SPAM

Streamlined Prefetcher-Aware Multi-Threaded Covert-Channel

Guide: Prof. Biswabandan Panda

In this project, we proposed a novel last‑level cache (LLC) covert‑channel attack that leverages hardware prefetchers by carefully tuning memory access patterns to selectively trigger or bypass them. Building on a literature review of Intel’s IP‑stride and stream prefetchers, I extended the state‑of‑the‑art Streamline attack to multi‑core systems with distinct sender and receiver access patterns: senders exploit prefetchers to accelerate encoding, while receivers circumvent them for precise latency measurements.

Key contributions include:

  • Accelerating covert communication by selectively triggering or bypassing hardware prefetchers, after a literature review of the Intel IP-stride and stream prefetchers.
  • Advanced the bit rate of the state-of-the-art covert-channel Streamline by over 70% in multi-core systems for comparable error rates.
  • Enhanced resilience against defenses by investigating mitigations of flush-based channels.
  • Tested attacks rigorously by injecting noise using stress-ng workloads co-running on other threads.
  • Co-authored the paper of the work published at IEEE Computer Architecture Letters 2025.

Link: Paper

References

2025

  1. SPAM: Streamlined Prefetcher-Aware Multi-Threaded Cache Covert-Channel Attack
    E. Kritheesh, and Biswabandan Panda
    IEEE Computer Architecture Letters, 2025