Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
June 1, 2018
Acumos/ONAP Integration - Vision
• Vision & Goal
o Treat ONAP/SDC as a special case of Federation to Acumos (via E5)
o On Acumos side, modelers will create models that are technically feasible to run in ONAP environment (e.g. DCAE)
When those models are published, they will be labeled as, for example, “ONAP/DCAE Compatible”
o On ONAP/SDC side, user can browse the available Acumos model catalog
The user can search the desired model and download from Acumos
The model can be deployed and run in ONAP/DCAE environment with the proper data feed
The ML results can be displayed by the DCAE Model Runner, or through other application running in DCAE
• Challenges
o Since DCAE has/needs special requirements/info to run a Docker Image, therefore a “wrapper” (which provides
DCAE required functionality, such as HealthCheck so it meets the DCAE design standard) needs to be included in
packaging the image on Acumos side.
o Data collection and feed to the model in DCAE environment
The data are collected by ONAP/DCAE data collectors and fed using DMaaP. Connecting ML components
together with DMaaP will be part of the SDC (Service Design Creation) process, in which the message router
configuration will be specified (e.g. DMaaP topics).
2
Acumos/ONAP Integration - Requirements
• Requirements
o A wrapper that provides DCAE required functionality, such as HealthCheck so it meets the DCAE design standard
o DCAE Model Runner with data broker-like functionality that takes data from DCAE collectors and feed the data to
the model (because DCAE currently does not support protobuf, Message Router is used in DCAE Model Runner)
o Combine the resulting collection of components into a Docker Image file
o Design UI pages to support the workflow
• Design Options
1) Rebuild from the artifacts of the existing generic model (note: excluding composite or non-Python models) +
“Generic DCAE Python Runner”, and repackage them into a single image that is ONAP compatible. Collect DCAE
data format & component specification in tabular form via Portal/UI.
2) Allow the modeler onboard DCAE ready models, leave everything the way it is and mark such models (either with a
flag or keyword) as DCAE compatible at onboard time (dependent on the newly proposed Common
Services/Module effort).
• Recommendations
o Short Term solution
Option 1)
o Long Term solution
Option 1) & 2)
3
Protobuf defines format
Python DCAE
Model Runner
of input/output of model
CLAMP Model
Data Or Docker
Model Source Manual for Image
Model Rebuild Docker
Process now
Image
JSON, Protobuf, (Onboarding/UI)
jar file, E5
model.zip
Federate
DCAE data format &
component
specification
Modeler ONAP/DCAE DCAE data format &
Component Specs component
(Portal/UI) specification
Blueprint
Custom-build blueprint for now. Full
Design & implement this process automation requires ONAP/SDC R2
4
Acumos/ONAP Integration – Work Efforts
• Python DCAE Model Runner (done)
o A wrapper that provides DCAE required functionality, such as HealthCheck so it meets the DCAE design standard
o Consume JSON data, and convert to model-usable native format
• Model Rebuild Process (Onboarding) (done)
o A new API is exposed to accept requests to convert existing solutions into DCAE-ready solutions. Unlike existing
Onboarding APIs, this request does not include the needed artifacts for onboarding; instead, it includes a reference
to an existing solution within Acumos
o The Onboarding server fields requests to this API, looking up the solutions and pulling the needed artifact from the
Nexus server where they reside
o A very similar process to normal Python onboarding starts, but using the DCAE-specific model runner code from
Paul and Tommy, and a slightly different Dockerfile template
o When this process finishes, the results of Docker build (the DCAE-ready microservice) is pushed into Nexus, along
with copies of the other artifacts, forming the new DCAE-ready version of the solution
• Model Rebuild Process (Portal/UI) (done)
o UI
Design & implement UI pages to support the rebuild process
5
Acumos/ONAP Integration – Work Efforts
• ONAP/DCAE Data Format & Component Specifications (Portal/UI)
o Backend Process
Read model metadata, and populate fields under “streams” - subscribes, format, version, config_key, type
As for the data type, there are three options:
1) Take input from user - this option is manual and requires the user be DCAE knowledgeable about
specific data types available in DCAE data sources (Note: not all the DCAE in ONAP is created equal,
and the DCAE compatible model may be customized to work in a particular ONAP environment.
Need E5 to support restricted mode)
2) Generate the type automatically – this option needs more research and deeper understanding of
ONAP/SDC/DCAE (ONAP currently does not support data type registry, so this option may not be
feasible for now)
3) Prepopulate the field and allow user to select and match the type between model input and DCAE
data sources - this option also needs more research and deeper understanding of ONAP/SDC/DCAE
Access Repo, and populate field of URI of the docker image on UI page (same issue as option 2)
o UI
1) Ask user to provide DCAE type information (option 1 for now, options 2 & 3 TBD)
2) Allow user to enter parameters
3) Allow user to adjust the default interval and timeout values for health check
4) Prepopulate field of URI of the docker image in Acumos repo
6
Acumos/ONAP Integration – Work Efforts
• E5
o Allow ONAP to pull public catalog in Acumos
o Allow ONAP to pull model & model artifacts with ONAP/DCAE compatible flag from Acumos
• ONAP (SDC) (future)
o Allow ONAP users to view the models in SDC catalog via ONAP adapter
o Connect ML components together with DMaaP and specify the message router configuration (e.g. DMaaP topics)
• ONAP/DCAE Model Deployment (future)
o Build blueprint
Custom-build blueprint
o Deploy/Run ML model
CLAMP or manual
o Display/plot results
Plotter program to show graphical results
7
User Flow: Acumos/ONAP Integration – Option 1
Modeler ONAP User
Portal/
Validation Onboarding
Marketplace
Search & select model in catalog Given a model ID, retrieve artifacts from
Search Rebuild
by functionality, category, repo, and rebuild ONAP/DCAE compatible
modeler name, tags, etc. Catalog model model with generic DCAE Model Runner
Assume the rebuilt model View View model list & model
Vision & Goal: Treat ONAP/SDC as a E5 details from Acumos in
is in private catalog Acumos
special case of Federation to Acumos SDC Catalog via E5
1) Publish model to public catalog Catalog
Publish
2) Invoke validation and display
results model
Download Download/transfer the
Validate model selected model to SDC via E5
Set the flag for the model in the catalog
model that it is ONAP/DCAE compatible
Config, Deploy & Run the
Run security scan, license Config, model in DCAE environment
compliance, and key word Custom-build blueprint for now. Full
automation requires ONAP/SDC R2 Deploy & using DCAE Model Runner and
scan (assume auto & pass) data format & component
Run model
specification
8
User Flow: Acumos/ONAP Integration – Option 2
Modeler ONAP User
Portal/
Validation Onboarding
Marketplace
Onboard/package a model
Login to Acumos platform Login to Onboard with DCAE requirement,
and land on Home page Acumos model and label it as ONAP/DCAE
compatible