Apr 06 2020 11:35 AM
Apr 06 2020 11:35 AM
One of the most impressive parts of moving applications into the cloud is the ability to apply different types of computing power to your application based on use-case. By taking advantage of all the different compute types within Azure, users access to various CPUs and GPUs that can be implemented. No upfront costs or concerns about racking the gear. No fighting for better support with your vendor or having to buy the latest and greatest to keep up. Azure provides you with a number of options to help implement these various compute options.
Different use cases will often require you to make decisions about what compute options you've selected. How will they improve portions of your application for the business? In this post we'll look at how you can use multiple types of compute options within a Kubernetes cluster in order to use different resources for different parts of their application.
This is a post on understanding Azure Kubernetes Service (AKS) node pools. This is a blog post that makes assumptions that you're done the Azure tutorials around fundamentals. It will also assume you know how to create and use the Azure Kubernetes Service to launch a cluster.
If you'd like to start with these, check out these Microsoft Learn Modules and blog posts:
Microsoft Docs Learn Azure Fundamentals
Microsoft Docs Learn Introduction to Azure Kubernetes Service
Kubernetes Terminology for Beginners
Azure Kubernetes Service - A Beginner's Guide.
From the official Kubernetes Docs
The Kubernetes node has the services necessary to run application containers and be managed from the master systems. A node is a worker machine in Kubernetes, previously known as a minion. A node may be a VM or physical machine, depending on the cluster. Each node contains the services necessary to run pods and is managed by the master components. The services on a node include the container runtime, kubelet and kube-proxy. See The Kubernetes Node section in the architecture design doc for more details._
It could be your local computer using minikube, a server in a datacenter or a virtual machine in the cloud.
You can use different types of CPUs and storage with management for your AKS managed nodes. A subset of VM's like hardware and configuration. Scale them based on your utilization.
From Create and manage multiple node pools for a cluster in AKS on Microsoft docs website:
In Azure Kubernetes Service (AKS), nodes of the same configuration are grouped together into node pools. These node pools contain the underlying VMs that run your applications. The initial number of nodes and their size (SKU) is defined when you create an AKS cluster, which creates a default node pool. To support applications that have different compute or storage demands, you can create additional node pools. For example, use these additional node pools to provide GPUs for compute-intensive applications, or access to high-performance SSD storage.
From Microsoft docs:
GPU optimized VM sizes are specialized virtual machines available with single or multiple NVIDIA GPUs. These sizes are designed for compute-intensive, graphics-intensive, and visualization workloads.
Tailwind Traders is the world's biggest fake company with a reference app used for this and many other examples. You can review this video from Azure Friday about the reference apps. You can also review the application repository here, it's open to the world to build with all the tools in this blog post.
The Tailwind Traders Website at its core is an e-commerce site with a number of extremely critical services that the company requires reliability and uptime in order to successfully execute new orders. In order for Tailwind Traders to understand their customers better, much of the experience the user has is stored in a NoSQL database in Azure Cosmos DB. Cosmos provides Tailwind Traders with a low latency and high throughput database that works with the MongoDB API. Daily, reports on this information are run and delivered into a web presentable front end.
The reports app uses a number of algorithms that the data science team at Tailwind Traders developed along with the application engineer; these are extremely compute-intensive and seem to be taxing the capabilities of the more general-purpose CPU offerings in Azure. The Tailwind Traders engineering team recently looked at processing times of their daily reports for and recognized that even though they were wise enough to architect their application into microservices, they felt that the return time of some reports (nearly 8-9 hours) had a potential of being reduced by utilizing GPUs available in Microsoft Azure.
Tailwind Traders will want to use standard CPUs for a portion of the web front end of my app. The company's CIO would like to have reports on their collected data processed quicker on GPUs in order for the business to best place the most popular and impressive products to potential customers based on data captured. The more information Tailwind Traders is able to gain on their customer's experience, the more they can improve it.
TWT's team will build an AKS cluster with multiple node pools. The two pools will contain different types of computing power to handle the microserviced application workloads.
To begin, the team will create a new cluster for AKS, in this case,
nodepool01 that will handle the web app front end using a Bs2 series VM for the node pool.
You can also create node pools and specify your specific CPU type with
az cli tool rather than do so by the portal:
# Create a resource group in East US az group create --name myResourceGroup --location eastus # Create a basic single-node AKS cluster az aks create \ --resource-group myResourceGroup \ --name cluster01 \ --vm-set-type VirtualMachineScaleSets \ --node-count 2 \ --generate-ssh-keys \ --kubernetes-version 1.15.7 \ --load-balancer-sku standard # Add a node pool to the cluster az aks nodepool add \ --resource-group myResourceGroup \ --cluster-name cluster01 \ --name nodepool02 \ --node-count 3 \ --kubernetes-version 1.15.7
This cluster will use B-series burstable VMs which are ideal for workloads that do not need the full performance of the CPU continuously, like web servers, small databases and development and test environments - exactly the solution for a web front end service and its required components such as a load balancer.
The TWT team chose to create a secondary node pool
nodepool02 that will handle my GPU intensive workloads, such as processing data for reports that are used to improve customer experience. The secondary node pool is created using a NC6s_v2 class virtual machine.
The TWT tech team next can create a node pool using the az aks node pool add command again. This time, specify the name
nodepool02, and use the
--node-vm-size parameter to specify the Standard_NC6 size:
az aks nodepool add \ --resource-group myResourceGroup \ --cluster-name myAKSCluster \ --name nodepool02 \ --node-count 1 \ --node-vm-size Standard_NC6 \ --no-wait
The TWT team can begin assigning specific pods to nodes by scheduling pods using taints and tolerations.
By providing a dedicates pool of GPU backed nodes, TWT will now be able to run their reports using the power of single or multiple NVIDIA GPUs.
There's more documentation available to you to find out the best way to implement node pools with AKS. Check out the different links and videos provided before so you can begin using this powerful option for those who want to diversify compute options within their AKS cluster.
Microsoft Docs: az aks nodepool
Create and manage multiple node pools for a cluster in Azure Kubernetes Service (AKS)
GitHub Actions for deploying to Kubernetes service
Azure Kubernetes Service (AKS)
Multiple node pools in Azure Kubernetes Service (AKS) | Azure Friday
Getting production ready in Kubernetes
If you're running into issues and need assistance with AKS or any other Azure service, reach out to me:
I am always happy to hear from developers and engineers who are trying to implement new services to improve their experience for their customers.