Download presentation
Presentation is loading. Please wait.
Published byΜαθθαῖος Καψής Modified over 5 years ago
1
Using Service Function Chaining for In-Network Computation
Adrien WION (Thales/Telecom ParisTech)
2
Service Function Chaining for In-Network Computation
Current practices: Centralized control infrastructure Problems: Single Point of Failure Scalability Legacy Interoperability No Incremental Deployment Under exploitation of in-network resources Proposal: Augment the IGP and make it function aware
3
Proposed approach: Divide & Conquer
Centralized Management and Orchestration (MANO) Distributed MANO Loss of policy server => no problem Difference NFV-R and classical routers Nodes called NFV-Routers: host vSF and share inforamtion Policy server Push high level configuration 2. Distributed MANO Share local information Local autonomous decision based on shared view
4
Exploiting the existing in-network resources
If you have a large network you have an IGP (Interior Gateway Protocol)
5
Announced address are actually VNF
Idea(s): Binding a prefix to a specific function
6
Leveraging on anycast addressing
Idea(s): Binding a prefix to a specific function + Anycast Addressing Advantages Prefix to select the function IGP metric to select the function instance
7
Augmenting network layer routing
@IPIDS @IPFW @IPIDS Network view IGP Augmented View IGP View Augmented IGP topology: Service mapped to an anycast prefix Node advertise available service Routing decision taken with shared topology Routing decision is applied per flow Introduce NFV-Router Routing decision taken hop-by-hop Toy example how our routing process work
8
NFV Router Architecture & Implementation
High Level Policies D-MANO VNF 1 VNF n Virtualization … Ressource Monitor Namespaces Connector Encapsulation VNF Routing Algorithm NSH* Route Injector Router Routing IGP OSPF *Network Service Header (RFC 8300) Network
9
Highlights on evaluation results
Source: Network emulation: NFV Routers LXC container Deployed on 48 nodes cluster 10 VNF (nodes with max betweenness centrality) Routing policy: Shortest Path to next VNF Hop-by-hop routing Load balancing on VNF VNF 1 VNF 2 NFV-R 1 NFV-R 2 NFV-R 3 NFV-R 4 NFV-R 5 NFV-R 6 NFV-R 7 NFV-R 8 NFV-R 9 NFV-R 10 The higher LSA update frequency, the higher the network traffic distribution stability
10
What did we achieve ? Fully distributed framework to chain in-network function No need to rely on fast responses from a controller => Resilience, Scalability Load balancing between VNF instance Interoperability with legacy network => No need of SDN architecture, may rely on distributed routing protocol like OSPF No configuration needed for adding new VNF instances Future Work Inter-Domain Service Provisioning VNF metrics Maintenance and Failure VNF Provisioning References:
11
Backup slides
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.