gone wild
play

Gone WILD Richard Wang, Dana Butnariu, Jennifer Rexford Key - PowerPoint PPT Presentation

Openflow-based Server Load Balancing Gone WILD Richard Wang, Dana Butnariu, Jennifer Rexford Key Tradeoffs Load Balancing 1. Fast Enough 2. Scalable Popular 3. Programmability Service 4. Cost Replica 1 C1 C2 C C3 Replica 2 Server


  1. Openflow-based Server Load Balancing Gone WILD Richard Wang, Dana Butnariu, Jennifer Rexford

  2. Key Tradeoffs Load Balancing 1. Fast Enough 2. Scalable Popular 3. Programmability Service 4. Cost Replica 1 C1 C2 C C3 Replica 2 Server Internet Replica 3 Data Center

  3. Key Tradeoffs Software LB 1. Fast Enough 2. Scalable 3. Programmability 4. Cost Replica 1 C1 C2 C C3 Replica 2 Internet Replica 3 Data Center

  4. Key Tradeoffs Dedicated HW 1. Fast Enough 2. Scalable 3. Programmability 4. Cost Replica 1 C1 C2 C $ $ C3 Replica 2 Internet Replica 3 Data Center

  5. Openflow Load Balancing S1 S3 37.5% Replica 1 50% NOX Replica 2 Controller Internet S2 S4 12.5% Replica 3 Data Center

  6. Naiive Solution Challenges 1. # Microflow Rules 2. Controller Load S1 S3 37.5% Replica 1 C1 50% NOX Replica 2 C2 Controller Internet S2 S4 12.5% Replica 3 Data Center

  7. Proactively Install Rules Issues 1. Wildcards Rules 2. Connection Affinity S1 S3 37.5% Replica 1 C1 50% NOX Replica 2 C2 Controller Internet S2 S4 12.5% Replica 3 Data Center

  8. Wildcard Rules on Client IPs 0 1 111*  R3 110*  R2 101*  R2 1 0 0 1 100*  R2 011*  R2 010*  R1 0 1 1 0 0 0 1 1 001*  R1 00*  R1 000*  R1 R1 R1 R1 R2 R2 R2 R2 R3 0 1 1 1 0 Switch 0 0 1 0 0 0 1 0 1 1 1 0 0 1 1 1 0 1 0 * * * * * * * * 1 3 4 Replica 1 Replica 2 Replica 3 8 8 8 37.5% 12.5% 50%

  9. Minimizing Wildcard Rules 0 1 111*  R3 110*  R2 R2 101*  R2 1 0 100*  R2 1 011*  R2 1*  R2 * R1 010*  R1 011*  R3 1 0 0 00*  R1 0 R1 R3 * Switch 0 0 1 1 0 1 * * 1 3 4 Replica 1 Replica 2 Replica 3 8 8 8 37.5% 12.5% 50%

  10. Connection Affinity C3 C2 C2  Replica 2 C1 C1  Replica 1 00*  Controller 00*  Replica 1 00*  Replica 2 C2  Replica 2 C1  Replica 1 Switch 0*  Replica 1 0*  Controller 0*  Replica 2 Old Rule Transition Rule New Rule Time

  11. Evaluation Old Rule After Before 3 Transition Replica 1 8 1 Replica 2 8 New Rule 4 0 Replica 3 8 36 Clients, 16 MB File, WGET every 10 s

  12. Multiple Switches S1 S3 37.5% Replica 1 C1 50% NOX Replica 2 C2 Controller Internet S2 S4 12.5% Replica 3 Minimize Number of Rules Opportunistically Combine Wildcard Rules

  13. Conclusion • Proactively Installing Rules in Openflow Transition Multiple Wildcards Rules Rules Switch 1. Connection Affinity 2. Effective Distribution Non-Uniformity 3. Fast Enough 4. Scalable Controllers 5. Programmability 6. Cost

  14. Questions?

Recommend


More recommend