#acumos-meeting: Architecture Committee
Meeting started by farheen_cefalu_a at 15:01:59 UTC
(full logs).
Meeting summary
- Review of Training module and Plan (farheen_cefalu_a, 15:03:34)
- Michelle: We have Alex S. from AT&T to show
the new design for license manager. (farheen_cefalu_a,
15:04:05)
- Licensing Plan (farheen_cefalu_a, 15:04:50)
- Alex: We have a wiki page under License
manager. License usage manager architecture. This is a draft and a
work in progress. It's high level but is filling up with more
details. High level overview intentions and concepts. What are the
impacts on acumos and how we are going to interact. License usage
manager we are using everywhere. The second piece we will get into
details. the third part is license (farheen_cefalu_a,
15:07:54)
- 5. the API of the license usage manager. This
is the api we expect acumos to use for license usage manager. This
is high level intentions and talk. We want to keep moments here.
We're have three big concepts. 1. Software and very fluid
definition for the softwaer. We can have sotware id by licensing
swid. It is like a version. 2. The second piece is license which
is essentially default rules and lim (farheen_cefalu_a,
15:09:54)
- rules and limits will overwrite what's defined
in the license and call also limit users or assets. 3. Assets are
what the company is paying money for Asset is one specific copy of
software. If you have a dvd it's a copy of the software and an
asset. If you take the dvd and install it on a laptop then you have
an asset in dvd and an installed software is a different state. Now
this is the impact to acumos. (farheen_cefalu_a,
15:11:43)
- we want to limit the number of fields like
software ids and license ids. These are 4 things we're talking
about and we're going to have toagree on the lifecycle of all of
them. Concept is we will touch the license usage manager to the
state machine. We want acumos to interrupt its workflow and ask for
license entitlements. We are going to work on limits and rules on
state diagrams. How many actions and tra (farheen_cefalu_a,
15:13:51)
- we also expect acumos to upload the license and
write to the usage manager. We have some defaults set. We also
understand that acumos is trying to do the composition and training
of the models and solutions. We have to track and acquire licenses
for chained solutions. Acumos doesn't have to make decisions or
tracking of the license usage. The license usage manager will do
this. Any questions? (farheen_cefalu_a,
15:15:32)
- Michelle: I want to clarify that we are not
referring to the state of the model is not private, company, or
public. Alex: No. The state machine will be what happens in catalog
to everything that can happen in deploy. Any action that happens
with the software. It's a little bit like DOM. I think i covered
right to use and license usage manager will merge them and figure
out what are the rules. We talked a (farheen_cefalu_a,
15:17:21)
- this is a high level state machine. I was
thinking acumos already has. We're going to have it in a generic
way that adds to the state machine. there is a big piece of catalog
deployment and download. i want to share and envision that is close
enough i have some extra. The things that we want to find from state
machine are 1. we want to know all the limits on each state. 2.
State transitions require lice (farheen_cefalu_a,
15:19:15)
- blic and interrupt and ask license manager that
is sent from state to state yes or no. Another similar but
important is deploy is a copy of the software and LUM will have to
assign a tag and LUID will send it back. We have two different
assets with totally different lifecycle. You can ask for
entitlement. You can ask for all the state diagrams of that vendor
for that model. We can find the rules to entit (farheen_cefalu_a,
15:21:17)
- state diagram. for example if you're going
from catalog to deployment state this is how we put rules and limits
on the state diagrams with license usage. If this state requires
acquisiton of the asset check or no check. Only one copy of this
software piece is allowed to be deployed at one time. if something
is deployed and released in the end the second deployment can
happen. Entitlement will say you hav (farheen_cefalu_a,
15:23:04)
- can have this restrictions on every state. You
can only have up to 5 assets. if it's federated in there is a role
here and it's limit is zero then we do not allow to change the
state. You will not go anywhere. It's still a work in progress and
we can put rules. This is applicable to acumos and we can
use. (farheen_cefalu_a,
15:24:45)
- Bryan: To clarify these wireframes are not for
the user UI. Alex: There are three steps. state diagram can be
used for each software piece. Second step we will put in default
rules ok you're allowed to put it in catalog but you are not allowed
to download or deploy. Step 3. if someone pays money then it
overwrites all the default rules. (farheen_cefalu_a,
15:26:47)
- Alex: This is intention. I am not saying this
will be ready for this release. We will get started. Michelle: In
the first phase we are going to check that there is a default
license there is a models. There is an entitlement that exists. It
may not go through detailed constraints but perform a basic
check. (farheen_cefalu_a,
15:28:01)
- Alex: Next slide start with inside of acumos.
Key table where we are going to touch our license fields
(farheen_cefalu_a,
15:29:26)
- Alex 3 software fields specific to software
models. It is the primary key to the software id table which says
it's a project developed by id. We are going to store that data in
license usage manager and send it to get it on the federation
step. (farheen_cefalu_a,
15:31:43)
- Bryan: At this point we dont have jira items to
push artifacts to license usage manager. Alex: If that happens
outside then we will look for software id. The second id is a
license id for this software. It would be the same and a one to
many relationship in software ids. Bring in the license info into
LUM it will not be done by acumos then we don't need to have the
second field. Bryan: There are two thing (farheen_cefalu_a,
15:33:54)
- for the right to use. Bryan: There will be
rights to use that are assigned to the platform. Site admin. Alex:
3. Right to use. I was thinking it was about user. Acumos can
decide. It's the rights to use. It's the first message that is
sent when the asset is acquired. LUM will figure out what to use
for that specific. As soon as LUM identifies. We're intterupting
state machine. One important piece is (farheen_cefalu_a,
15:36:37)
- server inside of acumos. any piece in acumos
can talk to license guard. Deliver back responses. Bryan: This is
represented on a draft. Alex; this is what we did in ASD and CCD.
And bottom table is when something is deployed. On deploy state we
can return back the license keys. Extra data that is not for boreas
but later. We are not sure how we will use license keys in general.
Next slide onboarding p (farheen_cefalu_a,
15:38:35)
- license guard will talk back to license
manager. This will happen at every state. See a glimpse of what is
happening in license usage manager. It has licensing rights to use.
look there. The last thing I have this swagger yaml for the API.
It is very draft. We can talk about API when things are starting to
freeze more. (farheen_cefalu_a,
15:40:22)
- Anwar: Michelle have we started implementation
for boreas? M: We are aiming to put something into sprint 2. We
haven't figured that out yet. Alex: Architecture is not yet
finalized until we define the apis. Until then we are not going to
start development until they have been identified. (farheen_cefalu_a,
15:41:56)
- ACTION: Michelle:
Let's capture in a line or two what is the plan to test in
boreas. (farheen_cefalu_a,
15:42:25)
- Michelle: The basic is that the supplier
provides default licenses and license manager. Anwar: Let's
consider the 5G use case. Would Ericsson be able to federate
licensed models? Alex; We will not do entitlements and put the
tables for open source and some tracking information. Asset is
acquired and released. Based on the collected data the supplier
will put limits. That will happen after Boreas. For b (farheen_cefalu_a,
15:45:23)
- ACTION: Michelle: Run
the Boreas delivery functions to TSC, Jack will understand what you
are implementing in the end. Check with Wen Ting. (farheen_cefalu_a,
15:46:40)
- Bryan: Do this on the release planning page so
everyone in open source can see. We have the scope on that page and
it needs to be updated. (farheen_cefalu_a,
15:47:19)
- Anwar: Make sure Michelle to tie down a use
case to this so you can demonstrate. We need a working
demonstration at the end of the release. (farheen_cefalu_a,
15:48:01)
- Anwar: Any update on your percentage increase
of your boreas design and architecture. Keep it there for
now. (farheen_cefalu_a,
15:48:51)
- Resource allocation (farheen_cefalu_a, 15:49:07)
- Anwar: We have a core team we have identified.
Let's sit down and review where we can move the resources. One
place is portal. I want to see what are the portal plans for
boreas. What are the other key big items for boreas. (farheen_cefalu_a,
15:50:07)
- ACTION: Mukesh send
Anwar the UI related big ticket items for resource
allocation. (farheen_cefalu_a,
15:51:30)
- Phillippe: Model onboarding is 100% because
functionality architecture design have been designed. From my side.
100% for onboarding. Anwar: And you are including ms minus docker
release. P: Yes but we have to discuss. It is in the scope of
onboarding. Catalog data model Chris: That is scheduled for sprint
2. I have no updates. Alex makes sure Chris gets apis for license.
This has to be done in advance (farheen_cefalu_a,
15:53:57)
- We aren't ready. Anwar: Let's make sure the
scope for Boreas release is agreed by the TSC. At the end of every
release we have marketing announcments and acumos AI day. We have
to make sure we know what we are going to deliver for boreas. Alex:
We need time to scope. (farheen_cefalu_a,
15:55:08)
- ACTION: Nat Get the
TSC and release management can track Licensing. (farheen_cefalu_a,
15:55:41)
- Training (farheen_cefalu_a, 15:55:50)
- Anwar: We'll have ML workbench Nifi,
jupyterhub/zeppelin AccuCompose, Tensboard? Bryan: Yes, this is the
scope. (farheen_cefalu_a,
15:56:53)
- ACTION: Bryan: Set up
a meeting with Anwar and Mukesh around training. (farheen_cefalu_a,
15:57:30)
- Design Studio, Kazi: What is included is
jupyter notebook, nifi, pipelines, AccuCompose. AccuCompose is DS
renamed under the ML workbench. Jupyter, pipelines, are not wholly
contained within the DS. ML Workbench contains Design studio that
has been renamed to AccuCompose. (farheen_cefalu_a,
16:00:09)
- Anwar: Which components is DS in scope for the
Boreas release. Replace Design Studio with ML Workbench.Jupyter
notebooks, Nifi pipelines, kubeflow moved to clio. Rapid miner? Adi
will have to address. Deployment Bryan? We're making progress on
whitebox and Nifi. Security? No change. Common Services: Guy we
haven't moved. We have the same things from last time. We have to
add the ONNX style generation. (farheen_cefalu_a,
16:04:56)
- model runner improvement. I expect a big
change for next week. Anwar: That concludes. Nat set up a time for
training and overlap with ml workbench. let's set up time.
(farheen_cefalu_a,
16:05:41)
- ACTION: Nat Set up
time for a deep dive into training. (farheen_cefalu_a,
16:05:54)
Meeting ended at 16:06:00 UTC
(full logs).
Action items
- Michelle: Let's capture in a line or two what is the plan to test in boreas.
- Michelle: Run the Boreas delivery functions to TSC, Jack will understand what you are implementing in the end. Check with Wen Ting.
- Mukesh send Anwar the UI related big ticket items for resource allocation.
- Nat Get the TSC and release management can track Licensing.
- Bryan: Set up a meeting with Anwar and Mukesh around training.
- Nat Set up time for a deep dive into training.
People present (lines said)
- farheen_cefalu_a (43)
- collabot` (3)
Generated by MeetBot 0.1.4.