Ilhem FAJJARI Orange Labs Network Towards a Telco-grade Orchestration for Cloud-Native Network Functions 1
Content Conte ntext t & mo motiva ivations tions Scope ope & ambitions bitions Achie hieve veme ments nts Futur ture work rk 2
Key industry trends Containers Automation Microservices . An approach to develop a A technology to give each “ Infrastructure as code ” single application as a application running on a approach to enable the suite of small services , server its own, isolated entire infrastructure each running in its own environment to run. deployment to be process and It holds the components expressed and controlled communicating with necessary to run the through software lightweight mechanisms desired software. Reduce complexity Innovate more rapidly Speed up the testing and offer a more and be more agile process and build responsive IT large, scalable cloud environment; applications 3
Need to increase network flexibility while reducing CAPEX and OPEX All these approaches are eminently suitable to be used in NFV environments NFV should be equally agile and automated Operator ambitions But making NF Because they have cloud- native won’t unique properties be so easy! Why? 4
How to make network functions cloud-native IDS FW Router FW Router IDS Gest OS Gest OS Gest OS Router IDS Host OS Hypervisor FW Host OS Hardware Hardware Traditional physical Virtualized NF 1.0 Cloud-native NF 2.0 NF Containters Virtual Machines That is alll? Absolutely not! 5
NGPaaS project NGPaaS project is an Horizon 2020 EU-funded project and part of the 5G-PPP Phase 2 initiative Started in June 2017 End in June 2019 Consortium: 12 partners - 2 universities - 1 research centre - 5 industrial partners - 4 SMEs 6 6
Accelerated data plane: some telco workloads require native sophisticated network models to support multi-homing with Deterministic performances: some telco workloads are affinity/anti-affinity, energy, etc. network performances to achieve very low latency and jitter Telco Specific scheduling policies: new placement constraints need be correctly steered configured together as a service through which traffic needs to Service function chaining : telco applications must be various QoS as specified by ETSI MANO CPU and memory performances sensitive and hence need predictable access to nges llenges ation challe estration rchestr es and orc ures e featur rade co-Grad Multi-network conncectivity : telco workloads may require Multi – Network rk connect ctivity Service ce Accelera erated Funct ction on data a plane chaining VNF to be supported Network-related, inter-component Specific c Deter ermi minist stic c sched eduling g perfor ormanc mances es polici cies es interne Groupe France Télécom 7
Rancher /2) Mesos Telco adoption is still very limited Docker swarm Telco Kubernetes Several container management & orchestration frameworks exist nges (2/2 … llenges ation challe estration rchestr es and orc ures e featur rade co-Grad Kuberntes is de facto standard Industrialized and mature for IT, but Swarm - Mesos - Kubernetes interne Groupe France Télécom 8 Jan, 2017 Jan, 2018
was originally designed and implemented by Google and then donated to the open source software for container orchestration at is Kubern rnetes etes (K8S) ) ? Cloud Native Computing Foundation What interne Groupe France Télécom 9
Kubernetes is ill adapted for cloud-native VNF orchestration for IT, but Telco adoption is still very limited Kuberntes lacks a number of mandatory features for the management and performance guarantee of Telco applications regard to NFV requirements Kuberntes is de facto standard Industrialized and mature Kubernetes need to be customized to close the gap with Several enhancements need to be integrated interne Groupe France Télécom 10
requirements container-based network functions NGPaaS objective: a Telco-grade Kubernetes for an automated management and orchestration of cloud-native enhanced orchestration of cloud native VNFs use case Scope ope - Design, develop and deploy a Kubernetes based system for an Objec jectiv tives es - Customize Kubernetes to close the gap with regard to NFV - A prototype of a telco grade Kubernetes with a first deployed interne Groupe France Télécom 11
orchestration Kubertenes enhancements towards a telco-grade # VNF Enhance ancement ent descrip cripti tion on requirem irement ent Multi i – Network connectiv ectivity ity 1 Multi-net etwo work Kuber ernet etes es with Kuber ernet etes es exten ensi sions s to support support & multi netwo work k multiple e netwo work k inter erfac faces es per POD data a plane e support and with DPDK K and SR-IOV IOV accel eler erat ation accel eler erat ation datapla aplane Service ice Accele elerated ated Functio ion accel eler erat ation data a plane chainin ining VNF 2 Deter erminist stic Kuber ernet etes es with Kuber ernet etes es exten ensi sions s to support Perfo forman ances ces EPA (Enhan anced ced cust stomized CPU pinning polici cies es and Platfo form huge e pages es Awaren eness) ess) Specif cific ic support Deter erminist inistic ic schedu eduli ling performances ances polici icies 3 Custom Extended ed Extensi sion of Kuber ernet etes es sched eduling to sched eduler er Kuber ernet etes es support new metrics cs (ex. Netwo work) k) sched eduling 4 SFC support Kuber ernet etes es with Kuber ernet etes es exten ensi sions s enabling service vice funct ction service vice funct ction chaining for NFV chaining interne Groupe France Télécom 12
orchestration Kubertenes enhancements towards a telco-grade # VNF Enhance ancement ent descrip cripti tion on requirem irement ent Multi i – Network connectiv ectivity ity 1 Multi-net etwo work Kuber ernet etes es with Kuber ernet etes es exten ensi sions s to support support & multi netwo work k multiple e netwo work k inter erfac faces es per POD data a plane ne support ort and with h DPDK K and SR-IO IOV accel eler erat ation Service ice Accele elerated ated accel eler erat ation datapla aplane Functio ion data a plane chainin ining accel eler erat ation VNF 2 Deter erminist stic Kuber ernet etes es with Kuber ernet etes es exten ensi sions s to support Perfo forman ances ces EPA (Enhan anced ced cust stomized CPU pinning polici cies es and Platfo form huge e pages es Awaren eness) ess) Specif cific ic Deter erminist inistic ic support schedu eduli ling performances ances polici icies 3 Custom Extended ed Extensi sion of Kuber ernet etes es sched eduling to sched eduler er Kuber ernet etes es support new metrics cs (ex. Netwo work) k) sched eduling 4 SFC support Kuber ernet etes es with Kuber ernet etes es exten ensi sions s enabling service vice funct ction service vice funct ction chaining for NFV chaining 13
Enhanced plateform awareness: CPU pinning Use of Control CPU Management Policies to enable CPU pinning Two strategies are supported by K8S: - none policy: relies on the default CPU affinity scheme - static policy: allows pods with certain resource characteristics to be granted increased CPU affinity and exclusivity on the node • Limitations: • the allocation algorithm for the static policy is topology aware best fit Not adapted for our needs e.g., DPDK deployment require the selection of CPUs belonging to a specific socket 14 14
Enhanced plateform awareness: Custom CPU pinning strategy Static policy Custom policy apiVersion: v1 apiVersion: v1 kind: Pod kind: Pod metadata: metadata: name: exclusive-1 name: exclusive-1 annotations: spec: PreferredNUMANodeId: "1" nodeName: node3 spec: containers: nodeName: node3 - image: quay.io/connordoyle/cpuset- containers: visualizer - image: quay.io/connordoyle/cpuset-visualizer name: exclusive-1 name: exclusive-1 resources: resources: limits: limits: cpu: 1 cpu: 1 memory: "256M" 15 15 memory: "256M"
Recommend
More recommend
