Honeyword is a representative ``honey" technique to detect intruders by luring them with decoy data. This kind of honey technique blends a primary object (from distribution $P$) with decoy samples (from distribution $Q$). In this research, we focus on two key Honeyword security metrics: the flatness function and the success-number function. Previous researchers are engaged in designing experimental methods to estimate their values. We've derived theoretical formulas on both metrics of the strongest $\mathcal{A}$ using the optimal guessing strategy, marking a first in the field. The mathematical structures of these metrics are intriguing: the flatness function has an expression as $\epsilon(i)=\sum_{j=1}^{i}\int_{0}^{+\infty}\tbinom{k-1}{j-1} f(x)G^{k-j}(x)(1-G(x))^{j-1}dx$. In particular, the most important one, $\epsilon(1)$ is $\frac{1}{k}(M-\int_{0}^{M}G^k(x)dx)+b$, where $M=\max_{x: Q(x)\neq 0}\frac{P(x)}{Q(x)}$, $b=\sum_{x: Q(x)=0}P(x)$, and $G$ is a cumulative distribution function derived from $P$ and $Q$. This formula provides a criterion to compare different honey distributions: the one with smaller $M$ and $b$ is more satisfactory. The mathematical structure of the success-number function is a series of convolutions with beta distribution kernels: $\lambda_U(i)=U\sum_{j=1}^{i}\int_{\frac{1}{k}}^{1} \frac{\phi(x)}{1-\phi(x)} \tbinom{U-1}{j-1} x^{U-j}(1-x)^{j-1}dx$, where $U$ is the number of users in the system and $\phi(x)$ is a monotonically increasing function. For further elaboration, we made some representative calculations. Our findings offer insights into security assessments for Honeyword and similar honey techniques, contributing to enhanced security measures in these systems.
翻译:暂无翻译