The project on the design of realistic uncheatable benchmarks strives to achieve a more sophisticated design methodology of computer performance benchmarks, based on the theory of computational complexity. The goal is to integrate ideas from randomization and cryptography with realistic existing benchmark in everyday use. There has been tremendous progress in theoretical computer science in recent years, especially with regard to the power of randomization and interaction. These ideas are well suited to the design of uncheatable benchmarks. The purpose of benchmark testing is to estimate the performance and reliability of hardware and software systems. By and large existing benchmark design has focused on "typical" programs and data sets for the system. The problem of making benchmarks resistant to tampering has been mostly ignored. The on-going research in this project has demonstrated the feasibility of applying ideas from computational complexity and cryptography to a wide variety of existing benchmarks, and making them more accurate, resistant to tampering and thus more trustworthy.***
