Applying POMDP to moving target optimization

Diversity maintains security by making the computing environment less standard and less predictable. Recent studies show that many randomization techniques, e.g. address space layout randomization (ASLR) significantly enhance system security simply through reducing the number of return to libc exploits [14]. However, "diversity" may incur significant overhead on the computing platforms. We study the problem of implementing diversity to trade off security performance with diversity implementation costs. We address this problem by formulating it as a partially observable Markov decision process (POMDP). An optimal solution considering a fixed amount of history can be obtained by transforming the POMDP optimization problem into a nonlinear programming (NLP) problem. Simulation results for a set of benchmark problems illustrate the effectiveness of the proposed method.