I will look at it next week and try to fix it.
I'm currently working on a more stable implementation that will also simplify the information model. The deadline for this work is in June but I hope to have it finished before that. Best regards, Arne