No Free Lunch for Avoiding Clustering Vulnerabilities in Distributed Systems

Emergent design failures are ubiquitous in multi-component, distributed systems. For example, leakage in microprocessors, congestion in airline networks, and outfit density in warships are design failures that emerge from the undesirable clustering of design elements. Here, we use methods of statistical physics to show that clustering vulnerabilities arise from generic interactions among competing degrees of freedom that describe the placement and connection of distributed system elements. We show that clustering vulnerabilities have multiple origins, including trade-offs between configurational and conformational entropy, and that avoiding clustering is only possible at the expense of introducing large variability in arrangement. Our results suggest that there is "no free lunch" for avoiding clustering vulnerabilities, but they can be quantified and managed via statistical physics approaches.