SDN Performance Analysis - Vincenzo Imperati

This project uses Mininet to simulate SDN topologies and analyze network performance using Ryu controller and D-ITG traffic generator.

Overview#

Software Defined Networking (SDN) performance analysis framework that provides comprehensive testing and evaluation of network topologies. The project combines Mininet’s network simulation capabilities with Ryu’s SDN controller framework and D-ITG’s traffic generation tools.

Technologies Used#

Mininet: Network emulator for creating virtual networks
Ryu: SDN framework for topology management and control
D-ITG: Distributed Internet Traffic Generator for network flows
Python: Custom topology and controller implementation

Quick Start#

Prerequisites#

sudo apt-get install mininet python3-ryu python3-pip

Install D-ITG following the official documentation.

Usage#

Note: The controller application and Mininet topology must be running on separate terminals.

Controller#

Start the controller by executing the run_controller.sh script:

bash run_controller.sh [-f <controller-file>]

Default controller: controller1.py (located in controllers directory)
Implements shortest-path routing with equal link weights

Mininet Topology#

Run the topology test using the test_topo.sh script:

bash test_topo.sh [-f <topology-file>] [--start <lambda-rate>] [--stop <lambda-rate>] [--step <lambda-rate>]

Default parameters:

Topology: topology2.py (located in topologies directory)
Start: 2 packets/second
Stop: 200 packets/second
Step: 2 packets/second

Generate Performance Plots#

Create graphs and analysis reports:

bash plot_graphs.sh [-f <topology-file>] [--start <lambda-rate>] [--stop <lambda-rate>] [--step <lambda-rate>]

Default parameters:

Same as topology testing
Generates plots for the specified topology and lambda rate range

Network Topologies#

Topology v1#

Configuration:

4 hosts (H₁, H₂, H₃, H₄)
3 switches (S₁, S₂, S₃)
Link properties: 20 Mbps bandwidth, 20ms delay, max queue size = 10

Traffic flows (unidirectional):

H₁ → H₃
H₁ → H₄
H₂ → H₃
H₂ → H₄

Test parameters:

Lambda rate: 2-200 packets/second (Poisson distribution)
Observation time: 20 seconds per measurement
Protocol: UDP packets

Performance Metrics#

The framework measures and analyzes:

End-to-end delay: Packet transmission times across network paths
Packet loss: Network reliability under varying load conditions
Bandwidth utilization: Link capacity usage monitoring
Throughput: Per-flow and aggregate network performance

Controller Features#

The SDN controller (controller1.py) provides:

Shortest-path routing algorithm
Dynamic flow table management
OpenFlow 1.3 compatibility
Automatic topology discovery
Multi-path support capabilities

Data Collection#

Performance data is extracted using:

Ryu: Bandwidth utilization statistics
D-ITG: End-to-end delay and packet loss measurements
Custom scripts: Automated data processing and visualization

Output#

The analysis generates:

Performance graphs and charts
Statistical analysis reports
Network utilization data
Comparative topology studies

Research Findings and Analysis#

This project conducted comprehensive performance analysis comparing two network topologies under varying traffic conditions. The research provides valuable insights into SDN behavior, bottleneck identification, and network optimization strategies.

Topology v1 Analysis#

Network Structure:

Linear topology: H₁-S₁-S₂-S₃ with H₂ connected to S₂, H₃ and H₄ connected to S₃
Traffic flows: (H₁→H₃), (H₁→H₄), (H₂→H₃), (H₂→H₄)
All flows follow shortest path routing with equal link weights

Key Findings:

Bottleneck Identification:

Link (S₂, S₃) identified as primary bottleneck, carrying 4 distinct flows vs. 2 flows on other links
Bottleneck saturation occurs at 60 packets/second lambda rate
Link utilization grows twice as fast for (S₂, S₃) compared to other links

Path-Length Impact on Latency:

H₁ packets experience 33% longer delay than H₂ packets due to extended path length
H₁ path: 4 hops (H₁→S₁→S₂→S₃→destination)
H₂ path: 3 hops (H₂→S₂→S₃→destination)

Packet Loss Behavior:

Linear packet drop growth begins at lambda rates > 100 packets/second
Links (H₃, S₃) and (H₄, S₃) reach maximum utilization slower due to upstream packet drops

Topology v2 Analysis#

Network Enhancement:

Added direct link (S₁, S₃) to eliminate bottleneck from Topology v1
Maintains same link characteristics (20 Mbps, 20ms delay, queue size 10)

Routing Optimization:

New shortest paths: H₁ traffic now uses direct S₁→S₃ link
Eliminated bottleneck: maximum 2 flows per link across entire topology
Link (S₁, S₂) becomes unused in optimal routing

Performance Improvements:

Enhanced Throughput:

Bottleneck saturation improved to 120 packets/second (2x improvement)
All active links show uniform, linear utilization growth
No preferential bottleneck formation

Latency Equalization:

Equal latency for all packet flows (H₁ and H₂ packets)
Path length normalization: all flows now traverse 3 hops

Packet Loss Reduction:

Negligible packet drop in standard testing range (2-200 pkt/s)
Extended testing shows packet drop threshold at 240 packets/second
2.4x improvement in sustainable traffic load

Comparative Performance Summary#

Metric	Topology v1	Topology v2	Improvement
Bottleneck Saturation	60 pkt/s	120 pkt/s	2x
Packet Drop Threshold	100 pkt/s	240 pkt/s	2.4x
H₁ vs H₂ Latency	33% difference	Equal	Eliminated
Max Flows per Link	4	2	50% reduction

Technical Insights#

Network Design Principles:

Single additional link can eliminate bottlenecks and double network capacity
Shortest-path routing automatically exploits topology improvements
Equal path lengths critical for latency consistency

Traffic Engineering Observations:

Flow distribution analysis enables bottleneck prediction
Link cost calculation methodology validated through empirical testing
Poisson traffic distribution provides realistic performance evaluation

SDN Controller Validation:

Shortest-path algorithm correctly adapts to topology changes
Dynamic flow table management handles route optimization automatically
OpenFlow integration enables real-time performance monitoring