Introduction
In modern networking, the decoupling of identity and location is a key strategy for improving scalability and mobility. Protocols like the Locator/ID Separation Protocol (LISP) implement this "ID/LOC split" by using a mapping system. When a border router (xTR) receives a packet for an unknown destination, it must "pull" the mapping from a global system via a control-plane channel.
However, this pull-based mechanism introduces significant security vulnerabilities. Attackers can exploit the control plane through three primary vectors:
- Control-Plane Denial-of-Service (DoS): Flooding the router with packets for non-existent destinations to exhaust rate-limiter resources.
- Channel Overflow: Overwhelming the memory buffer used to store validation "nonces" for pending Map-Requests.
- Cache Pollution: Forcing the eviction of legitimate entries from the local mapping cache by requesting a wide range of random prefixes.
Proposed Method: CMS-Based Rate Limiting
To mitigate these threats, this paper proposes a per-source rate-limiter implemented at the border router. Unlike traditional global limits, this approach tracks the request frequency of each individual internal source node. To achieve this at line-rate without the prohibitive memory costs of linear lookup tables, the authors utilize a Count-Min Sketch (CMS).
The CMS is a probabilistic data structure consisting of a 2D array of counters. When a cache miss occurs, the source ID is hashed $d$ times, incrementing the corresponding cells in the matrix. To estimate a source's frequency, the router retrieves the values from these $d$ cells and selects the minimum. If this estimate exceeds a predefined threshold $T$, the request is dropped. This "minimum" logic ensures that while the system may occasionally overestimate traffic due to hash collisions (false positives), it will never underestimate it, ensuring all heavy-hitter attackers are caught.
Evaluation and Results
The proposed system was evaluated using a custom simulation environment modeling LISP network behavior across various scales (50,000 to 500,000 nodes) and attacker densities (1% to 10%).
Key findings include:
- Memory Efficiency: The CMS significantly outperforms traditional tables. For a 50,000-node network, a standard table requires ~100KB, whereas the CMS achieved zero false positives with only 56KB.
- Graceful Degradation: The system allows for a trade-off between memory and accuracy. By tolerating a negligible false positive rate (<0.1%), the memory footprint for the same 50k-node network can be reduced to just 30KB.
- Scalability: While increasing the number of attackers requires more memory to maintain precision, the CMS scales sub-linearly, making it far more viable for high-density environments than per-flow state tracking.
Conclusion and Future Work
The use of Count-Min Sketches provides a robust, memory-efficient defense against control-plane exhaustion in ID/LOC protocols. By focusing on per-source tracking, the system effectively isolates and suppresses attackers. Future work may address the vulnerability to internal source IP spoofing and explore dynamic parameter tuning to adapt to shifting traffic patterns in real-time. Additionally, the authors suggest that combining this rate-limiting approach with advanced cache replacement policies (like LFU-Aging) would provide a comprehensive defense against cache-specific attacks.
Summary updated