ENSC 833 – AQM Shootout in ns-3

ns-3 Bufferbloat AQM FQ-CoDel PIE RED ECN (stretch)

Abstract

Bufferbloat, excessive latency caused by oversized network buffers under load, remains a persistent problem in home and edge networks. This project proposes a systematic evaluation of Active Queue Management (AQM) algorithms using the ns-3 network simulator to identify configurations that best mitigate delay while preserving throughput and fairness. A dumbbell topology will be used to emulate a residential bottleneck link with realistic propagation delays. Traffic will consist of 1 to N long-lived TCP flows competing with a latency-sensitive UDP flow representative of gaming or voice applications. The study will compare FQ-CoDel, PIE, and RED across parameter sweeps of bottleneck capacity, base RTT, queue discipline, and TCP congestion control variants such as CUBIC, BBR, and DCTCP. Key metrics include 95th percentile queueing delay, end-to-end latency, aggregate throughput, fairness using Jain’s index, and packet drop or mark rates. A stretch goal will enable Explicit Congestion Notification (ECN) to contrast marking-based and drop-based behaviors. The outcome will be a reproducible ns-3 experiment framework, comparative performance plots, and a recommendation matrix mapping optimal AQM choices to traffic mixes and network conditions typical of home environments.

1) Motivation & Problem Statement

Home networks often exhibit bufferbloat: large device buffers absorb bursts but can introduce sustained queueing delay under congestion. This harms interactive applications (gaming/VoIP) even when throughput remains high. This project quantifies the trade-offs between: (i) latency/queueing delay, (ii) throughput, and (iii) fairness for different AQM designs under realistic “many downloads + one interactive flow” workloads.

2) Topology & Network Model

Topology: Dumbbell (multiple senders → bottleneck → multiple receivers)
Bottleneck link: Capacity sweep (e.g., 10–200 Mbps) to represent typical home uplink/downlink
Base RTT: RTT sweep (e.g., 10–100 ms) to capture local vs wider-area paths
Queueing location: Bottleneck device queue discipline (AQM under test)
AQM algorithms: FQ-CoDel, PIE, RED

(You can adjust sweep ranges to match your target “home network” scenario.)

3) Traffic Workload

Bulk traffic: 1 to N long-lived TCP flows (file downloads / streaming)
Interactive traffic: 1 UDP flow (gaming/voice-like)
TCP variants: CUBIC, BBR, DCTCP (where applicable)
Run structure: Warm-up → steady-state measurement window

Workload is designed to stress the queue and reveal delay spikes and tail latency.

4) Experimental Variables (Parameter Sweeps)

AQM / queue discipline: FQ-CoDel vs PIE vs RED
Bottleneck capacity: [fill in your sweep values]
Base RTT: [fill in your sweep values]
Queue limit / target delay knobs: [document defaults + any tuned ranges]
TCP congestion control: CUBIC / BBR / DCTCP
ECN (stretch): off vs on (marking vs dropping)

5) Metrics

Queueing delay (tail): 95th percentile queueing delay at the bottleneck
End-to-end latency: UDP packet delay distribution (mean / p95 / max)
Throughput: Aggregate TCP goodput (and per-flow goodput)
Fairness: Jain’s fairness index across TCP flows
Loss/mark behavior: drop rate and (if ECN) marking rate

6) Results

This section will include plots comparing AQMs under each scenario. Suggested figures:

Tail latency vs throughput (e.g., p95 latency on x-axis, aggregate throughput on y-axis)
Queue delay CDF for each AQM under the same conditions
Per-flow throughput + Jain’s fairness across N TCP flows
Drop/mark rate vs offered load or RTT

Place images in a figures/ folder and embed like:

<img src="figures/latency_throughput.png" alt="Latency vs Throughput" style="max-width:100%; border:1px solid #ddd; border-radius:12px;">

7) Reproducibility

Code repo: [Add link]
ns-3 version: [e.g., ns-3.XX]
How to run:

# example commands (replace with your script + args)
./waf configure
./waf --run "aqm-shootout --aqm=FqCoDel --tcp=Cubic --rtt=40ms --bw=50Mbps --flows=8 --ecn=0"
./waf --run "aqm-shootout --aqm=Pie --tcp=Bbr --rtt=40ms --bw=50Mbps --flows=8 --ecn=1"

Outputs: raw traces (.pcap/.csv) + generated plots in results/ and figures/
Experiment matrix: a single config file enumerating sweeps (recommended)

8) Expected Outcome / Recommendation Matrix

Final deliverable will include a small “decision guide” mapping optimal AQM choices to conditions typical of home networks (capacity/RTT/flow count and traffic mix). Example table columns: Scenario, Best AQM, Key knob settings, Observed trade-off.

References

Nichols, K., and Jacobson, V. (2012). Controlling Queue Delay (CoDel). ACM Queue, 10(5).
Hoeiland-Joergensen, T., McKenney, P., Taht, D., Gettys, J., and Dumazet, E. (2015). The FlowQueue CoDel Packet Scheduler and Active Queue Management Algorithm. RFC 8290, IETF.
Pan, R., Natarajan, P., Baker, F., and VerSteeg, B. (2013). PIE: A Lightweight Control Scheme to Address the Bufferbloat Problem. RFC 8033, IETF.
Floyd, S., and Jacobson, V. (1993). Random Early Detection Gateways for Congestion Avoidance. IEEE/ACM Transactions on Networking, 1(4), 397–413.
ns-3 Consortium. (2023). ns-3 Network Simulator Documentation. https://www.nsnam.org/docs/