The Price of Meltdown and Spectre: Energy Overhead of Mitigations at Operating System Level

The Meltdown and Spectre hardware vulnerabilities shocked hardware manufacturers and system users upon discovery. Numerous attack vectors and mitigations have been developed and deployed. However, due to the deep entanglement in core CPU components they will be an important topic for years. Although the performance overhead of software mitigations has been examined closely, the energy overhead has experienced little attention---even though the energy demand is a critical cost factor in data centres. This work contributes a fine-grained energy-overhead analysis of Meltdown/Spectre software mitigations, which reveals application-specific energy overheads of up to 72 %. We further compare energy overheads to execution time overheads. Additionally, we examine subsystem-specific effects (i.e., CPU, memory, I/O, network/interprocess communication) and develop a model that predicts energy overheads for applications.

[1]  Luiz André Barroso,et al.  The Case for Energy-Proportional Computing , 2007, Computer.

[2]  Michael Stumm,et al.  An analysis of performance evolution of Linux's core operations , 2019, SOSP.

[3]  Thomas F. Wenisch,et al.  Breaking Virtual Memory Protection and the SGX Ecosystem with Foreshadow , 2019, IEEE Micro.

[4]  Salvatore J. Stolfo,et al.  CLKSCREW: Exposing the Perils of Security-Oblivious Energy Management , 2017, USENIX Security Symposium.

[5]  Daniel Gruss,et al.  PLATYPUS: Software-based Power Side-Channel Attacks on x86 , 2021, 2021 IEEE Symposium on Security and Privacy (SP).

[6]  Frank Piessens,et al.  Fallout: Leaking Data on Meltdown-resistant CPUs , 2019, CCS.

[7]  Michael Hamburg,et al.  Spectre Attacks: Exploiting Speculative Execution , 2018, 2019 IEEE Symposium on Security and Privacy (SP).

[8]  Wolf-Dietrich Weber,et al.  Power provisioning for a warehouse-sized computer , 2007, ISCA '07.

[9]  Manos Antonakakis,et al.  SoK: The Challenges, Pitfalls, and Perils of Using Hardware Performance Counters for Security , 2019, 2019 IEEE Symposium on Security and Privacy (SP).

[10]  Jeremy Kepner,et al.  Measuring the Impact of Spectre and Meltdown , 2018, 2018 IEEE High Performance extreme Computing Conference (HPEC).

[11]  Michael Hamburg,et al.  Meltdown: Reading Kernel Memory from User Space , 2018, USENIX Security Symposium.

[12]  Derek Chiou,et al.  GPGPU performance and power estimation using machine learning , 2015, 2015 IEEE 21st International Symposium on High Performance Computer Architecture (HPCA).

[13]  Michail Loukeris.  Efficient computing in a safe environment , 2019, ESEC/SIGSOFT FSE.

[14]  Ricardo Bianchini,et al.  Application transformations for energy and performance-aware device management , 2002, Proceedings.International Conference on Parallel Architectures and Compilation Techniques.

[15]  Marco D. Santambrogio,et al.  Adaptive and Flexible Smartphone Power Modeling , 2013, Mob. Networks Appl..

[16]  El-Sayed M. El-Alfy,et al.  Impact of Spectre/Meltdown Kernel Patches on Crypto-Algorithms on Windows Platforms , 2019, 2019 International Conference on Innovation and Intelligence for Informatics, Computing, and Technologies (3ICT).

[17]  Rüdiger Kapitza,et al.  Proactive Energy-Aware Programming with PEEK , 2014, TRIOS.

[18]  David Lorge Parnas,et al.  Review of David L. Parnas' "Designing Software for Ease of Extension and Contraction" , 2004 .