Page Comparison

1. Netconf mount on Honeycomb is working. Getting 404 error for netconf config message.
2. Have to test for Deny message for second CL from Policy.
3. Lack of extra VM for RAN Sim. Can have limited number of RAN Sim nodes.

1. Dependency on C&PS - C&PS team has consensus that initial implementation in R7 will meet needs of SON use case.
2. Dependency on ORAN - In order to prevent major code changes, we need pre-standard version of O-1, VES formats, yang models
3. Policy - Include Defer from CLC (high priority), Separate Drools instance (depends on Policy)
4. Modeling - Cell lifecycle - Assume cell addition etc is not part of ONAP. We can assume that data on list of cells is available to C&PS.
5. Incorporating ML-based use case. See last item.

Thanks to Vijay and Shankar PN who are PTLs or DCAE and OOF who joined the call. 
We discussed high-level guidance about incorporating ML-based SON use cases.
Consensus was that training of models should be done outside ONAP, and likely to be offline.
ONAP use cases should focus on demonstrating how an ML-trained model can be onboarded. 
This can be done as a recommendation model in OOF, and also as a DCAE MS which leverages work  done
in Frankfurt for the Acumos-DCAE Adapter. See link: Acumos DCAE Integration 

Guidance for new use case discussions:
Generation of data and training of ML-model must be done outside ONAP separately.
ONAP SON solution should include: pre-trained model, source (e.g. enhancement to RANSim) of data needed to apply the pre-trained model.

See this paper for a discussion of SON use cases and references for work applying different ML approaches to SON problems.
https://arxiv.org/pdf/1707.09300.pdf