Optimized placement in Openstack for NFV
7 likes2,601 views
The document discusses enhancing efficiency in OpenStack clouds through optimized Network Functions Virtualization (NFV) placement. It highlights the convergence of IT and telecom, the need for carrier-grade cloud solutions, and presents a smart scheduling approach to improve resource management while addressing energy efficiency and performance. The conclusion emphasizes NFV as a key use case for OpenStack and calls for community engagement in advancing scheduler capabilities.
![Conclusion
Β
β’β― NFV
Β Value
Β Proposition
Β
β’β― NVF
Β is
Β a
Β killer
Β use-Ββcase
Β for
Β Openstack
Β
Β
β’β― Call
Β for
Β community
Β action
Β
β’β― Scheduler
Β Gap
Β and
Β a
Β candidate
Β solution
Β [e.g.
Β SolverScheduler,
Β blueprint
Β exists,
Β
code
Β pushed
Β for
Β review
Β in
Β Icehouse]
Β
β’β― Cross-ΒβScheduler
Β API
Β w.
Β constraints
Β [e.g.
Β augment
Β server-Ββgroups
Β API
Β released
Β in
Β
Icehouse]
Β
β’β― Neutron
Β hooks
Β for
Β Virtual
Β Network
Β Services
Β (and
Β API)
Β
Β](https://image.slidesharecdn.com/optimizednfvplacement-final-140515160341-phpapp01/85/Optimized-placement-in-Openstack-for-NFV-17-320.jpg)
Optimized placement in Openstack for NFV
- 1. Increasing Β Infrastructure Β Ef/iciency Β via Β Optimized Β NFV Β Placement Β in Β OpenStack Β Clouds Β Yathiraj Β Udupi, Β Debo Β Dutta Β β Β Cisco Β Ram Β (Ramki) Β Krishnan Β -Ββ Β Brocade Β Β OpenStack Β Atlanta Β Summit, Β Β May Β 2014 Β
- 2. Who and why? Debo/Yathi Β -Ββ Β Cisco Β Cloud Β CTO Β ofLice Β Ramki Β -Ββ Β Brocade Β CTO Β ofLice Β Goal: Β Drive Β Innovative Β Open Β Source Β solutions Β for Β NFV Β with Β OpenStack Β
- 3. Our Thesis β’β― Toby Β Ford@AT&Tβs Β NFV Β talk Β on Β Tue, Β May Β 13th Β β’β― Worlds Β of Β IT Β and Β Telco Β are Β coming Β together Β Β β’β― Telco Β Cloud Β -Ββ Β OpenStack Β as Β the Β infrastructure Β foundation Β β’β― Goal: Β Transform Β OpenStack Β to Β a Β Carrier-Ββgrade Β cloud Β solution Β Β β’β― We Β deep Β dive Β into Β some Β high-Ββlevel Β gaps Β Toby Β identiLied Β β’β― We Β demo Β some Β initial Β progress Β
- 4. Agenda Β β’β― NFV Β Summary Β β’β― Cloud Β NFV Β Use Β Case Β β’β― Drive Β Innovation Β -Ββ Β EfLicient Β Resource Β Placement Β Strategies Β β’β― Extensions Β to Β OpenStack Β scheduler Β β’β― Conclusion Β Β
- 5. Network Functions Virtualization (NFV) NFV Vision Source: ETSI NFV White Paper β’β― Global Β movement Β by Β network Β operators Β -Ββ Β AT&T, Β Verizon, Β BT, Β CenturyLink, Β Deutsche Β Telekom, Β Telefonica, Β KDDI Β etc. Β β’β― General Β purpose Β Β hardware Β -Ββ Β OPEX Β and Β CAPEX Β savings Β β’β― Increased Β automation Β β Β OPEX Β savings, Β faster Β time Β to Β market Β β’β― New Β business Β models Β and Β value Β added Β services
- 6. NFV Use Case - NFVIaas Motivation Β Β β’β― Network Β Functions Β in Β the Β cloud Β β’β― Combined Β value Β β Β Infrastructure Β as Β a Β service Β (IaaS) Β β Β Compute/storage Β infra, Β Network Β as Β a Β service Β (NaaS) Β β Β WAN Β network Β infra Β β’β― Leverage Β NFV Β Infra Β of Β another Β SP Β β Β increase Β resiliency, Β reduce Β latency Β (CDN), Β regulatory Β requirements Β Β Β Where Β are Β we Β are Β today Β ? Β β’β― Compute/storage Β is Β treated Β independent Β of Β network, Β no Β energy Β efLiciency Β considerations Β β’β― Service Β value Β is Β not Β maximized Β NFV Use Case β NFVIaaS Source: ETSI NFV Use Cases
- 7. NaaS Virtualized Β Network Β Bandwidth Β Bandwidth Β Virtual Β Machine Β Virtual Β Machine Β Virtual Β Machine Β Virtual Β Machine Β WAN Bandwidth on Demand Data Β Center Β 1 Β Data Β Center Β 2 Β BeneLits Β β’β― Use Β WAN Β bandwidth Β as Β needed, Β avoid Β Lixed Β cost Β due Β to Β reservation Β (typically Β 1.5 Β times Β peak) Β β Β typically Β leverage Β MPLS Β technologies Β β’β― Popular Β use Β cases Β -Ββ Β Disaster Β Recovery Β , Β On-Ββdemand Β backup Β across Β WAN Β StorageStorage
- 8. NFVIaas (IaaS+NaaS) Virtualized Β Network Β Bandwidth Β Bandwidth Β Virtual Β Machine Β Virtual Β Machine Β Virtual Β Machine Β Virtual Β Machine Β Data Β Center Β 1 Β Data Β Center Β 2 Β Where Β do Β we Β want Β to Β get Β to Β ? Β β’β― Beyond Β WAN Β Bandwidth Β savings Β β’β― Optimal Β resource Β placement Β across Β DCs Β -Ββ Β Increase Β Energy Β efLiciency Β while Β maintaining Β multi-Ββtenant Β fairness Β and Β improving Β performance Β β Β CAPEX/OPEX Β savings, Β Improve Β QoE, Β Address Β regulatory Β requirements Β β’β― Popular Β use Β cases Β -Ββ Β Disaster Β Recovery, Β On-Ββdemand Β backup Β across Β WAN, Β Virtualized Β CDN Β Compute/Storage/WAN Bandwidth on Demand + Energy Efficiency Storage Storage
- 9. NFVIaas (IaaS+NaaS) β’β― Power Β usage Β in Β DCs Β -Ββ-Ββ Β servers Β Γ ο Β heavy Β hitter Β β’β― Server Β power Β proLiles Β typically Β non-Ββlinear; Β ~45% Β of Β peak Β power Β with Β ~20% Β of Β offered Β load; Β ~30% Β power Β in Β idle Β state Β β’β― InefLicient Β to Β keep Β servers Β powered Β on Β under Β low Β load Β conditions Β Energy Efficiency Issues SPEC Β Benchmark Β results: Β HP Β ProLiant Β DL380p Β rack Β server Β Source: Β http://i.dell.com/sites/doccontent/shared-Ββcontent/data-Ββsheets/en/Documents/Comparing-ΒβDell-ΒβR720-Ββand-ΒβHP-ΒβProliant-ΒβDL380p-ΒβGen8-ΒβServers.pdf Β
- 10. NFV β Huge opportunity for Openstack Energy Β aware Β joint Β scheduling Β of Β compute/storage/networking Β resources Β β Β example Β below Β β’β― NFV Β Customer Β submits Β job Β request, Β e.g. Β backup, Β with Β elasticity Β windows Β β’β― NFV Β Provider Β returns Β back Β information Β about Β time Β window Β to Β schedule Β backup Β β’β― Trigger Β other Β events Β e.g. Β Consolidate Β workloads; Β Finish Β one Β job Β and Β start Β and Β the Β next; Β Power Β down Β resources Β (especially Β servers) Β after Β job Β completion Β Β Β Our Β Solution: Β Smart Β Scheduler Β in Β Openstack Β How do we get there ? Solver Β Β Scheduler Β Adapted Β from Β ETSI Β NFV Β Architectural Β Framework Β
- 11. Users: Minimize costsβ¦ (Energy & Network Efficiency) Maximize Performance... Infrastructure: State (BigData?) (Storage/Network/Compute state, Energy Profiles, Policy/ constraints etc.) Smart Scheduling in Smart Scheduling in OpenStack for Optimized NFV Resource Placements Our Solution Smart Scheduler in Openstack β’β― Use analytics to determine current state of the Openstack deployment. β’β― Use resource management techniques to pick resources based on business constraints
- 12. Candidate Solution: Unified Constraints-based Scheduling A Smart Resource Placement Engine β’β― Unified constraints involving network, storage, compute, energy, etc. β’β― Global state + analytics β’β― Blazing fast implementations via Apache licensed third-party Solver libraries Sources: β’β― https://docs.google.com/document/d/1IiPI0sfaWb1bdYiMWzAAx0HYR6UqzOan_Utgml5W1HI/edit β’β― https://github.com/CiscoSystems/nova-solver-scheduler
- 13. Solver Scheduler: Smart Scheduling in OS Intelligent Placement Engine Plug in Plug in Scheduling Decision Cost Functions Constraint Functions Users: Minimize costsβ¦ (Energy & Network Efficiency) Maximize Performance... Infrastructure: Server State... Energy Profilesβ¦ Network Link Capacitiesβ¦ System Capacity...Sources: β’β― https://docs.google.com/document/d/1IiPI0sfaWb1bdYiMWzAAx0HYR6UqzOan_Utgml5W1HI/edit β’β― https://github.com/CiscoSystems/nova-solver-scheduler
- 14. An Example LP Problem Formulation Supply Β Demand Β Cost Β Metric Β to Β Β Minimize Β Constraints Β to Β satisfy Β Cost Β Variables Β to Β solve Β Scheduling can be Complex
- 15. DEMO: Smart Scheduling for NFV Service VMs with Compute/Storage Affinity Constraints Applicable Scenarios: 1.β― CDN NFV Service VMs that need data on certain storage volumes, on physical servers that are on or closest to the data. 2.β― Backup NFV Service VMs placement. Multinode devstack setup: -β― Host-1: (Controller, Compute node) -β― Host-2: (Compute node with demo_vol_1 Volume) -β― Host-3: (Compute node with demo_vol_2 Volume) Boot 2 VMs specifying the requested volumes to be close in proximity Results: Optimal placement by picking the two physical volume hosts: Host-2 and Host-3. Host-Ββ3: Β Β Host-Ββ2: Β Β Host-Ββ1: Β demo_vol_1 Β demo_vol_2 Β
- 16. Demo: Smart Scheduling with Compute-Storage Affinity
- 17. Conclusion Β β’β― NFV Β Value Β Proposition Β β’β― NVF Β is Β a Β killer Β use-Ββcase Β for Β Openstack Β Β β’β― Call Β for Β community Β action Β β’β― Scheduler Β Gap Β and Β a Β candidate Β solution Β [e.g. Β SolverScheduler, Β blueprint Β exists, Β code Β pushed Β for Β review Β in Β Icehouse] Β β’β― Cross-ΒβScheduler Β API Β w. Β constraints Β [e.g. Β augment Β server-Ββgroups Β API Β released Β in Β Icehouse] Β β’β― Neutron Β hooks Β for Β Virtual Β Network Β Services Β (and Β API) Β Β