Adaptive Performance-Aware MPC: Detecting and Autonomously Responding to Runtime Anomalies

While tools like Sequre (Smajlović et al.) and PIGEON (Harth-Kitzerow et al.) focus on compile-time or hardware-level optimizations for better MPC performance, they don’t address unpredictable runtime anomalies—such as sudden network bottlenecks, hardware failures, or workload spikes—that can arise in real-world deployments, especially in cloud or distributed settings. This idea proposes embedding anomaly detection agents (drawing from distributed systems monitoring) directly into MPC frameworks. When a deviation from expected performance is detected (e.g., via statistical profiling or machine learning models), the system could, for example, switch to more communication-efficient protocols, adjust batch sizes, or re-route computation to healthier nodes. Such adaptive, performance-aware MPC would be particularly valuable for long-running or resource-intensive applications (e.g., private ML training, as discussed in ReplayMPC by Bautista et al.) and would fill a gap where current frameworks require manual intervention or restart on failures. This could make MPC more robust and practical for mission-critical deployments and collaborative multi-organization scenarios.

References:

1. Sequre: a high-performance framework for secure multiparty computation enables biomedical data sharing. Haris Smajlović, Ariya Shajii, Bonnie Berger, Hyunghoon Cho, Ibrahim Numanagić (2023). Genome Biology.
2. ReplayMPC: A Fast Failure Recovery Protocol for Secure Multiparty Computation Applications using Blockchain. Oscar G. Bautista, Kemal Akkaya, Soamar Homsi (2023). International Conference on Smart Computing.
3. PIGEON: A High Throughput Framework for Private Inference of Neural Networks using Secure Multiparty Computation. Christopher Harth-Kitzerow, Yongqin Wang, Rachit Rajat, Georg Carle, Murali Annavaram (2025). Proceedings on Privacy Enhancing Technologies.