Exploring the Energy Overhead of Reversible Programs Executed on Irreversible Hardware

This paper investigates the energy efficiency of reversible programs executed on irreversible hardware. Motivated by Landauer’s principle, which suggests that information loss during program execution increases energy consumption, we employ Intel’s Running Average Power Limit (RAPL) technology to compare the energy consumption of “dereversibilised” reversible programs against that of straightforward irreversible implementations. Preliminary results indicate that, while the memory handling of dereversibilised reversible programs does not necessarily improve overall energy efficiency, the energy overhead remains constant, ranging from 6% to around 240% over C depending on the abstraction level of the problem at hand. For dereversiblised Janus, one can expect an energy overhead of 100% or less when operating on data structures that are easily represented in the target language. One may expect an energy overhead of less than 30% for dereversibilised Hermes; however, its domain is more specific and thus resembles the target language more closely. These findings indicate an interesting trade-off: dereversibilisation incurs a certain (constant) energy overhead, yet it uniquely produces two programs for the price of one. This will enable programmers to make informed decisions, balancing the energy overhead against the benefits of reversibility based on their specific needs. Future work will aim to further explore the impact of dereversibilisation on energy overhead and information security.