Exploring the Energy Overhead of Reversible Programs Executed on Irreversible Hardware
Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Standard
Exploring the Energy Overhead of Reversible Programs Executed on Irreversible Hardware. / Vadgaard, Lars Bo Husted; Kirkeby, Maja Hanne; Larsen, Ken Friis; Thomsen, Michael Kirkedal.
Reversible Computation - 16th International Conference, RC 2024, Proceedings. ed. / Torben Aegidius Mogensen; Lukasz Mikulski. Springer VS, 2024. p. 77-93 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 14680 LNCS).Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - GEN
T1 - Exploring the Energy Overhead of Reversible Programs Executed on Irreversible Hardware
AU - Vadgaard, Lars Bo Husted
AU - Kirkeby, Maja Hanne
AU - Larsen, Ken Friis
AU - Thomsen, Michael Kirkedal
N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.
PY - 2024
Y1 - 2024
N2 - 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.
AB - 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.
KW - Dereversibilisation
KW - Encryption
KW - Energy efficiency
KW - Reversible programs
U2 - 10.1007/978-3-031-62076-8_6
DO - 10.1007/978-3-031-62076-8_6
M3 - Article in proceedings
AN - SCOPUS:85195862754
SN - 9783031620751
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 77
EP - 93
BT - Reversible Computation - 16th International Conference, RC 2024, Proceedings
A2 - Mogensen, Torben Aegidius
A2 - Mikulski, Lukasz
PB - Springer VS
T2 - 16th International Conference on Reversible Computation, RC 2024
Y2 - 4 July 2024 through 5 July 2024
ER -
ID: 397034163