TL;DR
HYDRA is a new optimization framework that calculates the absolute minimum botnet size and traffic required to launch devastating Link-Flooding Attacks (LFAs) against Low Earth Orbit (LEO) satellite networks. By framing the attack as a resource-minimization problem, it reveals that LEO networks are significantly more vulnerable to targeted regional disruptions than previously thought, while providing a rigorous benchmark to test defensive mitigations.
What Problem Does This Paper Address?
Low Earth Orbit (LEO) satellite constellations like SpaceX's Starlink and Amazon's Project Kuiper are rapidly becoming critical infrastructure for global, low-latency internet access. However, unlike terrestrial networks backed by billions of devices, LEO networks rely on a topologically sparse, capacity-constrained mesh of Inter-Satellite Links (ISLs) and Ground-Satellite Links (GSLs).
Because satellite orbits are highly predictable, attackers can easily anticipate network bottlenecks. Prior work (such as the ICARUS framework) demonstrated that Link-Flooding Attacks (LFAs) can exploit these bottlenecks to congest links and disrupt service. However, previous models assumed massive, unoptimized botnets. This paper addresses a critical gap: quantifying the true resilience of LEO networks by finding the absolute minimum threshold of compromised satellite terminals an attacker needs to strategically disrupt communication between specific geographic zones.
Methods
The authors introduce HYDRA, a discrete modeling and optimization framework that formulates LFA variants as botnet minimization problems. Instead of relying on brute-force flooding, HYDRA mathematically optimizes bot selection and traffic allocation using Mixed-Integer Linear Programming (MILP), solved via the Gurobi Optimizer.
HYDRA evaluates four distinct attack scenarios:
- Snapshot Attack: Minimizing the active botnet size for a single point in time.
- Continuous Attack: Jointly optimizing across multiple time snapshots to sustain disruption as the satellite topology dynamically shifts.
- Flexible Botnet Configuration: Finding a single, static botnet layout capable of targeting any zone pair individually without needing reconfiguration.
- Maximized Simultaneous Attack: Co-optimizing a botnet to flood shared links and disrupt multiple zone pairs concurrently.
Additionally, the authors use HYDRA to evaluate the efficacy of five network-layer mitigation strategies: route diversification, link-triggered source throttling, ingress capacity policing, distance-based traffic constraints, and reactive botnet attrition.
Datasets
The authors evaluated their framework using simulated LEO constellation models based on public orbital parameters (Two-Line Element data) across 10 different scenarios (including Starlink, OneWeb, and Project Kuiper).
To simulate realistic internet traffic, they utilized public global GDP data to perform GDP-weighted sampling of global city pairs. This generated 250,000 source-destination traffic demands per snapshot over a 90-hour simulation window to establish realistic background network loads.
Key Insights and Ideas
- LEO Predictability is an Attacker's Asset: Because satellite paths are entirely deterministic, an attacker does not need a massive botnet. They can carefully coordinate a small, highly strategic subset of bots to achieve maximum disruptive impact.
- The Square-Root Scaling Law: Surprisingly, as an attacker targets more zone pairs simultaneously or flexibly, the required botnet size does not grow linearly—it follows a square-root curve ($y \propto \sqrt{x}$). This occurs because different regional attack paths frequently overlap and share the same underlying satellite bottlenecks, allowing a single bot's traffic to pull double duty.
- Path Diversity Alone Won't Save the Network: Multipath routing helps, but because the physical topology of a satellite constellation is inherently sparse, alternative routes often converge on the same regional bottlenecks, allowing optimized attacks to adapt easily.
Key Results
- Higher Efficiency than Prior Attacks: Compared to the state-of-the-art ICARUS baseline under identical stealth constraints, HYDRA reduces the required active botnet size by 33.9% on average while slashing aggregate attack traffic by 23%.
- High Temporal Persistence: In continuous attack scenarios over a 90-minute window, HYDRA achieved a >97% success rate in maintaining targeted congestion as the network topology dynamically evolved.
- Defense Efficacy: Among the evaluated defenses, Bottleneck-Minimize Routing (KBM) and hop-throughput constraints (which limit traffic on routes traversing too many satellites) proved to be the most effective mitigations, significantly raising the resource bar for attackers.
Limitations and Future Work
- Routing Abstraction: The base model abstracts routing as a single shortest path per node pair per snapshot, omitting real-time path dispersion or complex dynamic traffic engineering that operators might deploy.
- Perfect Knowledge Assumption: HYDRA assumes the attacker has highly accurate knowledge of the constellation's topology and background traffic matrices.
- QoS Trade-offs: The evaluated mitigations (like source throttling or distance-based constraints) successfully hinder attacks but introduce severe Quality of Service (QoS) penalties for legitimate users, such as increased latency or dropped bursts. Future work must explore defenses that maximize security while minimizing legitimate user degradation.
Summary updated