real tactical advantage to its clients with Kuber- netes and K3s. The Journey to SUSE RGS and K3s When looking at today’s connected envi- ronment, what is striking is the proliferation of IoT and edge these). It is also a core member of, and contributor to, the CNCF Foundation. Unsurprisingly then, when designing the SmartEdge infrastruc- ture, containers and Kubernetes were part of the plan from the microservices-centric strategy, in solving the most complex of infrastruc- ture management challenges. When it came to Kubernetes management, the team trialed a number of options. “K3s has been a foundational

tool that makes it easy to run Kubernetes locally. Minikube runs a single-node Kubernetes cluster inside a VM on your laptop for users looking to try out Kubernetes or develop with it day-to-day. 7 Out released as stable/GA with Kubernetes v1.13. It is now the suggested way of handling storage volumes inside Kubernetes clusters. Status within the Kubernetes project 10 How to contribute Working Groups

of thousands users signing up during commercial break. Show-time !! First row winner Second row winner Final winner Tease before commercial break End of last show Time 1 week - 1 hour 8 min 15

pods across availability zones, while still respecting resource access and availability concerns. When Kubernetes runs on vSphere, the hypervisor platform also supports an underlying tier of high availability failures How it works: • Kubernetes supports running a single cluster in multiple failure zones. • When nodes are started, labels are automatically added with zone information, based on tags pre-applied system’s memory as possible. Where does this lead? Node 0 32GB Node 1 21GB 2 CPU Nodes – NUMA host When Linux initially allocates a threads, it is assigned a preferred node, by default memory allocations

pods across availability zones, while still respecting resource access and availability concerns. When Kubernetes runs on vSphere, the hypervisor platform also supports an underlying tier of high availability failures How it works: • Kubernetes supports running a single cluster in multiple failure zones. • When nodes are started, labels are automatically added with zone information, based on tags pre-applied system’s memory as possible. Where does this lead? Node 0 32GB Node 1 21GB 2 CPU Nodes – NUMA host When Linux initially allocates a threads, it is assigned a preferred node, by default memory allocations

to 10k cores for a single experiment • Autoscaling capabilities: Pay for what you use, scale down when idle • Parallel training instead of sequential: huge time saver for large trainings Kubeflow • Machine containers, kubernetes and all that stuff • Pachyderm can version datasets and trigger new trainings when changes occur • Distributed File Systems • NFS • HDFS • … Classic DevOps solutions: • Containers • A/B testing and canary release of Models • Comparing Production accuracy vs expected accuracy when possible • Rolling-updates • … Resources • Source code for this talk: https://github.com/ritazh/kubecon-ml

them to figure out what went wrong when they had issues with their deployments. Blackbox Non-Technical Issues • We're not saying that centralized effort is bad when creating initiative to migrate Instead of starting with onboarding small number of teams, we went with the whole organization. When issues happened, we did not have the bandwidth to work on all of them at once. Start with a Bang

Case 2: Retry Pod should be recreate when the previous pod is failed. The previous node should be excluded. Case 3: Critical Deamonset Node should be tainted when critical Daemonset is unhealthy. Case

�������� �� ����������� Scheduling Isolation Efficient placement of service container and tasks When placed together, don’t affect each other Resource contention ���� ������ ���������� - Online workload

provisioning, updating, and managing server infrastructure. – Flexible Scalability • No compute cost when idle Benefits Serverless landscape defined in CNCF Kubernetes + Apache OpenWhisk Kubernetes