1. 1 Minimization of Network Power Consumption with Redundancy Elimination T. Khoa Phan* Joint work with: Frédéric Giroire*, Joanna Moulierac* and Frédéric Roudaut** * Mascotte project, ** Orange Lab IFIP Networking 2012 21-25 May 2012, Prague Czech Republic

2. 2 Motivation Redundancy Elimination (RE) Reducing traffic load Energy-aware routing (EAR) Reducing energy EAR + RE Reducing traffic load + Reducing more energy

3. 3  Measurements on energy consumption on routers show:  Small influence of traffic load [1]  Important parameter is the number of switch-on interfaces [1]  Energy-aware routing (EAR):  Routing solution minimizing the number of active links while satisfying link capacity constraint [2][3]  Simplified router architecture: a link is turned off means two interfaces of routers are turned off Energy Savings for Backbone Network [1][2][3] [Chabarek et al., INFOCOM’08; Chiaraviglio et al. IEEE/ACM ToN’11; Zhang et. al ICNP’10]

4. 4 Energy-aware routing (EAR) 0 12 3 45 7 6 9 8 11 1213 10 D 11,16 = 5 D 0,5 = 5 10 14 15 16 10 Link capacity Traffic demands: - From 0 → 5 - From 11 → 16 traffic volume 5 Energy-aware routing – turn off 10/18 links

5. 5 Traffic Redundancy Elimination cache  Is it to possible reduce traffic load on the Internet? –Anand et. al SIGCOMM’08 [4], SIGCOMM’09 [5] –Song et. al ICCCN’10 [6] → up to 50% of traffic can be eliminated

6. 6 Traffic Redundancy Elimination  Does it work for the real world? WAN Optimization Controllers (WOC) from Cisco, Juniper, etc. - Redundancy Elimination, TCP acceleration, Application acceleration [7][8] WAN WOC Synchronized compression database

7. 7 Energy-aware routing with RE-routers Energy-aware routing with RE – 50% of traffic can be compressed – turn off 10/18 links 01 2 3 45 7 6 9 8 11 1213 10 Compressed demand D 11,16 = 5 Compressed demand D 0,5 = 5 10 14 15 16 10 Traffic demands: - From 0 → 5 - From 11 → 16 with traffic volume 10 Link capacity Normal router RE-router

8. 8 Energy consumption of WOC Testbed with WOCs WOC and Watt-meter devices

9. 9 Energy consumption of WOC Single long live FTP connection Parallel FTP connections

10. 10 Optimal solution - Integer Linear Program: Objective function: Flow conservation constraint: Capacity constraint: RE-router constraint: Binary variable: x uv : link (u,v) is used or not w u : RE is enabled at “u” or not normal flow from s → t compressed flow from s → t compression ratio Energy-aware routing with RE-routers [2] [Chiaraviglio et al. IEEE/ACM ToN’11] [*]: Energy consumption of a WOC [2] *

11. 11 Heuristic algorithm: 2 steps  Step 1: all routers enable RE - find feasible routing for compressed demands with small number of active links  Algorithm 1: using shortest path → feasible routing  Algorithm 2: remove less loaded link and repeat Algorithm 1  Step 2: enable only necessary RE-routers → disable RE on the others to save energy Energy-aware routing with RE-routers

12. 12 Simulation results Atlanta network – 15 nodes, 22 edges Energy savings (%) Capacity/demand ratio Low traffic High traffic

13. 13 Simulation results Number of RE-routers Route length Capacity/demand ratio Number of RE-routers Route length (hop)

14. 14 Simulations for 10 general networks Simulation results Network topology |V| |E| Energy savings with different traffic volumes With RE-routerWithout RE-router highmediumlow highmediumlow Atlanta15 22 23.6%30.1%36.4%0%32%36% New York16 49 52.2%61.6%65.2%2%59%63% Nobel Germany17 26 23.9%33.9%36.8%0%35%39% France25 45 33.8%44%45.4%0%42%44% Norway27 51 36.2%45%47.5%12%43%47% Nobel EU28 41 27.7%30.9%34.2%12%32%34% Cost26637 57 25.3%33.6%35%3.5%32%35% Giul3939 86 36.5%48.4%51.1%0%45%50% Pioro4040 89 45.3%53.2%54.5%0%53%54% Zib5454 80 23.2%31.2%32.7%0%30%33%

15. 15 Conclusions To the best of our knowledge, this is the first work that considers energy-aware routing with RE. Real experiment to show power consumption of the WOC The problem is defined and modeled using Integer Linear Program (ILP) Propose heuristic algorithm and prove by simulations that our heuristics work well, approximate to the result of ILP. In high traffic volume, GreenRE can gain further 30% of energy savings

16. 16 Future work Other versions of the problem: –Some RE-routers have already been placed on the network → utilize the RE-routers for aggregating traffic –There are a limited number of RE-routers → find the best location to place the RE-routers on the topology More simulations on real network topology with real traffic demands and real redundancy distribution.

17. 17 Thank you for your attention! Q&A

18. 18 References [1] M. Gupta and S. Singh. “Greening of the Internet”. In Proceedings of ACM SIGCOMM, 2003 [2] L. Chiaraviglio; M. Mellia and Fabio Neri. “Minimizing ISP Network Energy Cost: Formulation and Solutions”. In IEEE/ACM Transactions on Networking, 2011. [3] M. Zhang; C. Yi; B. Liu and B. Zhang. “GreenTE: Power-aware Traffic Engineering”. In Proceedings of IEEE International Conference on Network Protocols (ICNP), 2010. [4] A. Anand et al. “Packet Caches on Routers: The Implications of Universal Redundant Traffic Elimination”. In Proceedings of ACM SIGCOMM, 2008. [5] A. Anand et al. “SmartRE: An Architecture for Coordinated Network-wide Redundancy Elimination”, In Proceedings of ACM SIGCOMM, 2009. [6] Y. Song; K. Guo and L. Gao. “Redundancy-Aware Routing with Limited Resources”. In Proceedings of ICCCN, 2010. [7] “BlueCoat: WAN Optimization”. http://www.bluecoat.com/ [8] T. Jr. Grevers and J. Christner. “Application Acceleration and WAN Optimization Fundamentals”. In Cisco Press, 2007

19. 19 Case 1: All routers are WOC-routers Parameters in the simulations:  Demands all-to-all  50% of traffic redundancy  Capacity/demand ratio (λ) Complete graph 5 vertices λ = 2 λ = 6 5 links can be turned off

20. 20. Non-redundant data with new signatures Redundant data replaced by signatures Encoded message

21. 21 Redundancy Elimination in Routers  There are two key challenges that hinder the deployment of redundancy elimination on routers:  Heavy computation  Large amount of memory for local database  Anand et al. in SIGCOMM’08 [4] and SIGCOMM’09 [5] → algorithm for eliminating traffic redundancy at router  Compression at 2.2Gbps  Decompression at 10Gbps (the prototype works on desktop 2.4GHz and 1GB for local database)

22. 22 Redundancy Elimination in Routers  Energy-aware routing → find routing solution for the core network  How much traffic load can be reduced if RE is integrated in routers across the Internet?  Real traffic traces from: 11 corporate network in US [4] a large university in US [5] 5 sites of a large corporate network in North America [6] → up to 50% of the traffic can be eliminated  A further 10-25% traffic load can be reduced when considering redundancy-aware routing algorithm [4]

23. 23 Energy-aware routing with WOC-routers Three cases of the problem:  Case 1: all routers on the network are RE-routers → only enable the necessary RE-routers  Case 2: some RE-routers have already been placed on the network → utilize the RE-routers for aggregating traffic  Case 3: there are a limited number of RE-routers → find the best location to place the RE-routers

24. 24 0 1 2 3 4 5 7 6 9 8 11 1213 10 Compressed demand D 11,16 = 5 Compressed demand D 0,5 = 5 10 14 15 16 10 Traffic demands: - From 0 → 5 - From 11 → 16 with traffic volume 10 Link capacity Normal router RE-router Step 1 – Algorithm 1: Find feasible routing Step 1 – Algorithm 2: Remove less loaded link and find feasible routing Step 2 – Disable unnecessary RE-routers Normal demand D 0,5 = 10 Normal demand D 11,16 = 10 Energy-aware routing with RE-routers