Requirement analysis of vim platform reliability in a three-layer decoupling environment for nfv

Gil Hellmann, Wind River, Xuesong Wang, Wind River, Qiao Fu, China Mobile, Jinglong Lv, China Mobile

A traditional non-virtualization environment is fully integrated with the hardware platform, software platform, and service application as indivisible whole. It makes failure detection and HA much easier in north - south directions. In a virtualization environment, especially three-layer decoupling, the hardware platform, NFVI/VIM, VNF and even MANO may all come from separate vendors. This provides operators with more choices and flexibility. However, it also introduces challenges on how to achieve the same level HA as with non-virtualization environment. In this session, we will cover the following:

• Common reliability strategy of traditional telecom network elements

• The general HA mechanisms of VIM (OpenStack) platform today

• What new requirements are brought to a VIM platform in the ETSI-MANO structure for a telecom-level high availability

• The impact of various types of failures on the VMs and business programs, and some fault handling strategies performed by VIM and MANO according to our test results.

1. NFV Reliability Analysis in Three-
Layers Decoupling Environment
Gil Hellmann, Wind River –Solutions Readiness
XuesongWang, Wind River –Solutions Readiness
Fu Qiao, China Mobile Research Institute –Network Department
JinglongLv, China Mobile Research Institute –Big Data Department

2. Topics
•Reliability in Non-NFV Platform
•NFV Three-Layers Decoupling Environment
•NFV HA architecture in Three-Layers Decoupling Environment
•NFVI/VIM Role in Three-Layers Decoupling Environment
•Single Layer VS Multi-layer Composed NFV End-To-End Reliability
Solution
•Platform Reliability Scenarios
•Conclusion

3. 3 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
RELIABILITY IN NON-NFV PLATFORM
§Different layers are integrated to each other closely by APIs
§Once a fault happens in specific layer, other layers can detect it quickly
§It is very hard to introduce 3rd party application
Application
FirmwareHardware
Platform
Application
FirmwareHardware
platform
High Availability

4. 4 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
NFV THREE-LAYERS DECOUPLING ENVIRONMENT
oss
NFVI
NFVO
Orchestrator
VNF
Manager(s)
Virtualization Layer
Virtual
Computing
Virtual
Computing
Virtual
Computing
Hardware Resource
Computing
Hardware
Network
Hardware
Storage
Hardware
EMS 1
VNF 1
EMS 1
VNF 1
EMS 1
VNF 1
VNF
Manager(s)VNF
Manager(s)
Service, VNF,
and
Infrastructure
Description
Virtualized
Infrastructure
Manager(s) (VIM)
•Hardware Layer
•Platform Layer
•Application Layer

5. 5 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
NFV RELIABILITY ARCHITECTURE IN THREE-LAYERS
DECOUPLING ENVIRONMENT
From bottom to top, Layered Approach to NFV High Availability
§NFVO is the brain of the system High Availability
§Various layers providing different aspects of overall Carrier Reliability.
NFV platform reliability can’t depend on only one layer.
§No cascading failures between layers
§Layers detect & recover from failures in adjacent layers with NFVO
authorization
NFVO
Hardware
VIM / NFVI
Application / VNFM / VNFC(VM)

6. 6 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
NFVI/VIM ROLE IN THREE-LAYERS DECOUPLING
ENVIRONMENT
§VIM directly or indirectly detect various failure
–Guest, OpenStack, Hardware, etc…
§Orchestration/NFVO can authorize VIM to deploy fastest HA solution at
first time with pre-defined behavior

7. 7 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
Single-layer VS Multi-layer Based NFV End-To-End
Reliability Solution
Solution 1：VNF LayerOnly
Reliability Solution
Solution 2：NFVI/VIM LayerOnly
Reliability Solution
Solution 3:Application+ NFVI/VIM
Layer Reliability Solution
Application
Assurance
•Application checker•VM health checker and / or Guest
health checker by VIM
•Application health checker & VM health
checker and / or Guest health checker by
VIM
•Application failover•VM failover (reboot or evacuation)•Application failover
•Graceful VM failover
•Quick switchover as HA only
application data
•Slow switchover as HA OS +
Application
•Quick switchover as HA only application
data
•Hot standby•No Hot Standby•Hot standby
•Specific solution per application•Generic solution for any application/OS•Integration per application
•No VM level HA for future failover•VM level HA for future failover•Specific solution per application +
GenericVM level HA for future failover
NFVI/VIM
Platform
Assurance
•No Detection on hardware or host•Detection on hardware failure and host
failure
•Detection on hardware failure and host
failure
•No failure isolation or debugging
ability
•Failure isolation and debugging ability•Failure isolation and debugging ability
•Forecast failure ability (hardware failure)•Forecast failure ability(hardware failure)

8. 8 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
PLATFORM RELIABILITY SCENARIOS
§Failure scenarios
–Node Power Failure
–Node Management Link Failure
–Service Network Failure
–Hardware Failure
–VNF Failure
–Hypervisor Failure
§We will exam each one from 2 perspectives:
–Recovery Driven by NFVO Layer
–Recovery Driven by VIM Layer

9. 9 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
NODE POWER FAILURE SCENARIO
Recovery Driven by NFVO Layer
NFV Orchestrator
Compute Node
KVM
+
Virtual Switch
Compute Node
KVM
+
Virtual Switch
VNFVNF
VIM
Control Node
VNF Manager
IPMI
OpenStack Management
ETSI-MANO
Compute Node
KVM
+
Virtual Switch
VNF HA
VNF

10. 10 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
NODE MANAGEMENT LINK FAILURE SCENARIO
Recovery Driven by NFVO Layer
NFV Orchestrator
Compute Node
KVM
+
Virtual Switch
VIM
Control Node
VNF Manager
IPMI
OpenStack Management
ETSI-MANO
VNF HA
Compute Node
KVM
+
Virtual Switch
VNF
Compute Node
KVM
+
Virtual Switch
VNFVNF

11. 11 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
HARDWARE FAILURE SCENARIO
Recovery Driven by NFVO Layer
NFV Orchestrator
VIM
Control Node
VNF Manager
IPMI
OpenStack Management
ETSI-MANO
VNF HA
Compute Node
KVM
+
Virtual Switch
VNF
Compute Node
KVM
+
Virtual Switch
VNFVNF
Compute Node
KVM
+
Virtual Switch

12. 12 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
SERVICE NETWORK / VNF FAILURE SCENARIOS
Recovery Driven by NFVO Layer
NFV Orchestrator
VIM
Control Node
VNF Manager
IPMI
OpenStack Management
ETSI-MANO
VNF HA
Compute Node
KVM
+
Virtual Switch
VNF
Compute Node
KVM
+
Virtual Switch
VNFVNF
Compute Node
KVM
+
Virtual Switch

13. 13 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
HYPERVISOR FAILURE SCENARIO
Recovery Driven by NFVO Layer
NFV Orchestrator
VIM
Control Node
VNF Manager
IPMI
OpenStack Management
ETSI-MANO
VNF HA
Compute Node
KVM
+
Virtual Switch
VNF
Compute Node
KVM
+
Virtual Switch
VNFVNF

14. 14 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
NODE POWER FAILURE SCENARIO
Recovery Driven by VIM Layer
NFV Orchestrator
Compute Node
KVM
+
Virtual Switch
Compute Node
KVM
+
Virtual Switch
VNFVNF
VIM
Control Node
VNF Manager
IPMI
OpenStack Management
ETSI-MANO
Compute Node
KVM
+
Virtual Switch
VNF HA
VNF

15. 15 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
NODE MANAGEMENT LINK FAILURE SCENARIO
Recovery Driven by VIM Layer
NFV Orchestrator
Compute Node
KVM
+
Virtual Switch
VIM
Control Node
VNF Manager
IPMI
OpenStack Management
ETSI-MANO
VNF HA
Compute Node
KVM
+
Virtual Switch
VNF
Compute Node
KVM
+
Virtual Switch
VNFVNF

16. 16 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
HARDWARE FAILURE SCENARIO
Recovery Driven by VIM Layer
NFV Orchestrator
VIM
Control Node
VNF Manager
IPMI
OpenStack Management
ETSI-MANO
VNF HA
Compute Node
KVM
+
Virtual Switch
VNF
Compute Node
KVM
+
Virtual Switch
VNFVNF
Compute Node
KVM
+
Virtual Switch
Graceful shutdown

17. 17 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
SERVICE NETWORK / VNF FAILURE SCENARIOS
Recovery Driven by VIM Layer
NFV Orchestrator
VIM
Control Node
VNF Manager
IPMI
OpenStack Management
ETSI-MANO
VNF HA
Compute Node
KVM
+
Virtual Switch
VNF
Compute Node
KVM
+
Virtual Switch
VNFVNF
Compute Node
KVM
+
Virtual Switch
Guest heartbeat

18. 18 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
HYPERVISOR FAILURE SCENARIO
Recovery Driven by VIM Layer
NFV Orchestrator
VIM
Control Node
VNF Manager
IPMI
OpenStack Management
ETSI-MANO
VNF HA
Compute Node
KVM
+
Virtual Switch
VNF
Compute Node
KVM
+
Virtual Switch
VNFVNF

19. 19 WHEN IT MATTERS, IT RUNS ON WIND RIVER.© 2017 WIND RIVER. ALL RIGHTS RESERVED.
CONCLUSION
§In 3 Layers decoupling, NFV reliability can not only depends on only one
layer.Various layers provides different aspects for Carrier Reliability,
where NFVI/VIM plays an important role.
§For the scenarios described in this presentation, tradeoff exists between
NFV reliability and cost:
–When NFVO is responsible to make all recovery decision it can result in high
latency under heavy load
–Recovery performance will be faster and lower latency when NFVO authorize
VIM to make some recovery decisions

20. Thanks You
谢谢