The Enhanced Cisco Container Platform
Download as PPTX, PDF0 likes273 views
The document discusses Cisco's Container Platform and provides the following key points: 1. Cisco's Container Platform provides a turnkey solution for production-grade Kubernetes container environments that is easy to acquire, deploy and manage on hybrid cloud infrastructures. 2. It features native Kubernetes integration that is 100% upstream compatible, integrated networking, management and security capabilities, and support for AI/ML workloads. 3. The platform architecture includes hardware from Cisco (UCS servers, Nexus switches), virtualization software (VMware, HyperFlex), and container-specific software like Kubernetes, Istio and Prometheus for orchestration, networking and monitoring of container workloads.
Ad
More Related Content
What's hot (20)
Similar to The Enhanced Cisco Container Platform (20)
Ad
More from Robb Boyd (20)
Ad
Recently uploaded (20)
The Enhanced Cisco Container Platform
- 1. Meenakshi Kaushik Wednesday, July 17, 2019 10:00 am The Enhanced Cisco Container Platform
- 2. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential Challenges when Deploying Kubernetes Clusters Source CNCF Survey DEVNET-1227 2
- 3. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential Cisco Container Platform Hybrid Cloud Optimized E.g.: AWS, Google, … Flexible Deployment Model VMware (air gapped) | OpenStack (air gapped) | Public cloud Integrated Networking | Management | Security | AI/ML Support Native Kubernetes (100% Upstream) Direct updates and best practices from open source community Turnkey Solution for Production-Grade Container Environments Easy to acquire, deploy and manage | Open and consistent | Extensible platform | World-class advisory and support DEVNET-1227 3
- 4. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential Control Plane Data Plane VM VM Control Plane Kubernetes Automation Orchestration Operations HX Connect Cluster/ Machine Controllers VM VM VM Cluster 1 Kubernetes Cluster1 Workloads Cluster1 Ops Pod Pod Pod VM VM VM Cluster 2 Kubernetes Cluster2 Workloads Cluster2 Ops Pod Pod Pod Kubernetes Fluentd Prometheus Kibana Hyperflex Contiv Storage (HyperFlex / VMware ) Networking (Nexus 9K) Compute Hardware (UCS) Hypervisor Layer (HyperFlex / VMware ) Cisco Container Platform Architecture VM VM Istio DEVNET-1227 4
- 6. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public#CLUS Cisco UCS HX/Compute Cisco Container Platform – Enterprise Kubernetes Hardened Container OS vSphere v6.5 U1 (and above), Non-VMware 100% Upstream Kubernetes NVIDIA GPU* Hardware * V100 3rd Party Compute / Storage Automated Installer (Wizard-Driven, Modular, Robust, Repeatable) Containerized Deployment (Separated Control and Data Planes) Day 0 Cisco Nexus 9000 (ACI) Cisco Blades, Racks (B / C) Day N Software Lifecycle Manager (Day 0 to N operations – Cluster Mgmt, Update/Upgrades, Patches, API) HA (Control Plane) Ubiquitous Security (Hardened Linux, RBAC, non-root, Secure External Registry, etc. ) Failure Recovery (Control Plane Snapshotting) Logging & Assurance (Logging, EFK - Analytics and Visualization stack) Integrated Tools (Visualization: Prometheus) Unified Management System (Multi-Tenant, Single Pane of Glass, GUI, REST API) Container Networking (Calico, ACI, Contiv) Performance Enhancement (Fast Data with Fd.io/VPP-Via Contiv, multi-gpu node-pools) Load Balancing (L3/L4: MetalLB, L7: nginx, Istio-Envoy, Ent Grade Integration-AVI, F5) Hybrid Cloud Integration (AWS EKS, Istio service mesh) PSOCLD-1043 6
- 7. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential Accelerate Innovation Developers Freedom to access the best platforms and tools for increased agility IT Ops One consistent environment to eliminate silos & drive efficiency Security team Visibility and control across one hybrid environment without slowing innovation DEVNET-1227 7
- 10. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential AI / ML On-Prem Source Kubeflow Community Survey DEVNET-1227 10
- 11. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public#CLUS Use Case – Multi-GPU as a Service • Manage AI/ML K8s workload similar to any other K8s App workload • GPU consumption automatic; Seamless scale up/down • Auto install, upgrade and manage dependency for CUDA drivers, device drivers and Kubernetes • Enable high performance distributed training/serving • MultipleGPU per worker nodes • GPU selection with high performance links between GPUs / CPUs DEVNET-1227 11
- 12. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential GPU Support On Cisco Container Platform Worker VM HYPERVISOR Worker VM Kubernetes Cluster 1 Kubernetes Cluster N Worker VMWorker VMWorker VM … GPU GPU UCS Server GPU Node Pool GPU Node PoolNon-GPU Node Pool UCS Server HYPERVISOR Cisco Container Platform v4.0 • Ubuntu 18.0.4 & K8s 1.13+ • Multiple GPU per worker node (Default) • GPU passthrough (native performance), TESLA V100 PCIe, … • Auto provisioning of Nvidia CUDA drivers, container runtime, NVIDIA device plugin: GPU workload provisioning acceleration New Feature GPU DEVNET-1227 12
- 13. Multi-GPU as a Service Demo
- 15. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public#CLUS Customers are looking for ML Lifecycle Manager Source Kubeflow Community Survey DEVNET-1227 15
- 16. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public#CLUS Make it easy for everyone to develop, deploy and manage portable, scalable ML everywhere DEVNET-1227 16
- 17. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public#CLUS Kubeflow Essentials General Applications Infrastructure (Cloud/On-Prem) Container Runtimes ML Workloads (Modelling, training, roll-out, serving, ...) Infrastructure (Cloud/On-Prem) Container Runtimes DEVNET-1227 17
- 18. © 2016 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2019 Cisco and/or its affiliates. All rights reserved. Cisco Public#CLUS Kubeflow Architecture Libraries and CLIs Higher Level Services Low Level APIs / Services Arena kfctl kubectl Katib Pipelines Notebooks Fairing TFJob PyTorchJob Jupyter CR Seldon CR Kubebench Metadata Orchestration Pipelines CR Argo Study Job MPIJob Spark Job Model DB TFX Developed By Kubeflow Developed Outside Kubeflow Adapted from Kubeflow Contributor Summit 2019 talk: Kubeflow and ML Landscape (Not all components are shown) DEVNET-1227 18
- 19. Kubeflow Demo
- 20. Thank you for watching.
Editor's Notes
- #3: https://www.cncf.io/blog/2018/08/29/cncf-survey-use-of-cloud-native-technologies-in-production-has-grown-over-200-percent/
- #7: Currently Cisco first starts with the requirement of a infrsstructure platfrom like Cisco Hyperflex running VMware as an example. Deployment of CCP doesn’t require internet access for deployment and therefore allows for air-gapped deployments. From there we deply virtual machine instances running Ubuntu OS fully back-end suppored by Canonical, included in the subscription. The Kubernetes releases are also fully back-end suppored by Google, also included in the subscription. We offer a very easy UI driven wizard to install the Kubenetes cluster that runs the CCP Control Plane . From the Control Plane, you can then again from a UI wizard or from API calls, easily deploy Kubernetes clusters (we call them tenant clusters) either on-premise or native AWS EKS, with other cloud providers to follow. With each on-premise tenant cluster, you get: A Choice of container networking: Cisco ACI, Contiv or Calio CNI plugin Persistent volume storage on using Vmware volume plugin or when on using Hyperflex, their storage pluing is immediately avaible. Every cluster comes with logged from a deployed ELF (elcastic search, fluentd and kibana) and monitoring with Prometheus and Grafana. Other selectable turnkey options like a Harbor Container registry, Istio, and AWS IAM authentication are also avaible. Finally, we deploy a new release each month and give the customer the opertunity to live migrate and upgrade clusters .
- #11: https://docs.google.com/presentation/d/100GXMDDKDyiANSZnvIN_l5kCNxKnsOOcTLq-34-OaJc/edit#slide=id.g5028e32a9a_2_136
- #13: In CCP 4.0, we now have the ability to attach multiple GPUs in the server and thru the VMware Passthrough option to a Kubernetes node and run AI/ML workloads defined using a GPU annotated node pool. GPU passthrough gives near native GPU performance compared to vGPU; good for machine learning workloads. GPU passthrough also provides better security. VM with GPU passthrough cannot be live migrated CCP installs nvidia-container-runtime=1.0.0+docker17.03.2-1, cuda-drivers
- #16: https://docs.google.com/presentation/d/100GXMDDKDyiANSZnvIN_l5kCNxKnsOOcTLq-34-OaJc/edit#slide=id.g51ff2feecb_0_234