1. Towards Virtual Routers as a Service 6th GI/ITG KuVS Workshop on “Future Internet” November 22, 2010 Hannover Zdravko Bozakov

2. Towards Virtual Routers as a Service Talk outline  Virtualization overview  Use case: virtual routers as a service  Problem statement  Resource allocation algorithm  Virtual router location selection  In brief  Virtual router architecture  Live-migration  Summary and outlook 2

3. Network Virtualization Overview Virtualization aims to decouple logical and physical network resources and increase network flexibility Variable mapping of physical and logical entities  Slice network hardware for multiple customers  Handle multiple network devices using a single control plane Live-migration of logical routers  Load balancing (capacity, routing tables, CPU)  Scheduled router maintenance  Energy conservation 3

4. Virtual Routers as a Service 4

5. On-demand provision of connectivity over core network  Enterprise branch offices  Regional providers  University campuses Single virtual router for edge interconnection  Reduction of customer management overhead  Consolidation of provider resources and transparent remapping Port relay nodes (PRN)  Forward traffic to virtual router (root node) 5

6. Problem Statement 6 What we have:  Backbone network  Weighted graph G with weights W  Link utilization U and capacity C (u/c)  Customer requirements  Subset of edge nodes Γ  Capacity demand D for edge nodes γ What we need:  Optimal location of VR root node R  Optimal paths from R to Γ satisfying capacity constraints W=1 R

7. Path Selection Algorithm Trivial case: unlimited backbone capacity  For each γ calculate shortest path to R (e.g. Dijkstra)  Does not work for capacity constrained networks Solution with constraints: flow network theory (successive shortest paths) 7 SRC DST w=0 C=1 w=0 C=4

8. Root Node Selection Optimal location of root node R  Minimize the cost S of bandwidth consumed by the VR links V  Root selection using total enumeration  Nodes with insufficient resources are pruned (e.g. capacity, CPU, memory) Example: optimal root node locations with total cost S=4 8

9. Virtual Router Architecture Root node: hardware accelerated virtual router  Control plane virtualization using standard VMs on commodity servers  Programmable data plane using Openflow- enabled switches Port relay nodes (PRN)  Forward packets based on L2 virtual router addresses along computed paths  Openflow implementation 9

10. Live Router Migration Virtual router architecture allows live router migration  Setup outbound PRN paths for new root node R*  Clone forwarding table from old root R and remap physical ports  Control plane continuously updates routing tables on R and R*  Asynchronously setup inbound paths for R*  Tear down old paths and root node 10 Watch the demo during the break!

11. Conclusion and Outlook Conclusion  On-demand connectivity using single logical router instance reduces management overhead  Presented approach allows optimal computation of paths to a router located within network core  Basic root node selection strategy  Architecture is capable of live-migration Outlook  Refine path selection algorithm and analyze alternative approaches  Optimize root node selection method  Detailed evaluation of live-migration performance  Implement and evaluate fallback strategies 11