Discussion items 

NOTES

Scott Blandford - working on updating the Casablanca Scale Out Use Case page.

Ron Shacham - started a discussion on Beijing Pairwise Testing.

• CLAMP has been doing pairwise with SDC/Policy to test the distribution and policy creation. Some issues on Policy API, CLAMP is working with Michael Mokry. Mike is busy with AAF issues but should go to him for any issues and if he cannot help Pamela Dragosh will find another Policy SME to help CLAMP team.
• Gervais-Martial Ngueko - noted that the way the scope is used to group policies by Control Loop. In the Policy UI, you can see the policies grouped together. The scope=Control Loop Name. Not clear what is the Policy ID sent to DCAE once CLAMP has created one. Is it the Policy Name without the Scope?
  
  • Ron Shacham shared an email that highlighted the confusion.
  • Pamela Dragosh shared that the current API is not going to work long term. Scope was originally intended as name/value pairs for refining policies, but the implementation became more of a gloried database call. Cannot fix that immediately due to timing. The policy team hopes to have a better API starting with development in Casablanca and perhaps finishing by Dublin.
  • Alex Shatov - just having CLAMP in the scope is good enough for the matching condition in the Policy API for the DCAE Policy Handler to get the appropriate you need.
  • Pamela Dragosh talked about how that would work for Beijing but it required DCAE to sort through and make a decision as to which component gets a policy. It would be better in the future to have more refinement for the policies for a Control Loop so it would be easier for the DCAE Policy Handler to be able to query from Policy and then distribute to the appropriate DCAE micro service instance. Policy makes the decisions while DCAE does the enforcement.
  • Discussion around the "Config_" being in the policy name. What is needed is policyScope start with "CLAMP". Scope prefix is rough filtering for DCAE Policy Handler, then deployment handler determines where to deliver the policies. Alex will reply via email the requirement from DCAE to CLAMP.
  • Ron talked about Lusheng Ji setting up an entire DCAE deployment in windriver lab that is now available. CLAMP is starting testing in that lab. SB-07. Pam indicated this is where Marco Platania is doing vFW/vLB testing. Ron indicated we are only exercising deployment piece.

Pam asked Randa Maher for status on APPC Daylight issues to do pairwise testing. Randa able to merge Nitrogen, but having issues with A&AI. Team is still troubleshooting. There are issues with nexus and windriver being upgraded. Waiting for a fix from CCSDK for APPC builds to work again. Done some testing successfully using Carbon.

Scott Blandford led discussion on Scale Out Use Case:

• Wiki is located here for Beijing: Manual scaling functional requirement
• New wiki page for Casablanca is being worked on here to match HPA Use Case Template: Scaling Use Case Extension. Scott talked about minutes from Scale Out Use Case meeting:
  • Policy make a simple request to SO, SO pull from A&AI and other systems.
  • Can SO get all the information it needs to do auto scale out? Need Steve Smokowski for help on this
  • Need to be sure CLAMP has what it needs at Design time to create the policies. Ron confirmed the information will be there in the CSAR, eg. the VF Module Type.
  • Being sure that while another instance is being scaled out, need to hold any other scaling requests until that request finishes. Need Policy Guard policies defined that discussion should start around.
• Scott working on the template and specifically the goals for the Use Case. Scott needs to talk to China Mobile team regarding the VOLTE Use Case use the UseCase UI and VFC. There is some overlap with regards to auto scaling and would like to work closely with China Mobile and Huawei on this.
• Scott shared the auto scale out work flow using Option C (per his slides). Showed what needs to be done for Casablanca and the overlap with the manual scale out that was done in Beijing.
  • What triggers the Scale Out? Where does customer ordering come from and go through? These are stretch goals to be pursued beyond Telemetry triggering the scale out. Unsure if appropriate for Casablanca. Pam recommends using PlantUML to help depict these flows.
  • Question: Where the decision for which location to Scale? Scott, we are scaling a component of the VNF. The state of the environment the VNF must be in the same site. We'd like to consider if the site is full can we pick another site? Another stretch is service scaling to another site. There is still a whole list of things needed to go beyond the basic scenario. How to automate the features you can do in VID?
  • Data in the Scaling Request: vnf Instance ID and VF Module Type. Both are available to CLAMP/Policy, should be able to find those details and put those in the Policy. Pam would like to show some more detail on that in the Operational Policy.
  • Question: Use the guard to limit the number of scale ups and/or the scale in? Pam we should focus on the scale up (NOTE: Pam will look into the scale in guard policy for Casablanca)
  • What does a scaling policy looks like? Bandwidth, CPU Utilization, what other areas? Which service component gets scaled vs the component that identifies or triggers the Control Loop Event?
  • How do we capture the Common Actor API? Should it be functional requirement sponsored by the CL Sub committee and use the Scale Out Use Case to help drive that? Requires changes to VID/SO and subsequently VFC. We want simplification in the API, enough information for the Controller to do its job balancing helping VID/Policy keep the calls simple. Pam needs help from the community to help drive this: Control Loop Sub Committee to help drive functional requirements. Can we put together a sub-team for this? Need the whole ONAP community to work on this. When companies bring ONAP into their production they may want to extend this. Pam's presentation regarding this: https://wiki.lfnetworking.org/display/LN/ONAP+Project+Specific+Breakouts?preview=/327933/328253/ONAP%20March%202018%20-%20Control%20Loop%20Sub%20Committee.pdf
• Scott: uses both the Scale Out Use Case call and the Control Loop calls to help coordinate all the aspects of this Use Case. Pam: Would like to get on the SO and VID weekly meetings to further discuss Common Actor API with those teams.
• Scott: Where does DCAE end and Policy begins? Pam: The PlantUML would help with that.

RECORDING