High-Level Plan

[taham0]

Deliverables

1. End-to-end prototype of facial recognition system deployed on android devices
2. Fault injection framework
3. Report including evaluation

---

Application Use-case

1. Intruder detection: multiple devices work together to identify an intruder
2. Test Taker ID Verification System: use age, gender and other features to ID test taker

System details

1. Use EdgeFace. Implement features around this model.

OR

2. Use open-source facial recog app. Extend this application with a communication protocol like gRPC. Implement redundancy, and consistency between clients. Connect with server and db.

Redundancy

1. N-version programming (makes the most sense to me)
2. Recovery Blocks (but latency)
3. N Self-checking (acceptance test will add to latency)

Fault Categorization

Broadly, what kind of faults do we want to induce?

(1) Network: node crashes, slow processing, packet loss and jitter, network partition (2) Software: no result, incorrect result (false positive, false negative)

[adityavardhanpadala]

• Presentation outline

• Communication Protocol

  • gRPC
  • raw sockets

• Android App

  • Host <-> Client comms
  • Client <-> Chaos Engine comms

• Chaos Engine outline

  • Simple TCP Server
  • JSON messages to every client
    • check [[system_overview.excalidraw]]