[PPT] - The New AQM Kids on the Block: An Experimental Evaluation of CoDel PowerPoint Presentation

SLIDE 1

The New AQM Kids on the Block: An Experimental Evaluation of CoDel and PIE

Naeem Khademi <naeemk@ifi.uio.no> David Ros <dros@simula.no> Michael Welzl <michawe@ifi.uio.no> 17th IEEE Global Internet Symposium – 2014 Toronto, ON, Canada

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 1 / 21

SLIDE 2

Outline

AQM AQM mechanisms considered CoDel PIE ARED Experimental Setup A Basic Test Parameter Sensitivity Conclusions and Future Work Q&A

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 2 / 21

SLIDE 3

AQM

Active Queue Management (AQM)

◮ Problem: Standard loss-based TCP’s congestion control plus

Large unmanaged buffers in Internet routers, switches, device drivers,... (a.k.a Bufferbloat)

◮ Cause: Latency issues for interactive/multimedia applications ◮ Solution: AQM tries to signal the onset of congestion by

(randomly?) dropping/marking packets

AQM Goals

◮ Maintain low average queue/latency ◮ Allow occasional packet bursts ◮ Break synchronization among TCP flows

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 3 / 21

SLIDE 4

AQM mechanisms considered

The New AQM Kids on the Block...

◮ Two very recent proposals:

◮ (FQ_)CoDel (IETF 84) ◮ PIE (IETF 85) mandatory in DOCSIS 3.1 CM

◮ Some older AQMs dating back to early 90’s/00’s (*RED, REM,

BLUE, CHOKe,...)

◮ Designed to be better than RED, just like CoDel and PIE

◮ Little academic literature available on CoDel and PIE

Literature (bold = peer-reviewed)

CoDel PIE FQ_CoDel Wired, sim [NJ12][GRT+13][WP12] [Whi13] [Whi13] [Whi13] Wired, real-life [GRT+13]

Wireless (any)
NOTE: [WP12] and [Whi13] are on DOCSIS 3.0 while [GRT+13] has tests with

Low-Priority congestion control.

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 4 / 21

SLIDE 5

AQM mechanisms considered

The New AQM Kids on the Block (cont.)

AQM Deployment Status

◮ (W)RED is available on plenty of HW but mostly "turned off"

Mentioned Reasons for Lack of Deployment

◮ Bad implementation (?) ◮ Hard to tune RED params ◮ Sally Floyd’s ARED (2001 technical report, available in Linux)

adaptively tunes RED params aiming for a certain target queuing => with fixed BW maps to a "target delay"

◮ Target delay can be set in ARED, CoDel and PIE

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 5 / 21

SLIDE 6

AQM mechanisms considered CoDel

The New AQM Kids on the Block (cont.)

CoDel – Controlling Delay

◮ Tries to detect the standing queue by measuring minimum sojourn

delay (delaymin) over a fixed-duration interval (default 100 ms)

◮ Uses timestamping ◮ If delaymin > target for at least one interval, enters dropping mode

and a packet is dropped from the tail (deque)

◮ Next dropping time: Dropping interval decreases in inverse

proportion to the square root of the number of drops since the dropping mode was entered

◮ Exits dropping mode if delaymin ≤ target ◮ No drop when queue is less than 1 MTU

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 6 / 21

SLIDE 7

AQM mechanisms considered CoDel

The New AQM Kids on the Block (cont.)

CoDel Assumptions

◮ 100 ms is nominal RTT assumed typical on the Internet paths ◮ interval = 100 ms; assures protection of normal packet bursts ◮ A small target standing queue (5% of nominal RTT) is tolerable for

achieving better link utilization

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 7 / 21

SLIDE 8

AQM mechanisms considered PIE

The New AQM Kids on the Block (cont.)

PIE – Proportional Integral controller Enhanced

◮ Lightweight as it uses delay estimation instead of timestamping ◮ Uses a Proportional Integral (PI) controller design ◮ Uses trend of latency (increasing or decreasing) over time to

determine the congestion level

◮ E[T] as current estimated queuing delay during every tupdate, N as

current queue length and µ is the draining rate E[T] = N/µ

◮ Randomly drops on enque based on probability p

p = p + α ∗ (E[T] − Ttarget) + β ∗ (E[T] − E[T]old)

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 8 / 21

SLIDE 9

AQM mechanisms considered ARED

The New AQM Kids on the Block (cont.)

ARED – Adaptive RED

◮ Tries to solve RED’s main problem of parameter tuning to keep the

average queue length (¯ N) around a desired target_queuing

◮ target_queuing = (th_max + th_min)/2 ◮ Observes ¯

N to make RED more/less aggressive

◮ Updates RED’s pmax adaptively (every 500 ms by default) using an

AIMD policy

◮ Only useful in fixed-BW scenarios

(target_delay = target_queuing/BW)

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 9 / 21

SLIDE 10

Experimental Setup

◮ Traffic: 60 sec (or 300 sec if RTT=500 ms) of TCP traffic by iperf, repeated for 10 runs ◮ AQM iface: GSO TSO off, BQL=1514, txqueuelen=1000 ◮ TCP: Linux default with reno ◮ Topology: Dumbbell with 4 sender-receiver pairs Model Dell OptiPlex GX620 CPU Intel(R) Pentium(R) 4 CPU 3.00 GHz RAM 1 GB PC2-4200 (533 MHz) Ethernet Broadcom NetXtreme BCM5751 RTL-8139 (AQM interface) RTL8111/8168B (Dummynet router) Ethernet driver tg3 8139too (AQM interface) r8168 (Dummynet router) OS kernel Linux 3.8.2 (FC14) Linux 3.10.4 (AQM router) (FC16)

10 Mbps 100 Mbps 100 Mbps 100 Mbps 100 Mbps 100 Mbps 100 Mbps A Q M 100 Mbps 100 Mbps Bottleneck router Dummynet router

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 10 / 21

SLIDE 11

Experimental Setup

Experimental Setup (cont.)

◮ AQM parameters used unless otherwise noted.

CoDel

interval=100 ms target=5 ms

PIE

parameters in [pie]. PIE Parameter Default value tupdate 30 ms Ttarget 20 ms α 0.125 β 1.25

ARED

parameters in [FGS01]. ARED Parameter Default value interval 500 ms α min(0.01, pmax/4) β 0.9 th_min 0.5 ∗ target th_max 1.5 ∗ target

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 11 / 21

SLIDE 12

Experimental Setup

Experimental Setup (cont.)

◮ RTT is measured on per-packet basis using Synthetic Packet

Pairs (SPP) tool [spp]

◮ Gives a very precise distribution of perceived RTT on the path

◮ Goodput is measured per 5-sec intervals

◮ long-term throughput/goodput does not reflect AQM performance

ver time (e.g. bursts of packet drops are not desired)
N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 12 / 21

SLIDE 13

A Basic Test

Single TCP Flow (RTTbase=100 ms)

100 110 120 130 140 150 160 170 180 190 200 210 220 230 240 1 1 1 5 5 5 10 10 10 20 20 20 30 30 30 RTT (ms) Target Delay (ms) CoDel PIE ARED

(a) Per-packet RTT

3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 9 9.5 10 1 1 1 5 5 5 10 10 10 20 20 20 30 30 30 Goodput (Mbps) Target Delay (ms) CoDel PIE ARED

(b) Goodput Per-packet RTT and goodput. Bottom and top of whisker-box plots show 10th and 90th percentiles respectively.

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 13 / 21

SLIDE 14

A Basic Test

A Basic Test (cont.)

◮ A similar trend can

be observed between CoDel and RED in a different test in [NJ12]

CoDel vs. RED from K. Nichols, “Controlling Queue Delay” [NJ12] FTP traffic mix w/ and w/o web-browsing and CBR applications and RTTs from 10∼500 ms.

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 14 / 21

SLIDE 15

Parameter Sensitivity

Parameter Sensitivity (cont.)

Target Delay

100 110 120 130 140 150 160 170 180 190 200 210 220 230 240 250 1 1 1 5 5 5 10 10 10 20 20 20 30 30 30 RTT (ms) Target Delay (ms) CoDel PIE ARED

(c) Light

100 110 120 130 140 150 160 170 180 190 200 210 220 230 240 250 1 1 1 5 5 5 10 10 10 20 20 20 30 30 30 RTT (ms) Target Delay (ms) CoDel PIE ARED

(d) Moderate

100 110 120 130 140 150 160 170 180 190 200 210 220 230 240 250 1 1 1 5 5 5 10 10 10 20 20 20 30 30 30 RTT (ms) Target Delay (ms) CoDel PIE ARED

(e) Heavy Per-packet RTT. Light, moderate and heavy congestion scenarios (4 senders and RTTbase=100 ms). Light, moderate and heavy congestion correspond to 4, 16 and 64 concurrent TCP flows respectively.

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 15 / 21

SLIDE 16

Parameter Sensitivity

Parameter Sensitivity (cont.)

Target Delay

5.5 6 6.5 7 7.5 8 8.5 9 9.5 10 1 1 1 5 5 5 10 10 10 20 20 20 30 30 30 Goodput (Mbps) Target Delay (ms) CoDel PIE ARED

(f) Light

8.5 9 9.5 10 1 1 1 5 5 5 10 10 10 20 20 20 30 30 30 Goodput (Mbps) Target Delay (ms) CoDel PIE ARED

(g) Moderate

8.5 9 9.5 10 1 1 1 5 5 5 10 10 10 20 20 20 30 30 30 Goodput (Mbps) Target Delay (ms) CoDel PIE ARED

(h) Heavy

Goodput. Light, Moderate and Heavy congestion scenarios (4 senders and RTTbase=100 ms).
N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 16 / 21

SLIDE 17

Parameter Sensitivity

Parameter Sensitivity (cont.)

CoDel’s Dropping Mode Interval – Target Delay=5 ms

100 110 120 130 140 150 160 170 180 5 5 5 30 30 30 100 100 100 RTT (ms) Interval (ms) Light Moderate Heavy

(i) Per-packet RTT

8 8.5 9 9.5 10 5 5 5 30 30 30 100 100 100 Goodput (Mbps) Interval (ms) Light Moderate Heavy

(j) Goodput 4 senders, RTTbase=100 ms.

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 17 / 21

SLIDE 18

Parameter Sensitivity

Parameter Sensitivity (cont.)

PIE’s t_update Interval – Target Delay=5 ms

100 110 120 130 140 150 160 170 180 190 200 210 220 230 240 5 5 5 30 30 30 100 100 100 RTT (ms) Interval (ms) Light Moderate Heavy

(k) Per-packet RTT

8 8.5 9 9.5 10 5 5 5 30 30 30 100 100 100 Goodput (Mbps) Interval (ms) Light Moderate Heavy

(l) Goodput 4 senders, RTTbase=100 ms.

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 18 / 21

SLIDE 19

Conclusions and Future Work

Conclusions

◮ ARED: Only performed worse than CoDel or PIE with small

number of flows

◮ CoDel: Dropping mode interval can be reduced to lower the delay

without degrading the goodput much

◮ PIE (as implemented in Linux): Long distribution tail for low

target delays

Future Work

◮ More realistic traffic types (here, only bulk TCP traffic) including

bursty traffic

◮ Simulations for environment parameters that cannot be produced

with our testbed

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 19 / 21

SLIDE 20

Q&A

Bibliography

[FGS01] Sally Floyd, Ramakrishna Gummadi, and Scott Shenker. Adaptive RED: An Algorithm for Increasing the Robustness of RED’s Active Queue Management. Technical report, 2001. [GRT+13]

Y. Gong, D. Rossi, C. Testa, S. Valenti, and D. Taht.

Fighting the Bufferbloat: On the Coexistence of AQM and Low Priority Congestion Control. 2013. [NJ12] Kathleen Nichols and Van Jacobson. Controlling Queue Delay. Queue, 10(5):20:20–20:34, May 2012. [pie] PIE Linux code (Cisco). ftp://ftpeng.cisco.com/pie/linux_code/. [spp] Synthetic Packet Pairs. http://caia.swin.edu.au/tools/spp/. [Whi13] Greg White. A Simulation Study of CoDel, SFQ-CoDel and PIE in DOCSIS 3.0 Networks. Technical Report, CableLabs, April 2013. [WP12] Greg White and Joey Padden. Preliminary Study of CoDel AQM in a DOCSIS Network. Technical Report, CableLabs, November 2012.

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 20 / 21

SLIDE 21

Q&A

More on experimental results: N. Khademi, D. Ros, and M. Welzl, “The New AQM Kids on the Block: Much Ado About Nothing?”, Technical Report 434, Department of Informatics, University of Oslo, 23 October 2013, available at http://urn.nb.no/URN:NBN:no-38868

N. Khademi, D. Ros, M. Welzl

The New AQM Kids on the Block... April 28, 2014 21 / 21