Schedule, agenda, and minutes summary/details for meetings of the Acumos TSC Community Committee.
Weekly Meeting Logistics
Zoom 2: https://zoom.us/j/9266141474
Tuesdays 9 AM ET
will be led by Reuben Klein
- Confirm Schedule Change for July
- Update Plan for Documentation
- Discuss agenda of Release Planning call on July 13
- Talk about recording and prioritization of features
- Review Demeter backlog for Point Release 1 requirements
Today's discussion points:
Christina Harter, CMP
LF Events & Program Management
The Linux Foundation
T: +1 (817) 308-5838 (PST zone)
Free tickets are available if interested (worth 50 USD)
Virtual booths are available:
Proposed to move this meeting to 9:00 AM ET in order to get more folks from China and/or Europe
100% of participants voted yes
Reuben Klein will update the calendar in Lists.io - Jacqueline Serafin is a good contact to ensure we are doing it right.
Release schedule proposal from Natarajan Subramanian
Sprint 4 will end in May
The project will Issue 2 add'l maintenance releases on Demeter
Propose that we plan to release Elpis around June of 2021
Trying to do one major release each year
Trying to avoid a tight schedule during the low productivity months (Aug and Dec).
Plan for maintenance releases to take 90 days
for Demeter.1 and Demeter.2
Is this in conformance with LFAI policy? (2 rels per year is the norm, this would be 3 releases 1 major and 2 dot releases)
Reuben Klein can raise this with LFAI TAC
We should influence the TAC to limit project acceptance to those projects which are willing to build interoperability with other projects in LFAI
Reuben Klein is the new AT&T TAC member
Demeter ETE1/2/3 ends this Friday
Demeter ETE 4 will be a two week sprint
Starts next Monday
Focus on critical defects. No new features.
We will begin working on branching on
Demeter announcement should be by
Jessica Kim raised the issue that no nominations have been submitted for TSC Chair by the deadline. What should we do?
Anwar suggested "taking down" the nomination. This must mean postponing the election for a time.
Normally the vote is extended for one week in this event.
TSC chair must be the first election of the year, so this is a critical milestone.
In addition to current backlog, Farheen Cefalu would like to submit a series of UI user stories that should make a big diffference.
Murali Vuppari suggested opening all necessary Epics in JIRA but do not assign them.
This will put them into the backlog for the next release
The license management PoC revealed a number of features that are needed for the License Management plan to work
Michelle Martens will document these. They should be prioritized for the next release.
Should we change the schedule of the Elpis release?
There are dead periods (summer months and December-January) that make it difficult to get anything done.
Perhaps we look at 9 months as a schedule goal (March instead of November after a 6 week break at end of year?)
add 2 months to Demeter maintenance release (Demeter.1 defects primarily) and then
begin Elpis and have it run the next 6 months.
Patch releases (Demeter.2 for high visibility defects/no new features) could be on an independent schedule running in parallel to the next major release
Elpis could focus on resolving the backlog
Murali Vuppari can review some of the backlog and identify which JIRA items are relevant and which are not.
Then we can prioritize the important stories
This must be completed in May
Another important area is to complete the migration to cloud service
Try to get feedback from the community on what to emphasize
Identify opportunities to team up with other LFAI projects
Integration of ONNX requires more Python resources
Import ONNX objects
Enhanced model runner for ONNX
Introspection to create protobuf
Create ONNX microservice
We should pick this up as an opportunity to increase the capabilties
How does ONNX runtime fit in? Is it part of ONNX O/S or a separate project?
Common development/test environment for the community
Currently only TechM and AT&T can work in a common environment
Our Azure development environment will go away at the end of 2020
We need to find somehting more universal and available to the entire community
perhaps Msft would be willing to donate an environment
They are a supporter of the ONNX community
We would need to make sure that Acumos does not become an Azure specific tool
We are still in need of a Release Manager to take over for Natarajan Subramanian
Looking for volunteers
This is an appointed role
Can Huawei, Ericsson, Nokia, Orange step up? There are other companies in LFAI, but we don't see any interest from them.
The name Elpis has been chosen for the "E" Release: Release Page
We will list requirements on this page - there is a large backlog built up in the last 3 releases. This backlog should be scanned for Elpis Epics
Demeter release is scheduled for May - Elpis work will begin around Memorial Day (end of May)
Nat is not able to do the release management so we will need a volunteer to lead this effort
This is not an elected position - it must be someone with the experience and time to do this job
We should draw from other LFAI efforts and teams to enhance Acumos features
MLFlow - Model target management
SoaJS - CI/CD (is it easier to improve our Jenkins scripts than to incorporate SoaJS?)
We need a way to organize a discussion of alternatives and strategy. Many members of the community have experience with a range of tools
This discussion can then be used to set up Elpis requirements
The Elpis release should not be too ambitious from a feature perspective
Less development/more integration
We need to organize leadership on this to provide a direction
We need experience in potential tools which are integration candidates - this should not be a wide net. We want consensus around one-two candidates
Less interested in the perfect component than one that covers 60-70% of the functionality. but cuts our development time significantly
Michael Rooke suggests that we concentrate on Marketplace aspects of Acumos
Reuben Kleinthinks deployment to multiple environments with CI/CD is important
Demeter Sprint 4 is scheduled to complete this week
Workbench team is somewhat delayed (2 weeks?)
License management is somewhat behind as well
SoaJS is also an issue at this time
We will not have to move the release target date, but the testing schedule will be impacted.
If too many items come in at the last minute it may create an unsustainable timeline
We may have to compromise on features if we cannot maintain our schedules
Schedule time at next meeting to assess where Demeter stands after completion of Sprint 4
Use Case Leads need to get a description and scope written up
Please go to the Use Cases page and provide a good additional details
Follow the Link to your use case and add a clear description of a process that we want to demonstrate
This should describe an end-to-end scenario which captures a complete function of Acumos (not just a feature)
The Scope should lay out the steps of the use case and what they accomplish
The Timeline should lay out objectives culminating in a demo to the Community Committee at the end of May (when Demeter is released)
The demo will be recorded and put onto the Wiki as a training item.
The Dependencies section should identify the other use cases, JIRA items and resources that you are depending on to do the demo
You can also show which other use cases depend upon this one.
Why are we building these use cases?
The idea here was to create an end-to-end scenario which demonstrates a meaningful application of Acumos.
This should demonstrate a real world value and not just an abstract feature.
The process identifies all of the features needed to accomplish a real-world task.
It demonstrates that each dependent feature cited in the use case works correctly
It demonstrates that the individual features that have been built all work together
This reestablishes a practice that was implemented pre-Athena in which end-to-end scenarios were created describing not just features but end-to-end flows
These use cases can be use as demos of the new capabilities introduced with each release
Ideally there should be use cases that incorporate all the priority features in the release.
If not, then why were those features considered high priority? They ought to have a real-world value that can be demonstrated.
The list of approved use cases have been compiled on a single wiki page. each with a link to an individual Use Case Project Page
We reviewed the Use Case project pages and identified the sections that need to be completed
It is the responsibility of the Team Lead to assemble a team and list them on the Project Page
It is the responsibility of the team to identify the scope of the use case and set a schedule for completion of milestones
We will send out email to the teams that have been identified asking them to fill in the pages as they see fit
They should set up team meetings and teams should report progress from time to time to the Community Committee
This can include demonstrations of the use case as far as it has been completed
It is the responsibility of the team to identify any missing dependencies and set up JIRA items accordingly
Each project should be reviewed with the Community Committee at least at inception and at completion
We will review with the TSC to find out whether linkage of release dates and use case completion should be made.
Michelle Martens spoke about RTU editor and applying it to Acumos - can portal launch the RTU editor?
Should we have a License Admin role?
The Lic. Admin would be responsible for setting the RTU information in the license agreement as well as other terms and conditions
We need a way to restrict access to license administration screens
Michelle Martens will bring this up in the Architecture Committee meeting
Farheen Cefalu pointed out that Acumos allows role definition and assignment now, so this does not require a new feature
We need a user story which tests/exercises this role
Work with the portal team to get this feature incorporated
As today is a holiday overseas, attendance was limited today.
We will postpone additional discussion until the next meeting
Planning for the Use Cases, above, should move forward to include complete descriptions, team definition, milestones, schedule and status
We will set up a template for a Use Case wiki page for each one.
We will ask the Use case leads to fill these pages out and to walk the Community Committee through their plans at upcoming meetings
Reuben Klein will send out an email to all the leads to explain what is needed.
We made a recording of this meeting and have attached it here: Community Meeting Recording 1-28-20
We reviewed the general categories of use cases and why we are proposing them
We wish to establish the value of machine learning in contrast to predictive analytics
Exploit the generality of ML when applied to different datasets
Create a library of good models for adopters to make use of as a jump start
Identify product gaps that result from the difficulty of (if any) of tying together multiple predictors
Identify product gaps in the areas of training and lifecycle.
We discussed MLFlow
The advantages of using existing opensource projects to accelerate Acumos development with fewer resources
The benefit for the LFAI community which is derived from closer interoperability between Projects.
Allow users to adopt a full set of AI components which are known to interoperate.
We watched a video demonstration from the MLFlow team
The TSC has sent out a request for nominations for a Chair for the Community Committee.
Natarajan Subramanian has sent out an email to the distribution list as well.
Looking for a response by the end of this week.
Nat will send out a reminder if we have no nominees before the next TSC meeting.
Syed Anwar Aftab suggested we contact other people directly to ask them if they are interested in taking this role
How do we get better focus and more interest in the Community Committee?
We should continue to discuss new desirable features and designs
The focus ofo this meeting should be on use cases and user experience
List areas where we are discussing use cases from the Demeter meeting
Use case for Bidrectional communications: An exception occurs on a model running in a partner site - exception data transmitted to publisher. (review with Wenting Sun )
Monetization of model development: Build models, advertise them and sell/support licenses
Michelle Martens brought up participation on sub-commitees and commtment of resources from the Community
What can be done to encourage additional participation?
We need to have incentive for the community to contribute and commit to Acumos
We could ask the LFAI community which is doing successful outreach to make interoperability with existing projects (such as Acumos) a requirement for graduation of any LFAI incubation project (e.g. ONNX, Horovos, Milvus, etc.)
Discussed several use cases:
Network Disruption Predictors:
Given a fault condition, predict severity (i.e. services impacted, scope and timing of any service disruption) at a particular location based on available logs.
VNF Failure Predictor: Automatically analyze and detect early signs of anomalies of VM behavior in the network. Proactively predict potential faults or failure of VMs in the network. (Could take ML models already developed and apply to ONAP)
Infrastructure Predictor: Predict resource utilization at a given future point in time based on current performance trends using a historical context (May extrapolate node occupancy or load from current trends. Could factor in busy periods and cyclic volatility)
Automated planning and purchasing; Learn when to repair or order replacement equipment based upon an increased likelihood of service impacting failures in the network.
Learn to identify unusual traffic patterns and compensate (adapt flows around failures/overloads etc.)
ML Enabled Service Assurance (Tingting’s project)
Self-healing of network failures
Report or improve performance of a network resident application by adjusting network parameters.
Improve precision of technician dispatch
Improve proactive care – generate contact information for impacted customers using subscription data
Third Party Application Support:
Enterprise Machine-to-machine applications
Edge Router monitoring for traffic anomalies that may indicate intrusion
Use Cases that can justify Acumos work and specific features
IoT device applications - e.g. webcam monitoring of network equipment, home security,
factory/assembly line monitoring - predict failures. schedule maintenance, profit/loss calculation. scheduling, capacity management
Boreas/Clio Migration - People need to update licenses (republish models) when Boreas models are run in Clio
AT&T is supporting a PoC on Clio to demonstrate that license management actually works in production
Working with Ericsson licensed models and TechM licensed models
We lacked a quorum to get a good concensus on how we want to work with TM Forum
Several use cases were discussed and new ones suggested
Reuben Klein will provide feedback to TM Forum on a preliminary list of use cases and objectives
We will wait for a larger audience to finalize this work
As the next 2 weeks are holiday weeks, our next meeting will be on Jan 7
Continued discussion setting up the Demeter Release Project Status JIRA page that was begun in the PTL meeting
Murali managed the discussion and updated the page cited above
Bryan discussed OA&M EPICs
Sayee discussed MLWB EPICs
Discussed value statement - Business value of Demeter priority features
PTLs to complete these statements
Discussion at next week's meeting on each
Seeking Community agreement on value of each priority item
We need review of ACUMOS-3773
Should be ready by this Friday and could be closed
Check at next call if we are done with this, or discuss further as necessary
LUM data accuracy validation
RTU data accuracy - ACUMOS-3639
Demeter epics are due Nov 22
User stories done by Nov 29
We did not have a quorum to discuss ACUMOS-1040. We will need a meeting dedicated to discuss this and understand what is required.
A design is required for this feature. Propose to create a design document and vote on that.
Reviewed LUM demo - but only went through about 50% because of time and questions
We need to have a review of the detail with security SMEs
If we were to have an automated way of creating trust in any process, what would we use?
LUM is basing artifacts on Acumos identifiers. Is there a validation process that can confirm these identifiers in the LUM?
No meeting due to Lockup Demeter Planning meetings Wednesday-Friday.
See lockup details at this link.
next release and the requirements has to be aligned to the success criteria.
Manoop:/Farheen : Community committee gather the requirements and discuss it with Architecture committee to finalize it.
Karthic and Sachin presented the AutoML concept. Karthic shared the CLMP work in progress AutoML solution. We need to further discuss in which project AutoML should be developed. Reuben suggested on of the following: model onboarding, model design or training. We agreed that AutoML is an important candidate for Boreas release.
Bryan presented the challenges we have as a community – what are the expectations we have about the trust we put on the models.