SwarmFlawFinder: Discovering and Exploiting Logic Flaws of Swarm Algorithms

This is the project page for the submission "SwarmFlawFinder: Discovering and Exploiting Logic Flaws of Swarm Algorithms."

This repository includes all the developed tools, data, and results. In addition, supplementary materials to the paper, not included due to the paper's space limit, can be found below.

Code

Developed Tools and Code

This sub page has code used in this paper including tools, data, and fixes (with original algorithms).
Link to open the subpage

Supplymentary Materials to the Paper

1. Illustration of Swarm Attack Strategies

In Section IV. Design; A. Test-run Definition and Creation, we mention "Attack Strategy (S)" and refer to this webpage to illustrate the attack strategies.
Link to open the subpage

2. Multiple Attack Drones Scenario Example

In Section IV. Design; D. Testing with Multiple Attack Drones, we refer to this webpage for an example scenario with multiple attack drones to elaborate how our system handle multiple drones.
Link to open the subpage

3. Algorithm Selection (for Evaluation)

In Section V. Evaluation; A. Experiment Setup (and Appendix A. Algorithm Selection), we refer to this webpage for the details of selection criteria of the algorithms we evaluated.
- The complete list of 73 algorithms we searched
  - The list of algorithms we have tried but failed to run
    - This page will provide the detail of why we were unable to run the algorithms and so include our studies.

4. Evaluation of Fixes

In Section V. Evaluation; B. Effectiveness in Finding Logic Flaws, we refer to this webpage for the quality and performance overhead of our suggested fixes.
- This sub page includes
  1. The quality of fixes for A2, A3, and A4 (A1 is in the paper)
  2. Performance overhead of the fixes and integrated fixes (for all algorithms)
Link to open the subpage

5. Spatial Distribution of Test-cases

In Section V. Evaluation; C. Effectiveness of DCC in Fuzz Testing, we refer to this webpage for the spatial distribution of test-cases generated by algorithms A2, A3, and A4.
Link to open the subpage

6. SwarmFlawFinder and Random Testing Approach with Search Space Restrictions

In Section V. Evaluation; C. Effectiveness of DCC in Fuzz Testing; 3. Impact of Searching Space on Random Testing, we refer to this webpage for the additional experiments of SwarmFlawFinder and the random testing approach with 3 different search space restrictions.
- This subpage provides additional results for SwarmFlawFinder under different search space restrictions.
Link to open the subpage

7. Details of the Number of Additional Attack Drones and Overhead

In Section IV. Design; D. Testing with Multiple Attack Drones, we refer to this webpage for the details of the number of additional attack drones and overhead.
Link to open the subpage

A. Additional Materials for Clarifications

We present additional materials that might be useful in understanding our work but are not referred to in the paper.
- A.1. Statistics of Random Testing Approach and SwarmFlawFinder
  - To support the random testing approach is less effective than SwarmFlawFinder, this page shows the number of identified unique attacks from the random testing taking an example of A₁.
  - Link to open the subpage
- A.2. Analysis on Mission Failures Exclusively found by Random Testing Approach
  - This page explains that the instances on the top of the gray circle are variants of what SwarmFlawFinder already identified.
  - Link to open the subpage
- A.3. Activated Attack Strategies during Evaluation
  - We report what attack strategies were frequently activated (might be effective) during testing for each algorithm.
  - Link to open the subpage

ElaineYao / swarmflaw