Exam DP-100 topic 4 question 1 discussion

So the right answers should be gpu-cluster for test and AKS for production?

Why not AKS cluster?

Correct answers: - Test: ds-workstation notebook VM - Production: AKS-compute cluster Test: ds-workstation notebook VM Since the requirement for the testing environment is that it must not require an authenticated connection, then none of the cluster options (CPU, AKS, GPU) would be fit for use as a Testing resource. This is because these online cluster resources do not recommend and highly discourage to have unauthenticated connection due to obvious security concerns. DSVM offers an isolated or controlled environment, where unauthenticated access is temporarily allowed, but this would be an exception rather than the norm. Production: AKS-compute cluster The AKS (Azure Kubernetes Service) cluster is better for both (1) low latency and (2) scalability, unlike GPU cluster which is designed for low latency not for scalability, which is a requirement for the PROD env.

Another misleading question: there is no clue about cost requirements, which will lead to the proper answer...

Question like this one on 4-Sept-2023 exam

Test: A) ds-workstation notebook VM For testing purposes, using a data science workstation notebook VM would be ideal. Since you're only testing the API responses and not focusing on large scale inferencing, a fully provisioned cluster would not be necessary. Production: f) aks-compute cluster Azure Kubernetes Service (AKS) cluster is best suited for production deployment of your machine learning model. AKS offers capabilities like auto-scaling and load balancing, ensuring that your model can handle a large number of requests and perform with low latency during inferencing. It is also not necessary for the compute resource to have a GPU for inferencing, making the AKS cluster a cost-effective option.

I would go for a cpu-compute cluster for testing

CB. Explanation: For testing, a CPU-based compute cluster should be sufficient since the primary requirement is to verify that the web service returns predictions in the expected JSON format. A CPU-based compute cluster is relatively cheaper and can handle moderate to low traffic during testing. For production, an AKS (Azure Kubernetes Service) cluster is recommended as it offers scalable and efficient orchestration of containers for high traffic applications. Since the mobile app is expected to receive multiple requests from end users, a scalable and reliable production environment is required. The AKS cluster provides an authenticated connection, and Kubernetes can scale the deployed model horizontally to handle a large number of requests. Note that GPU-based compute clusters may offer faster inferencing performance but are relatively expensive and may not be necessary for this specific project's requirements. Additionally, the deployment of GPU-based clusters may require additional configuration and setup, which may not be practical for testing and production.

Box1: I think the answer is here: "When using the Azure Machine Learning SDK v2 on a compute instance or on an Azure Virtual Machine, you can use a managed identity for Azure. This workflow allows the VM to connect to the workspace using the managed identity, without storing credentials in Python code or prompting the user to authenticate. Azure Machine Learning compute clusters can also be configured to use a managed identity to access the workspace when training models." So VM sounds good, as we are talking about testing mode on an inference process https://docs.microsoft.com/en-us/azure/machine-learning/how-to-setup-authentication?tabs=sdk Box2: AKS. Definitely the best solution when dealing with low latency and scaling needs on inference process https://docs.microsoft.com/en-us/azure/machine-learning/concept-compute-target

PROD is AKS for sure, test 3 options -- local -- contain instance -- aks it does not required authentication, I will vote for local machine

"Although compute targets like local, and Azure Machine Learning compute clusters support GPU for training and experimentation, using GPU for inference when deployed as a web service is supported only on AKS." https://docs.microsoft.com/en-us/azure/machine-learning/concept-compute-target#deploy

On 03 March 2022

Train: GPU, test: DS, Production: AKS

Azure compute cluster is not supporting real-time inference, only batch inference I think workstation notebook vm works with the question request https://docs.microsoft.com/en-gb/azure/machine-learning/concept-compute-target https://docs.microsoft.com/en-gb/azure/machine-learning/how-to-deploy-local-container-notebook-vm

what is the final good answer please?

So the correct answers are AKS for both?

on 2/9/21