FOUNDATIONS OF INTENT- BASED NETWORKING

Transcription

1 FOUNDATIONS OF INTENT- BASED NETWORKING Loris D Antoni Aditya Akella Aaron Gember Jacobson

2 Network Policies Enterprise Network Cloud Network Enterprise Network 2

3 3 Tenant Network Policies Enterprise Network Reachability: A can talk to B A B Waypoints: C to B traffic goes through a Firewall C

4 4 Cloud Network Policies Tenant 1 Network Policies Tenant 2 Network Policies 100Gbps S 1 S 2 100Gbps Network isolation: Tenant 1 and 2 s traffic must not affect each other Network resource management Fault tolerance S 3 S 4 S 5 S 6 100Gbps 100 Gbps 100Gbps S 7 S 8 S 9 S 1 0

5 5 Intent-based networking High level language to specify policies INPUT Synthesize Policy compliant network configurations OUTPUT

6 6 GENESIS SYNTHESIZING FORWARDING TABLES IN MULTI TENANT NETWORKS Kausik Subramanian [Subramanian, D Antoni, Akella, POPL17]

7 7 Software-defined Networks SDN Controller Centralized controller Programmable switch rules: Enforcing policies using conventional enforces policies Match: Packet headers distributed S 1 networks S 2 is difficult Action: Forward to next switch SSH traffic at S 3 is forwarded to S 7 SSH S 3 S 4 S 5 S 6 S 7 S 8 S 9 S 10

8 8 Support for complex and diverse policies High level language to specify policies Genesis Switch forwarding tables Enforcing certain policies is NP complete Genesis uses Satisfiability Modulo Theories (SMT) solvers to synthesize forwarding tables

9 9 Outline of the Talk Motivation Synthesis of forwarding tables in Genesis Scaling to large workloads: Tactics Genesis extensions and conclusions

10 10 Synthesis Approach High level policies + Topology INPUT Constraints on Fwd and Reach Abstract Representation (Fwd, Reach) Paths from Fwd and Reach solution Forwarding tables OUTPUT

11 11 Semantics of (Fwd, Reach) Fwd(S 1, ID) = S 2 : Switch S 1 forwards to S 2 Reach(S 2, ID) = 1: Specifies that S 2 is reachable in 1 step from source Reach(S 1, ID) = 0 Reach(S 2, ID) = 1 Reach(S 3, ID) = 2 Fwd(S 1, ID) = S 2 Fwd(S 2, ID) = S 3 S 1 S 2 S 3

12 12 Reachability Constraints SRC S 3 Reach(S 3, ID) = k 1 S 1 S 4 S 5 Reach(S 2, ID) = k 1 Reach(S 4, ID) = k S 2 Fwd(S 3, ID) = S 4 DST If a switch is reachable in k steps, one of its neighbors must be reachable in k 1 steps

13 13 Policy Constraints Waypoint: Blue Tenant specifies Reach(S 4, ID) = k path must traverse through S 4 S 2 S 4 Isolation: Blue Tenant and Red Tenant paths do not share any link S 1 S 5 Traffic Engineering: Using SMT OPT S 3 (S 3, ID1) (S 3, ID2)

14 THE END? 14

15 15 Baseline Synthesis Evaluation Setup Genesis implemented in Python, uses Z3 SMT solver Multi tenant isolation: Each tenant has a single reachability policy, and all tenant paths are mutually isolated Medium sized fat tree datacenter topologies

16 16 Baseline Synthesis Evaluation To scale to large networks and workloads, we need to further algorithmic insights and optimizations Exponential Complexity Synthesis time for over 60 tenants takes >5000s

17 17 SCALING TO LARGE WORKLOADS TACTICS

18 18 Tactics: Motivation Edge to edge paths: 272 Large search space Use network structure to specify path properties Core Aggregate Edge

19 19 Tactics as regular expressions Core No Edge Tactic: Not (Edge.* Edge.* Edge) Aggregate Edge

20 20 Tactics: Constraint Reduction Genesis uses tactics as a search strategy to eliminate constraints No Edge Tactic ensures no intermediate edge switch Reach(C 1 ) = k 1 Reach(A 1 ) = k 1 C 1 A 1 Reach(S) = k S Reach(E 1 ) = k 1 E 1

21 21 Tactics: Algorithmic Properties Specified using a restricted subset of regular expressions Sound and Complete algorithm for enforcing them Policy agnostic The operator can develop a repository of tactics based on their topology

22 22 Tactics: Evaluation Multi tenant isolation workload Valley Free Tactic and No Edge Tactic Valley Free Tactic speedup: 400x

23 23 Outline of the Talk Motivation Synthesis of forwarding tables in Genesis Scaling Genesis: Tactics and Divide and Conquer Genesis extensions and conclusions

24 24 Genesis Extensions Genesis Rich Policy Language Synthesis using SMT Resilient Paths Network Repair

25 25 Network Resilience Single path: Not resilient Cloud network Link failure S1 S2 S3 t resilience: For events under t arbitrary link failures, there exists a valid path

26 26 Policy-compliant Resiliency Backup path Cloud network 1 resilient Isolation policy S1 S2 S3 For 1 resilience, backup path must be edge disjoint from original path Sound transformation of input policies to provide t resilience

27 27 Minimal Reactive Network Repair Cloud network Policies Policies Best repair: Minimize change overhead Genesis uses MaxSMT

28 28 Network Repair Evaluation Multi tenant isolation workload One switch failure, network repair such that number of switches affected is minimized For larger workloads, repair is faster than re synthesis.

29 CONCLUSION 29

30 High level policies on paths and switches INPUT Genesis Switch forwarding tables satisfying policies OUTPUT OSPF and BGP configurations Efficient optimal repair 30