Robin.io - IT Press Tour #40 Sep./Oct. 2021

Transcript

1. 1 Enhanced Kubernetes Platform Automation and Orchestration W. Brooke Frischemeier Sr. Director 5G and Edge
2. Robin Overview Robin Products Why It’s Better Conclusion Detailed Info In Backup
3. Robin Overview Robin Products Why It’s Better Conclusion Detailed Info In Backup
4. Robin.io Accelerates Deployment and Automates Lifecycle Management of Enterprise Stateful, 5G and Edge Applications on Kubernetes World’s first deployment of cloud-native 5G Stack in production Trusted by F1000 companies for their mission critical Storage and Network applications Headquartered in San Jose Built highly differentiated technology with at least 2 years+ head start over competition Marquee customers – Worlds first e2e containerized 5G network, in production for 2 years - Rakuten Finalist Strong partnerships
5. 5 Customer Proven Results Customer proven efficiencies , CAPEX & OPEX • First production containerized 5G stack with millions of subscribers • 40% reduction in OpEx scalable orchestration & automation for RAN and Core • 50% reduction in CapEx by enabling Open RAN and Core on commercial hardware • 80% reduction in deployment time: from 10 days to minutes • 30% Faster for running VNF vs legacy + CNFs Trusted and proven platform at scale • 6 Petabytes big data apps (Palo Alto Networks) • 2.5 Petabytes ELK, Kafka, 300 nodes (USAA) • 400 Production Databases (Sabre)
6. Ease of us, no experts required, reducing time to outcome Massive scale Remove operations & resource silos High Performance Rapid adaptability and flexibility Not Checking Boxes…Doing It Better
7. Robin Overview Robin Products Why It’s Better Conclusion Detailed Info In Backup
8. 8 Robin Solutions Remove Background of Jewelry ROBIN Cloud Native Storage Kubernetes Application Workflow Manager Automated NF Performance Tuning Multi-tenant Observability ROBIN CNP Virtual Networking Deploy Anywhere Baremetal VM ROBIN MDCAP Advanced Workload Placement • Robin builds the industry’s best Enhanced Kubernetes Platform and Bare Metal to Services Automation Suite, deployable in both private and all of the top public clouds • That means, in a single click or automated event we install, upgrade and manage every aspect from bare-metal bring up to service lifecycle management • Robin deploys and manages the entire service stack with – Pre-integrated lifecycle automation – Comprehensive multi-tenant observability stack • Robin, with Rakuten Mobile, deployed the industry’s first production, end-to-end, containerized, 5G solution – 5G Core, Multi-vendor Open RAN, MEC framework – As well as parts of the 4G network • In this deployment Robin has demonstrated improved performance, with better resource utilization, while reducing operations silos and lifecycle timelines, from days to minutes Resource Event Planning
9. 9 Works any where Google Cloud Platform Baremetal, VM Robin Cloud Native Platform App-aware Storage Application-Aware Storage • Automates all major lifecycle events such as backup, snapshot, migrate, clone etc. • Industry leading app awareness tied to the elements and roles that make up your database’s and pipeline’s, as well as your application configuration, networking, meta-data and lifecycle states Enhanced Kubernetes Application Workflow Manager Advanced Workload Placement Multitenant Observability Optimized Performance Sherlock Advanced Planner Planning Tools • Resource dependency analysis • Multi-domain failure analysis • Impacts of cluster, networking, server, storage etc. Robin enhancements for better performing VMs and Containers Easy to use interface that doesn’t require Kubernetes expertise to use it Multi container runtime support Advanced Declarative Workload Placement • Policy-based modeling with no hardcoding • Highly granular NUMA-awareness with Affinity+Anti-affinity, CPU/GPU Pinning, HugePages, networking and much more • Super easy to use interface Virtual Networking Carrier-grade networking • Persistent IPs & Multiple IPs • Overlays/underlays: Calico, OVS, VLAN, SR- IOV • Multiple NICs • Dual-stack IPv4/IPv6 • Network policies Pre-integrated Lifecyle Management 1-Click or API-driven end-to-end Automation Deploy, Scale, Heal, Upgrade, Snapshot, Clone, Backup, entire application pipelines
10. 10 MDCAP Metal to Service Orchestration At Scale OSS/BSS Clouds On- prem Edge Co- location Edge Far Edge Edge Central Edge Local Core ~100s 100s 100s 100ks 10ks 10ks 1ks Bare Metal-a-a-S & LCM • BIOS, BMC configurations • NIC, SSD, FPGA, NVMe, RAID, firmware upgrades • OS installation, Drivers, Services and Software Packages • Upgrades – BIOS, Firmware, Drivers and OS • Prep for observability • Multiple Server Vendors, SKUs and Profiles Cluster LCM • Choice of cluster – Anthos, OpenShift, k8s, Robin, Rancher • Design clusters and profiles • Cluster instantiations, scaling, healing and termination • Upgrades, Prep for observability Application LCM • Instantiation, Healing, Scaling, Upgrades • Data Management – Snapshot, Clone, Backup, Restore, Import • Migrate to other clusters Network Function LCM • Helm charts, Operators, YAML, Custom • Instantiation, Healing, Scaling, Upgrades, Prep for observability Network Service LCM • Design Network Service across clusters • Instantiation, Healing, Scaling, Upgrades Methods Of Procedures Management • Appliances, routers, pagers/mail Inventory Management • Discovery, Health, Liveness and Readiness probes • Network Service to hardware component visibility • Version control • Observability DU DU CU-CP CU-UP UPF SMF AMF UPF MEC LB, DB, Q, ELK, CDN Analytics AI/ML Cassandra Hadoop Redis, Graph DB Inventory MOPs Services Multi-Data Center Automation Platform (MDCAP) RAN-a-a-S 5GC-a-a-S P5G-a-a-S MEC-a-a-S
11. Robin Overview Robin Products Why It’s Better Conclusion Detailed Info In Backup
12. 12 Automated Workload Placement – Everywhere NUMA-aware, Auto-discovered Resources, Modeled, Not Hardcoded Declarative, Intent- based, Input Models Your Intent Finds resources across Clusters and DCs Deploys the Model – Configures For You Manages The Lifecycle MDCAP Regional DC Staging CNP Edge DC Far Edge DC CNP CNP CNP CNP I P H L H D I P H L H D I P H L H D
13. 13 Declarative, Easy To Use – No K8s Expertise Needed Entire Application Pipeline is up and running in minutes, not days Deploy Data Analytics Application Pipeline • All components NUMA and networking components are modeled into the service and auto-discovered • K8S components auto created (No searching, no hardcoding, no K8s expertise needed) • Storage & Networking provisioned • Data-locality, anti/affinity, NUMA- aware policies enforced What additional components need to be deployed Compute Storage Scale Specify Data-locality, Anti/Affinity Constraints and Placement Hints Pre-integrated, per application lifecycle configuration
14. 14 Everything Is Modeled As A Variable No Hunting For Specific Configuration Details, No Hardcoding Inventory Policy Auto- Scaling Observability Discovery Life Cycle States I P S L O D Clear Advantage That Cannot Be Replicates By Legacy Designs
15. 15 Unifying And Streamlining Multi-domain Operations OCHESTRATOR Bare-metal Servers Containers VMs Services PNFs, e.g. Network Switches ELEMENTS Racks RAN-a-a-S EXECUTORS Ansible IPMI REST API Custom API SSH Git Object Store Config Database Jenkins CONFIGURATION Setup BIOS Install OS Configure Networking Power Cycle Upgrade OS Deploy Robin FUNCTIONS Workflow Studio Scale Parallelism Log Storage Inventory Reports CI/CD Git Create Nwk. Slice Add New Service Change Configs LCM Action WORKFLOWS Robin Clusters Ideal Automation Toolset For Multi-domain Network Slice LCM
16. 16 Easy Onboarding – Reduced Integration Times Helm, SmartHelm, Robin Bundles, Servers and etc. • Pre-onboarded hardware and API integration, e.g. many sever vendor specific implementations • NFs, services and applications instantiated your defined workflows are autogenerated for you via: – Helm – simple paste and ingest model with CNP – SmartHelm Annotation language that extends Robin Modeling to Ingested Helm Charts § Insert Variable arguments into helm directly from CNP § Customize LCM operations § Robin will help the customer do this – Robin Bundles – super-operator § Workflow-based language to model interactions between, NFs and supporting applications § Can start with Helm or SmartHelm ingests
17. Robin Overview Robin Products Why It’s Better Conclusion Detailed Info In Backup
18. 18 -as-a-Service Deployments, Provider and Private 10’s of Robin Clusters 10,000’s of Robin Clusters 1,000s of Robin Clusters OSS Edge 1 Edge 2 Edge N BBU 1 BBU N DU EMS CU BareMetal-as-a-Service (BIOS, OS, FPGA, Robin K8S Install, Monitor, Upgrade) EDGE RADIO CORE ROBIN MDCAP MOPs Manager (Highly scalable and extensible MOPs Management across 100s of 1000s of Network Elements) Centralized Management of 10,000s of RAN, Edge & Core clusters Policy based Closed-loop Automation Define & Execute Complex Workflows across RAN, Edge, Core RIC 5GC-as-a-Service (Deploy, Migrate, Heal, Monitor, Scale, Upgrade) 5GC MEC MEC MEC-as-a-Service (Deploy, Heal, Monitor, Scale, Upgrade) CU RAN-as-a-Service (Deploy, Heal, Monitor, Scale, Upgrade)
19. Ease of us, no experts required, reducing time to outcome Massive scale Remove operations & resource silos High Performance Rapid adaptability and flexibility Not Checking Boxes…Doing It Better
20. www.robin.io
21. Additional Detailed Content
22. 22 1 Minute Left hand side is the word for word script using a simple picture Remove Background of Jewelry ROBIN Cloud Native Storage Kubernetes Application Workflow Manager Automated NF Performance Tuning Multi-tenant Observability Resource Event Planning ROBIN CNP Virtual Networking Deploy Anywhere Baremetal VM ROBIN MDCAP Advanced Workload Placement • Robin builds the industry’s best Enhanced Kubernetes Platform and Bare Metal to Services Automation Suite, deployable in both private and all of the top public clouds • That means, in a single click or automated event we install, upgrade and manage every aspect of your – Bare metal servers – K8s clusters – Network Functions & Supporting Applications – Physical devices and Methods Of Procedures (MOPs) – We then deploy the entire service with pre-integrated lifecycle automation and a comprehensive multi-tenant observability stack • Robin, with Rakuten Mobile, deployed the industry’s first production end-to-end, containerized, 5G solution…and we manage VMs too – In this deployment Robin demonstrated improved performance, with better resource utilization, while reducing operations silos and lifecycle timelines, from days to minutes • With Robin, there is no special expertise needed to operate it; there is no searching and no hard coding of resources needed, Robin’s advanced modeling automatically does the for you work for you
23. 23 Robin Provider Mobile, Edge & Private Solutions EDGE FAR-EDGE CENTAL DC O-RAN Private 5G/LTE 5GC/EPC MEC
24. 24 VMs and Containers – Breaking Silos Better Than K8s Frees you from your vendor’s containerization roadmaps • VMs and Containers in the same Worker Node • CNP can also run standards-based KubeVirt • Robin CRI is better then KubeVirt – Robin can run host passthrough and bypass virtualization layer § Direct access to Pinned CPUs per NF § Better Performance and app isolation, lower jitter – Robin can mimic their OpenStack environment better § ROBIN OVS-CNI with VLAN support for VMs to mimic OpenStack for VM PODs in same subnet § Integration with SR-IOV / FPGA / GPU device plugins similar to OpenStack • Unified Operations – Runs like any other Pod (defined in PodSpec) as do containers, not custom definition like KubeVirt that uses VirtualMachineInstances § Fewer caveats – less human error § Same operations rules and LCM as containers § Less to figure out § Easier to integrate – Access to Robin Liveliness probes – better end to end application visibility and failover behavior • Completely shared resource pools One Platform Shared Resource Pools Containers & VMs Container
25. 25 Better NUMA Awareness Than K8s + Other Stuff • Being able to say you are NUMA aware is not the same thing as – Being able to see every NUMA node in every server – Having automated workload placement that can use it • Standard K8s NUMA awareness groups all resources in a worker node into one group without knowing which physical device it is on – This can lead to config failures or mis-configs • Robin CNP is NUMA aware down to every physical node, higher granularity than OTS K8s • Affinity/Anti-Affinity across any component in this slide • CPU and CPU sibling isolations • Not just GPU awareness - can discover and categorize different models/types as different workloads require different GPU types • Multiple networks (Overlay and Underlay) and Multiple IP addresses per Pod/NF • Everything is MODELED Isol CPU OS CPU NUMA 1 GPU Memory FPGA Guest OS Container 4 cores 5G/3x1Gi SRIOV/FPGA/GP U VM 2 cores 5G Mem 1x1Gi HP SRIOV/ FPGA/GPU NUMA 0 GPU Guest OS VM 4 cores 5G / 3x1Gi SRIOV/GPU Container 2 cores 5G Mem 1x1Gi HP SRIOV/ FPGA/GPU Calico FastPath TOR Switch Mgmt TOR Switch P 1 P2 SR-IOV P 1 P2 SR-IOV OVS Memory Hugepages Hugepages
26. 26 K8s Master Nodes K8s Worker Nodes Kubelet CRI Docker Extended Docker Shim LxC Virtual Machines CSI Robin Storage POD Storage Network Robin Worker Daemon Set IPAM Storage Coordinator Monitoring K8S API Server Helm Operator Robin Master DaemonSet Plan Manager and Overlay Scheduler RBAC, Multi-tenancy Storage Mgmt K8S Scheduler CONTROLLER MANAGER ReplicaSet Controller StatefulSet Controller Deployment Controller Service Controller Robin Controller Stock K8S components Robin Components Observability Logging Closed-loop Automation CNI Calico OVS BOND SRIOV MULTUS Enterprise-grade storage stack Snapshots, Clones, QoS, Replication, Backup, Data rebalancing, Tiering, Thin-provisioning, Encryption, Compression Flexible networking Hostname Resolution, Underlay Networking, Multiple-NICS, Persistent IPs, Dual stack IPv4/IPv6, Static IPs, Source Based Routing Additional K8s Plugins Device Plugin App Planning NUMA Awareness CPU Pinning Isolated Cores Huge Pages Policy-based, Multi-Service Affinity, Anti-affinity CNP Under The Covers
27. 27 Advanced Networking Provider NFs need robust connectivity options • IP Persistency across start, stop, heal and migration • SR-IOV underlay networks for high throughput, low jitter, redundancy as well as NF interconnect • Open vSwitch underlays to extend corporate operations networks to NFs • CNI plugin customization • Per-pod Multi-IP Network Support (Multus) • Calico overlays • NIC Bonding for redundancy and throughput • IPv4/v6 Support • Built in metalLB load balancer OSS, BSS, Management Networks Applications, NFs
28. 28 HW-SW NF Performance Tuning Every vendor has specific requirements – onboarding must be automated – no hardcoding • Operating System (OS) versions and configuration • Basic Input/Output Services (BIOS) versions • Field Programmable Application Integrated Circuits (FPGA) updates • Non-Uniform Memory Access (NUMA) affinity between cores, memory and network underlays • Isolated sibling processors • Real-time Kernel specifications, for better scheduling control • Command line parameters when executing the program • Logically and physically diverse Single Root Input Output Virtualization (SR-IOV) underlay networks for high throughput with low jitter/latency • HugePages support • Multiple IP address requirements • Storage network configuration modeling
29. 29 Traditional – Legacy Cloud Design K8S Cluster K8S Master EXT Load Balancer WAN Router WAN Modem Internet Default Virtual Network Ingress (L7 L3) POD POD POD POD POD Resident 2 Applications (Micro-Services) Resident 1 Applications (Micro-Services) POD Legacy web applications have a simple connectivity model
30. 30 Provider Network Design K8S Cluster OSS BSS Monitoring Logging Analytics K8S Master Default Virtual Network Ingress (L7 L3) POD POD POD POD POD Resident 2 Applications (Micro-Services) Resident 1 Applications (Micro-Services) POD Internet EXT Router SDLAN/SLB/UPF/CU/DU/5GC SDEWAN CNF Virtual Network1 Virtual Network2 Provider Network 3 Provider network via OVS Multiple SR-IOV Underlays NGFW Provider/Telco solutions are much more complex and require additional network connectivity options To legacy operations systems and high performance underlay networks