refining network intents for self driving networks
play

Refining Network Intents for Self-Driving Networks Arthur Selle - PowerPoint PPT Presentation

Refining Network Intents for Self-Driving Networks Arthur Selle Jacobs Ricardo Jos Pfitscher, Ronaldo Alves Ferreira, Lisandro Zambenedetti Granville UFRGS UFMS Budapest, Hungary August 24, 2018 Self-Driving Networks High-level


  1. Refining Network Intents for Self-Driving Networks Arthur Selle Jacobs¹ Ricardo José Pfitscher¹, Ronaldo Alves Ferreira², Lisandro Zambenedetti Granville¹ ¹UFRGS ²UFMS Budapest, Hungary August 24, 2018

  2. Self-Driving Networks High-level Architecture Intent Adapt Process Learn Operator Monitor Config Network ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 2

  3. Self-Driving Networks High-level Architecture Intent Adapt Process Learn Operator Monitor Config Network Focus of our work! ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 3

  4. Nowadays... ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 4

  5. Nowadays... Higher-level ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 5

  6. Nowadays... Higher-level PGA NetKat COCOON JANUS ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 6

  7. How to deploy intents expressed in natural language? 7

  8. A Network Intent Refinement using Nile 8

  9. A Network Intent Refinement using Nile 1. Receive intents expressed in natural language 9

  10. A Network Intent Refinement using Nile 1. Receive intents expressed in natural language 2. Use Nile to ask for operator feedback 10

  11. Intent Refinement By Example Experimental Service Chaining scenario, using SONATA-NFV and Mininet Original scenario ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 11

  12. Intent Refinement By Example “Please add a firewall and an IDS from Iperf client to server” Original Intent ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 12

  13. Intent Refinement By Example “Please add a firewall and an IDS from Iperf client to server” Original Intent ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 13

  14. Intent Refinement By Example DialogFlow.com “Please add a firewall and an IDS from Iperf client to server” Original Intent ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 14

  15. Intent Refinement By Example “Please add a firewall and an IDS from Iperf client to server” Extracted entities ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 15

  16. Intent Refinement By Example “Please add a firewall and an IDS from Iperf client to server” Extracted entities ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 16

  17. Intent Refinement By Example “Please add a firewall and an IDS from Iperf client to server” Extracted entities Neural Sequence to Sequence learning model, using Recursive Neural Networks. ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 17

  18. Intent Refinement By Example define intent testIntent: from endpoint('iperf client' ) to endpoint('iperf server' ) add middlebox('firewall'), middlebox( 'ids') Nile intent ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 18

  19. Intent Refinement By Example Is this what you want? define intent testIntent: from endpoint('iperf client' ) to endpoint('iperf server' ) add middlebox('firewall'), middlebox( 'ids') Nile intent YES NO ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 19

  20. Intent Refinement By Example Is this what you want? define intent testIntent: from endpoint('iperf client' ) to endpoint('iperf server' ) add middlebox('firewall'), middlebox( 'ids') Nile intent YES NO ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 20

  21. Intent Refinement By Example define intent testIntent: from endpoint('iperf client' ) to endpoint('iperf server' ) add middlebox('firewall'), middlebox( 'ids') Nile intent ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 21

  22. Intent Refinement By Example define intent testIntent: from endpoint('iperf client' ) to endpoint('iperf server' ) add middlebox('firewall'), middlebox( 'ids') Nile intent Nile compiler to SONATA-NFV commands ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 22

  23. Intent Refinement By Example # deploy vnfs vim-emu compute start -n fw <params> vim-emu compute start -n ids <params> # chain vnfs vim-emu network add -b -src iperf-c:c-eth0 -dst fw:in vim-emu network add -b -src fw:out -dst ids:in vim-emu network add -b -src ids:out -dst iperf-s:s-eth0 Compiled SONATA-NFV commands ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 23

  24. Intent Refinement By Example # deploy vnfs vim-emu compute start -n fw <params> vim-emu compute start -n ids <params> # chain vnfs vim-emu network add -b -src iperf-c:c-eth0 -dst fw:in vim-emu network add -b -src fw:out -dst ids:in vim-emu network add -b -src ids:out -dst iperf-s:s-eth0 Compiled SONATA-NFV commands ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 24

  25. Intent Refinement By Example “Please add a firewall and an IDS from Iperf client to server” Resulting scenario ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 25

  26. Evaluation (i) The accuracy we can achieve with different sizes of training datasets, aiming to find the optimal ratio between dataset size and prediction accuracy. (ii) The impact of the operator feedback on the accuracy of predictions over time to determine if it improves accuracy. ● 5 dataset sizes: 100, 500, 1000, 2000, 5000 entries. ○ ○ 20% validation split. ● We generated the datasets automatically with random sets of entities and Nile intent pairs, combining a different number of middleboxes, endpoints, traffic matching rules, time, and QoS requirements in each intent. ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 26

  27. Results (i) The accuracy we can achieve with different sizes of training datasets, aiming to find the optimal ratio between dataset size and prediction accuracy. ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 27

  28. Results (ii) The impact of the operator feedback on the accuracy of predictions over time to determine if it improves accuracy. ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 28

  29. Results (ii) The impact of the operator feedback on the accuracy of predictions over time to determine if it improves accuracy. ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 29

  30. Summary “How to deploy network intents expressed as natural language?” Using our refinement process + Nile Low-level of technical knowledge required Feedback from user allows to learn over time “What’s next?” Fully implement Nile compilation into OpenFlow and P4 backends. Further evaluate the end-to-end proposed solution. ACM SIGCOMM 2018 Afternoon Workshop on Self-Driving Networks (SelfDN 2018) 30

  31. Thank you! Arthur Jacobs asjacobs@inf.ufrgs.br github.com/NetworkIntentAssistent 31

Recommend


More recommend