# Daily Note 13/08/2020 ###### tags: `Daily Notes` , `Acumos` ## Name : Christofel Rio Goenawan ## University : Bandung Institute of Technology (ITB) --- ## Schedule: 1. Study Detailed Technical Problem in Acumos AI Latest Versions. 2. Study Design Consideration in Acumos AI. 3. Study Acumos Platform Architectures from Paper. 4. Continue to try installing Acumos AI in NTUST Server. ## Outcome : 1. Understand Detailed Technical Problem in Acumos AI Latest Versions. 2. Understand Design Consideration in Acumos AI. 3. Undestand Acumos Platform Architectures from White Paper. 4. Still Couldn't Give Admin Permission to Installer of Acumos AI. ## Further Plan : - Continue to deploy Acumos AIO in NTUST server - Study more detailed about AIO Installation in Kubernets --- ## Daily Log ### 1.Study Detailed Technical Problem in Acumos AI Latest Versions.<mark>(9.00)</mark> - Study more detail explanation in [Hitachi's Presentation](https://events19.linuxfoundation.org/wp-content/uploads/2018/07/OSSJ2019-Machine-learning-lifecycle-management-with-Acumos-AI-platform-across-multiple-environment-r2.pdf ) and other sources. ### 2.Study Design Consideration in Acumos AI. <mark>(10.00)</mark> - Study more detail explanation in [Hitachi's Presentation](https://events19.linuxfoundation.org/wp-content/uploads/2018/07/OSSJ2019-Machine-learning-lifecycle-management-with-Acumos-AI-platform-across-multiple-environment-r2.pdf ) and other sources. ### 3.Study Acumos Platform Architectures from Paper. <mark>(13.00)</mark> - Study more detail explanation in [White Paper](https://cache.techmahindra.com/static/img/pdf/throught-leadership/thought-leadership-how-acumos.pdf) and other sources. ### 4.Continue to try installing Acumos AI in NTUST Server.<mark>(15.00)</mark> - Continue to try to deploy Acumos AIO in NTUST server using Prep- Deploy Process based one previous [study notes](https://hackmd.io/@christofel04/TEEP_Daily_Notes_10_7_2020). --- ## Report ### 1. Technical Problem in Acumos AI >In this note Writer use [Hitachi's Presentation](https://events19.linuxfoundation.org/wp-content/uploads/2018/07/OSSJ2019-Machine-learning-lifecycle-management-with-Acumos-AI-platform-across-multiple-environment-r2.pdf) as reference. In Hitachi's testing of Acumos AI latest release, there are some technical problem in Acumos AI as below. 1. **At Installation** - **acumos_k8s_prep.sh**: Experienced **network corruption during k8s setup code**. - Components are installed **via Helm charts, manifests, and shell scripts**. Understanding all of them are **headache**. - **Reinstallation** or continue installation after fixing minor problem is not so easy task. - Installer automatically **erase all installed components beforehand**. 2. **Runtime Restrictions** - Some quick fixes are required to run on **multi-node configurations**. - **/etc/hosts** are **rewritten manually**. Some of them are **not consistent**. 3. **Architectures** - System configurations **changed so often**, because currently rearchitecting for Openshift, generic k8s, and docker environments. - Example: API management layer , Kong(Athena) to Ingres(Boreas, for Kubernetes) - It takes a **lot of time to upgrade and validate** the functionality of the deployments of different architecture. - So it better to watch and learn the difference of architectures for every releases and updates for a while. --- ### 2. Design Consideration in Acumos AI >In this note Writer use [Hitachi's Presentation](https://events19.linuxfoundation.org/wp-content/uploads/2018/07/OSSJ2019-Machine-learning-lifecycle-management-with-Acumos-AI-platform-across-multiple-environment-r2.pdf ) as study sources. #### 1. Good Points in Design Concept and API design are well-designed, and may have the rooms for future extension. The points can be seen as below. 1. **Generating solution deployment package as a “zip with install scripts.” are convenient, due to its portability and extensibility to support many environments**. ![](https://i.imgur.com/2TDIsm0.png) 2. **Concept of “Managing Projects and Resources” in Acumos AI is very convenient for data scientists to archive all related resources at once**. ![](https://i.imgur.com/PeTveDR.png) #### 2. Support Recomendation There are some support recommendation for further Acumos AI release. 1. **Project Extension to Add Capability to Manage Extra Components in "Projects"** - It is better to have **simple extension to add managed resource to the “Project”**. - Example resources are: - Experiment management service like MLflow. - Data sets to **track appropriate data sets for training machine learning models**. 2. **Easy Way to Specify Whether Jupyter Requires "GPU's" or Not** - It would be better to **add some hints or guideline to specify the requirements for infrastructure**.(e.g. “GPU” required. etc.) - ***Kubeflow*** can **specify the full requirements by specifying manifest**. 3. **Additional Soultion Deployment Export Plugin For Non- Kubernetes and Edges** - It would be better to **standardize the workflow to on-board and export solution deployment package** even if target environment is non-k8s. - **Extra package generator** other than “k8s” platform is a good option. --- ### 3. Acumos Platform Architectures >In this note Writer use [White Paper](https://cache.techmahindra.com/static/img/pdf/throught-leadership/thought-leadership-how-acumos.pdf) as study sources. The simple Acumos AI Platform Architectures can be seen as below. ![](https://i.imgur.com/odeXfHI.png) From the architectures above we can see there are 4 main components as below. 1. **Function for Onboarding Model from ML Tools** 2. **Set of Common Services for Microservice and Docker Creation** 3. **Marketplace Function** 4. **Supporting Operations and Admin Capability** There's also ***design studio*** that **provides a graphical interface for chaining together multiple models, data translation tools , filters and output adapters into a full end- to end solution** that can be deployed into any runtime environment to accelerate AI applications. The design studio can be seen as below. ![](https://i.imgur.com/7zPVJDo.png) The design studio **enables users to compose and stitch individual models into more complex, aggregrate models that can be used in ML applications**. --- ### 4. Continue to try installing Acumos AI in NTUST Server >This deployment is continuation from [yesterday's notes]( https://hackmd.io/@christofel04/TEEP_Daily_Notes_11_8_2020 ). In this note Writer use [Previous Notes](https://hackmd.io/@christofel04/TEEP_Daily_Notes_10_7_2020) as study sources. After checking the Open Shift cluster installation , there are issues when setting cluster up in Open Shift. :::danger In Writer's installation the setting freeze after installing kubeflow as below. ![](https://i.imgur.com/27U2fYv.png) But after look at [reference](https://github.com/openshift/origin/issues/21253) , it should end up showing Open Shift user account and server as below.. ![](https://i.imgur.com/WoHOBzT.png) ::: :::warning After Writer search in internet and from reference it found out this issue **often happens for CentOS with Open Shift veruson 3.11 like Writer's**. ::: :::info **Next Writer Will Continue to Solve the Open Shift Issue in NTUST Server** ::: --- ## Reference 1. https://docs.o-ran-sc.org/projects/o-ran-sc-ric-plt-vespamgr/en/latest/overview.html 2. https://events19.linuxfoundation.org/wp-content/uploads/2018/07/OSSJ2019-Machine-learning-lifecycle-management-with-Acumos-AI-platform-across-multiple-environment-r2.pdf 3. https://cache.techmahindra.com/static/img/pdf/throught-leadership/thought-leadership-how-acumos.pdf