QoS-aware Energy-Efficient Algorithms for Ethernet Link Aggregates - PowerPoint PPT Presentation

QoS-aware Energy-Efficient Algorithms for Ethernet Link Aggregates in Software-Defined Networks Pablo Fondo Ferreiro Miguel Rodríguez Pérez Manuel Fernández Veiga September 15, 2018 1

Context

Context Previous work on Aggregates of Energy Effjcient Ethernet Links Straightforward Solution Power ofg unused links • Slow response time • What about half used links? 2

EEE Links Active Figure 1: Energy-Effjcient Ethernet model. Retrieved from [1]. Refreshing 𝑢 r 𝑢 r 𝑢 w Active Waking up Quiet Low Power Mode Quiet Refreshing Quiet 𝑢 s Sleeping 3 • Formally IEEE 802.3az. • Low Power Idle (LPI) state. • Sleeping and waking up is not instantaneous. 100 Normalized Energy Usage (%) 80 60 40 20 EEE Link 0 0.001 0.01 0.1 1 Load ........

Problem statement Goal 4 Minimize energy consumption in bundles of EEE links leveraging SDN. λ 1 λ 2 λ 3 λ 4 λ 100 Normalized Energy Usage (%) 80 60 40 𝜇 𝑗 = 𝜇/4 20 Equitable share 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Load

Problem statement Goal 𝜇 𝑗 } 𝑙=1 ∑ 𝑗−1 1-link bundle 5 Minimize energy consumption in bundles of EEE links leveraging SDN. λ 1 λ 2 λ 3 λ 4 λ 100 Normalized Energy Usage (%) 80 60 40 𝜇 𝑗 = min {𝐷, 𝜇 − 20 1 Link Bundle Ideal Behavior 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Load

Problem statement Goal 𝜇 𝑗 } 𝑙=1 ∑ 𝑗−1 2-link bundle 5 Minimize energy consumption in bundles of EEE links leveraging SDN. λ 1 λ 2 λ 3 λ 4 λ 100 Normalized Energy Usage (%) 80 60 40 𝜇 𝑗 = min {𝐷, 𝜇 − 20 1 Link Bundle 2 Link Bundle Ideal Behavior 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Load

Problem statement Goal 𝜇 𝑗 } 𝑙=1 ∑ 𝑗−1 4-link bundle 5 Minimize energy consumption in bundles of EEE links leveraging SDN. λ 1 λ 2 λ 3 λ 4 λ 100 Normalized Energy Usage (%) 80 60 40 𝜇 𝑗 = min {𝐷, 𝜇 − 1 Link Bundle 20 2 Link Bundle 4 Link Bundle Ideal Behavior 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Load

Problem statement Goal Minimize energy consumption in bundles of EEE links leveraging SDN. Theoritical solution Presented in [2], provides a • Packet level algorithm • Assumes real time access to individual occupation of each output port SDN Solution • Cannot take real-time decisions based on queue occupation • Will use ONOS for portability 6 • Needs flow level operation

SDN Algorithm

SDN Application Main Tasks • Flow identification • Flow characterization • Port allocation 7

Flow definition Challenge Which fields of the packets will identify our flows? • We need: • Enough flows to distribute them along the bundle. • Few flows to keep flow tables small. • Flows with predictable demand. • Two alternatives: Flow tagging vs field matching . • We will aggregate IP flows: • MAC flows can be insuffjcient (e.g., transit networks). • Transport flows would be excessive. 8

Flow rate estimation Figure 2: Average error in the estimation of the flow rate for difgerent periods. 9 100 EWMA α = 0.2 EWMA α = 0.4 90 EWMA α = 0.6 EWMA α = 0.8 80 estimation error (Mbps) previous 70 60 50 40 30 20 10 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 sampling period (seconds) Use rate of previous interval with sampling rate around 0 . 2 s

Port Allocation In essence, a bin packing problem. Heuristics Greedy Corresponds to first fit decreasing. A flow level water-filling. Bounded Greedy Variation to reduce loses: |𝑔𝑚𝑝𝑥𝑡| Conservative • Balanced distribution among needed ports. • Safety margin to further avoid losses. • Note: Energy consumption raises very rapidly with traffjc load. 10 Maximum usable capacity of a link: 1 − 𝑐𝑝𝑣𝑜𝑒

Conservative Algorithm EEE energy usage rises rapidly with load. Behavior 11 Basis • Determines the number of needed links • Distributed flows evenly among the links 2-bundle link 100 Normalized Energy Usage (%) 80 60 40 20 Ideal share Conservative share 0 0 5 10 15 20 Incoming traffic load (Gb/s)

Conservative Algorithm EEE energy usage rises rapidly with load. Behavior 11 Basis • Distributed flows evenly among the links • Determines the number of needed links 2-bundle link 100 Normalized Energy Usage (%) 80 60 40 20 Ideal share Conservative share 0 0 5 10 15 20 4-bundle link Incoming traffic load (Gb/s) 100 Normalized Energy Usage (%) 80 60 40 20 Ideal share Conservative share 0 0 5 10 15 20 25 30 35 40 Incoming traffic load (Gb/s)

Conservative Algorithm EEE energy usage rises rapidly with load. Behavior 11 Basis • Determines the number of needed links • Distributed flows evenly among the links 2-bundle link 100 Normalized Energy Usage (%) 80 60 40 20 Ideal share Conservative share 0 0 5 10 15 20 8-bundle link Incoming traffic load (Gb/s) 100 4-bundle link 100 Normalized Energy Usage (%) Normalized Energy Usage (%) 80 80 60 60 40 40 20 Ideal share Conservative share 0 0 5 10 15 20 25 30 35 40 20 Incoming traffic load (Gb/s) Ideal share Conservative share 0 0 10 20 30 40 50 60 70 80 Incoming traffic load (Gb/s)

Experimental setup • Topology: Two switches connected by 5 EEE interfaces 10 GBASE-T. • We have used real traffjc traces retrieved from CAIDA [3]. • Baseline: Equitable algorithm. • Metrics: • Energy consumption • Packet losses • Packet delay 12

Results: Energy consumption (a) 32 . 5 Gbit / s trace. • Theoretical bound for the consumption of the 32 . 5 Gbit / s: 78 . 5 % . Figure 3: Normalized energy consumption (bufger = 10000 packets). (b) sampling period = 0 . 5 seconds. 13 98 100 Greedy Bounded-Greedy 96 Conservative Equitable real energy consumption (%) 94 80 energy consumption (%) 92 90 60 Greedy Bounded-Greedy 88 Conservative Equitable 86 40 84 82 20 80 78 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 6.5 13.0 19.5 26.0 32.5 sampling period (seconds) rate (Gbps)

Results: Packet losses (a) 32 . 5 Gbit / s trace. Figure 4: Packet loss percentage (sampling period = 0 . 5 seconds). (b) bufger = 10000 packets. 14 18 1.8 Greedy Greedy Bounded-Greedy Bounded-Greedy 16 1.6 Conservative Conservative Equitable Equitable 14 1.4 12 1.2 packet loss (%) packet loss (%) 10 1 8 0.8 6 0.6 4 0.4 2 0.2 0 0 10 100 1000 10000 100000 6.5 13.0 19.5 26.0 32.5 buffer size(packets) rate (Gbps)

Results: Packet delay (a) 32 . 5 Gbit / s trace. Figure 5: Average per packet delay (bufger = 10000 packets). (b) sampling period = 0 . 5 seconds. 15 4500 10000 Greedy 4000 B-Greedy Conservative average delay (microseconds) average delay (microseconds) Equitable 3500 1000 3000 Greedy 2500 Bounded-Greedy 100 Conservative 2000 Equitable 1500 10 1000 500 0 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 6.5 13.0 19.5 26.0 32.5 sampling period (seconds) rate (Gbps)

QoS-aware algorithms

Problem statement Goal Provide low-latency service while reducing energy consumption. • The previous algorithms manage to reduce energy consumption. • However, they increase the delay of the packets. • We consider now the QoS latency requirements of the flows. • Two types of traffjc: • Best-efgort. • Low-latency. • Modifications to the previous algorithms. 16

Solutions Figure 6: Spare Port. Figure 7: Two Queues. high-priority queue of the assigned ports. 2. Low-latency flows are allocated to the flows. Spare Port Two Queues 1. Apply energy-effjcient algorithm to all the 17 1. Apply energy-effjcient algorithm to empty port. 2. Low-latency flows are allocated to the most best-efgort flows. Port 1 low − latency flows best − effort flows Port 2 Port 3 Port 4 low − latency flows high − priority queue Port 1 best − effort flows low − priority queue Port 2 Port 3 Port 4

Simulations • Same topology: 5-link bundle of 10 GBASE-T EEE interfaces. • Real traces for best-efgort traffjc. • Synthetic traffjc for low-latency packets. • Baseline: Conservative algorithm. • Parameters: • Bufger = 10 000 packets. • Sampling period = 0 . 5 seconds. • Metrics: • Delay of low-latency packets. • Delay of best-efgort packets. • Energy consumption. 18

Results: Delay of low-latency packets (a) 32 . 5 Gbit / s trace. Figure 8: Average delay of low-latency packets. (b) 45 . 5 Gbit / s trace. 19 1000 1000 average delay (microseconds) average delay (microseconds) 100 100 Conservative Conservative Spare Port Spare port T wo Queues T wo queues 10 10 1 1 0.1 1 10 100 1000 0.1 1 10 100 1000 low-latency rate (Mbps) low-latency rate (Mbps)

Results: Delay of best-effort packets Figure 9: Average delay of best-efgort packets (32 . 5 Gbit / s trace). 20 1100 Conservative Spare Port 1000 T wo Queues average delay (microseconds) 900 800 700 600 500 400 300 200 0.1 1 10 100 1000 low-latency rate (Mbps)

Results: Energy consumption Figure 10: Normalized energy consumption (32 . 5 Gbit / s trace). 21 100 Conservative Spare Port T wo Queues 80 energy consumption (%) 60 40 20 0 0.1 1.0 10.0 100.0 1000.0 low-latency rate (Mbps)

Conclusions