BUFFERDeBLOAT A SURVEY OF DIFFERENT WAYS TO COUNTER-BUFFERBLOAT - PowerPoint PPT Presentation

BUFFERDeBLOAT A SURVEY OF DIFFERENT WAYS TO COUNTER-BUFFERBLOAT

Problem Problem Problem Simulation Setup Result Analysis Suggestion Bufferbloat is the existence of excessively large (bloated) buffers in systems, AFFECTED particularly network communication systems. VoIP Gaming Video chat + DARK BUFFER High latency and jitter Reduce throughput Hard to detect Larger buffer trend

Problem Problem Problem Simulation Setup Result Analysis Suggestion CONGESTION Previous study shows that CONTROL the combination of solutions at router and end-to-end Saturate the bottleneck doesn’t work as supposed. Keep / add low delay Response fast to congestion Yield quickly to TCP flows + ACTIVE QUEUE MANAGEMENT Avoid congestion Avoid global synchronization

Simulation Problem Simulation Setup Result Analysis Suggestion Simulation SIMULATION ANALYSIS OPTIMIZATION Use extensive parameters Who are influential factors Avoid configuration traps Avoid most uncertainties What are their impact Choose better parameters Output evaluation matrix When do they matter Advise on future design Plot intuitive graphs How they interact Why they cause problem

F A I R Setup QLEN TCP% L E F F DROP LINUX NS2.33 PYTHON VEUSZ Problem Simulation Setup Result Analysis Suggestion Setup SYSTEM EVALUATION Bottleneck link efficiency η = 𝑦 𝑗 /𝐷 𝑗 TCP breakdown 𝑈𝐷𝑄% = 𝑦 𝑗 𝑦 𝑘 / 𝑘∈𝑈𝐷𝑄 𝑗 Average queue length 𝑚 = 𝑏𝑤𝑕(𝑚 𝑗𝑜𝑡𝑢 ) 2 / 𝑂 ∙ 𝑦 𝑗2 Jain fairness index 𝐺 = 𝑦 𝑗 𝑗 𝑗 Packet drop rate 𝐸𝑆𝑈% = 𝑞 𝑘 / 𝑞 𝑗 𝑘∈𝐸𝑆𝑃𝑄 𝑗

6ms * 5 F I X 4ms 6ms 8ms 10ms C C P WLD B U F C A P D E L 0.25 Setup 0.5 1 2 4 6 8 10 100 2ms RUN: 10 300 10% 400 DropTail RED CHOKe DRR SFQ CoDel AQM ~ FLOW: 5 Reno + 5 CCP (+10%) ~ 100% TCP-Reno Ledbat TCP-LP TCP-NICE (pkts/Mbps) TIME: 60s 200 Problem Simulation Setup Result Analysis Suggestion Setup PARAMETERS

TCP-Reno F A I R DROP L E F F TCP% QLEN B U F + C A P AQM C C P DropTail Result Problem Simulation Setup Result Analysis Suggestion Result BUF and QLEN is positive correlated. Low CAP limit QLEN. CAP and DROP is negative correlated. QLEN DROP 350 9.00% 8.00% 300 RED 7.00% BUF=100 250 SFQ BUF=200 6.00% CoDel BUF=300 200 5.00% BUF=400 4.00% BUF=100 150 BUF=200 3.00% 100 BUF=300 2.00% BUF=400 50 Ledbat 1.00% 0 0.00% 0.25 0.5 1 2 4 6 8 10 CAP

TCP-Reno F A I R DROP L E F F TCP% QLEN B U F + C A P AQM C C P Result RED Problem Simulation Setup Result Analysis Suggestion Result CAP and QLEN/DROP is negative correlated. CAP and FAIR is positive correlated. QLEN DROP/FAIR 6.00 120.00% DropTail 5.00 100.00% SFQ 4.00 80.00% CoDel QLEN DROP 3.00 60.00% FAIR 2.00 40.00% 1.00 20.00% Ledbat 0.00 0.00% 0.25 0.5 1 2 4 6 8 10 CAP

Ledbat F A I R DROP L E F F TCP% QLEN B U F + C A P AQM C C P DropTail Result Problem Simulation Setup Result Analysis Suggestion Result QLEN/DROP is same as DropTail+TCP-Reno. CAP and TCP% is positive correlated. CAP and FAIR is negative correlated. TCP%/FAIR 100.00% 90.00% RED 80.00% SFQ 70.00% CoDel 60.00% TCP% 50.00% 40.00% FAIR 30.00% TCP-Reno 20.00% 10.00% 0.00% 0.25 0.5 1 2 4 6 8 10 CAP

Ledbat F A I R DROP L E F F TCP% QLEN B U F + C A P AQM C C P Result RED SFQ CoDel Problem Simulation Setup Result Analysis Suggestion Result DROP is same as RED+TCP-Reno. QLEN of CoDel is opposite to other AQM. Trends of TCP% and FAIR are opposite. QLEN TCP%/FAIR 25.00 100.00% DropTail 90.00% 20.00 80.00% QLEN_RED 70.00% QLEN_SFQ QLEN_CoDel 15.00 60.00% TCP%_RED 50.00% FAIR_RED TCP%_SFQ 10.00 40.00% FAIR_SFQ 30.00% TCP%_CoDel TCP-Reno FAIR_CoDel 5.00 20.00% 10.00% 0.00 0.00% 0.25 0.5 1 2 4 6 8 10 CAP

B U F QLEN C C P TCP% F A I R QLEN C A P DROP TCP% F A I R Result C A P DROP QLEN F A I R C A P DROP B U F C A P DROP L E F F TCP% QLEN B U F + C A P DROP QLEN F A I R AQM Problem Simulation Setup Result Analysis Suggestion Result AQM eliminate the impact of BUF. We choose 400pkts / 4Mbps for further experiment CAP and QLEN/DROP is negative correlated. to ensure a stable state and reasonable bufferbloat. AQM/CCP has combined impact on TCP%/FAIR. Ex. CoDel

Result CoDel SFQ DRR CHOKe RED AQM A Q M QLEN TCP% L E F F DROP F A I R Problem Simulation Setup Result Analysis Suggestion Result AQM force TCP and Ledbat into same priority. QLEN is decided by each AQM’s setting. AQM-CoDel has a higher TCP%. QLEN TCP%/FAIR/DROP 25.00 100.00% DropTail 90.00% 20.00 80.00% 70.00% QLEN 15.00 60.00% TCP% 50.00% FAIR 10.00 40.00% DROP 30.00% 5.00 20.00% 10.00% 0.00 0.00% RED CHOKE DRR SFQ CoDel

CoDel Result F A I R DROP L E F F TCP% QLEN A Q M AQM DropTail RED CHOKe DRR SFQ Problem Simulation Setup Result Analysis Suggestion Result AQM-DRR has a noticeable lower fairness. AQM-SFQ has perfect inter/intra fairness. AQM-CoDel has perfect intra fairness. FAIR DropTail RED CHOKe DRR SFQ CoDel FAIR_CCP FAIR_TCP

TCP% QLEN CoDel SFQ DRR CHOKe RED DropTail AQM A Q M Result F A I R DROP L E F F Problem Simulation Setup Result Analysis Suggestion Result AQM-SFQ has a high DROP_TCP/DROP_CCP. AQM-CHOKe/DRR behave similar. AQM-RED/CoDel/DropTail behave similar. DROP DropTail RED CHOKe DRR SFQ CoDel DROP_CCP DROP_TCP

Result A Q M QLEN TCP% L E F F DROP F A I R Problem Simulation Setup Result Analysis Suggestion Result TCP% and FAIR is closely related. We think AQM-SFQ has an a brief advantage in QLEN at this level have no bufferbloat effect. overall performance here. Use a simple addition score system to value. TCP% F_TCP + F_CCP D_TCP/D_CCP Score RED 2 198.88% 2 1.10 4 8 CHOKE 3 199.08% 2 1.88 3 8 DRR 3 180.51% 5 2.22 2 10 SFQ 3 199.93% 1 424.05 1 5 CoDel 1 198.79% 2 1.09 4 7 LEFF TCP% QLEN FAIR F_TCP F_CCP DROP D_TCP D_CCP DropTail 99.69% 98.24% 305.16 50.45% 97.41% 99.96% 2.20% 2.22% 2.86% RED 98.32% 57.29% 3.92 97.34% 99.44% 99.44% 8.77% 9.15% 8.30% CHOKE 98.17% 52.55% 5.68 99.23% 99.59% 99.49% 10.98% 13.97% 7.42% DRR 99.69% 54.05% 13.74 90.25% 97.24% 83.27% 5.15% 7.63% 3.44% SFQ 99.69% 53.21% 21.64 99.55% 99.99% 99.93% 2.81% 5.14% 0.01% CoDel 99.01% 63.50% 7.45 92.55% 99.26% 99.53% 7.02% 7.27% 6.66%

Result CoDel SFQ DRR CHOKe RED AQM C C P QLEN TCP% L E F F DROP F A I R Problem Simulation Setup Result Analysis Suggestion Result CCP-TCP-LP has a little higher TCP%. (* but lower LEFF under AQM-RED/CHOKe.) (* but lower F_CCP under even AQM-DropTail.) TCP%_RED TCP%/FAIR/DROP TCP%_CHOKe 100.00% TCP%_DRR DropTail 90.00% TCP%_SFQ TCP%_CoDel 80.00% FAIR_RED 70.00% FAIR_CHOKe 60.00% FAIR_DRR 50.00% FAIR_SFQ 40.00% FAIR_CoDel 30.00% DROP_RED 20.00% DROP_CHOKe DROP_DRR 10.00% DROP_SFQ 0.00% Ledbat TCP-LP TCP-NICE DROP_CoDel

CoDel F A I R DROP L E F F TCP% QLEN C C P AQM Result RED CHOKe DRR SFQ Problem Simulation Setup Result Analysis Suggestion Result With SFQ, value tuple is (424.05, 1.29, 85.52). We think CCP-Ledbat has an a brief advantage in CCP-Ledbat penalizes more on TCP flows. overall performance here. {D_TCP / D_CCP} 3.50 DropTail 3.00 DropTail 2.50 RED 2.00 CHOKe 1.50 DRR CoDel 1.00 0.50 0.00 Ledbat TCP-LP TCP-NICE

Result CoDel SFQ DRR CHOKe RED AQM W L D QLEN TCP% L E F F DROP F A I R Problem Simulation Setup Result Analysis Suggestion Result LEFF/QLEN and WLD is positive correlated. AQM-RED/CHOKe has more “stable” QLEN. AQM-DRR/SFQ has higher LEFF. QLEN LEFF 25.00 100.00% DropTail QLEN_RED 90.00% QLEN_CHOKe 20.00 80.00% QLEN_DRR 70.00% QLEN_SFQ 15.00 60.00% QLEN_CoDel 50.00% LEFF_RED 10.00 40.00% LEFF_CHOKe 30.00% LEFF_DRR 5.00 20.00% LEFF_SFQ 10.00% LEFF_CoDel 0.00 0.00% 1 2 3 4 5 6 7 8 9 10

CoDel F A I R DROP L E F F TCP% QLEN W L D AQM Result RED CHOKe DRR SFQ Problem Simulation Setup Result Analysis Suggestion Result DROP and WLD is positive correlated. Trends of TCP% and FAIR are opposite. TCP%/FAIR/DROP 100.00% TCP%_RED TCP%_CHOKe DropTail 90.00% TCP%_DRR 80.00% TCP%_SFQ TCP%_CoDel 70.00% FAIR_RED 60.00% FAIR_CHOKe 50.00% FAIR_DRR FAIR_SFQ 40.00% FAIR_CoDel 30.00% DROP_RED DROP_CHOKe 20.00% DROP_DRR 10.00% DROP_SFQ 0.00% DROP_CoDel 1 2 3 4 5 6 7 8 9 10

CoDel F A I R DROP L E F F TCP% QLEN W L D AQM Result RED CHOKe DRR SFQ Problem Simulation Setup Result Analysis Suggestion Result WLD has an impact on intra-protocol fairness. AQM-DRR has higher F_TCP under most WLD. F_CCP 100.00% DropTail 95.00% RED CHOKe 90.00% DRR SFQ 85.00% CoDel 80.00% 75.00% 70.00% F_TCP 70.00% 75.00% 80.00% 85.00% 90.00% 95.00% 100.00%