Power Analysis of WEP Encryption Jack Kang Benjamin Lee CS252 Final Project Fall 2003
Outline Background and Motivation Objective Theory Experimental Methodology Experimental Results Conclusions Future Work & Directions Questions
Background and Motivation (1/4) The Digital Divide Gap between the digitally empowered and digitally poor, between developing and developed nations Can information and communication technologies (ICT) close the gap? There are social AND economic reasons to solve this problem
Background and Motivation (2/4) Problems More talk than action Financial sustainability Coordination of activities Scope E-governance
Bottom of The Pyramid (BOP) Prahad argues that it is profitable to serve the poor Multinational Corporations have financial incentive to step in Background and Motivation (3/4) Prahalad, C.K. and Hammon, Allen, Serving the World's Poor, Profitably, Harvard Business Review, 9/2002.
Background and Motivation (4/4) So what about the technical problems? Low-cost Low-power Intermittent Connectivity User Interfaces for populations with multiple languages and low levels of literacy Shared accesses as a possibly dominant use mode Limited skilled workforce for maintenance
Objective Evaluate high-level software optimizations and low-level hardware configurations for reducing power dissipation applied to WEP encryption Provide a framework for further study in wireless communication infrastructure for developing regions
Theory – Loop Unrolling A compiler technique that extends the size of loop bodies by replicating the body n times The loop exit condition is then adjusted accordingly Why is power saved? More efficient front end – less branches means the fetch unit is able to fetch large blocks without being interrupted by control decisions Less branches in the code means reduced power dissipation of the branch prediction hardware
Theory – Cache Optimizations Choices in associativity and block sizes will affect the miss rate of the cache. Power can be saved if we can reduce the miss rate. No need to go off chip Better performance means we may be able to lower the clock frequency (and thus voltage levels) and still meet minimum performance needs
Experimental Methodology Software WEP encryption Software is cheaper (low-cost) Easier to upgrade (limited maintenance) SimpleScalar Simulates hardware and software configurations Wattch Provides power estimation
Wired Equivalent Privacy (1/3) Overview wireless standard Provides wireless network with security equivalent to wired network Confidentiality Access Control Data Integrity
Wired Equivalent Privacy (2/3) Encryption Hirani, Sohail A. Energy Consumption of Encryption Schemes in Wireless Devices. Master’s Thesis. University of Pittsburgh, April 2003.
Wired Equivalent Privacy (3/3) Decryption Hirani, Sohail A. Energy Consumption of Encryption Schemes in Wireless Devices. Master’s Thesis. University of Pittsburgh, April 2003.
SimpleScalar (1/2) Baseline Simulation - Microprocessor In-order issue No branch prediction Minimal number of functional units Integer ALU Floating Point ALU Integer Multiplier/Divider Floating Point Multiplier/Divider
SimpleScalar (2/2) Baseline Simulation – Memory L1 Instruction Cache 16-KB cache 32-byte blocks Full associativity L1 Data Cache 16-KB cache 32-byte blocks 4-way associativity Unified L2 Cache 18-KB cache 32-byte blocks 4-way associativity
Wattch (1/2) Overview Framework for analyzing and optimizing microprocessor power dissipation at the architectural level Wattch v1.02 SimpleScalar Interface Simulated PISA instruction set Built on Pentium 4/x86 platform
Wattch (2/2) Conditional Clocking Styles NCC – No conditional clocking CC1 – Simple conditional clocking Zero power dissipation with zero accesses CC2 – Aggressive conditional clocking (ideal) Linear power dissipation with fractional accesses CC3 – Aggressive conditional clocking (non-ideal) 15% power dissipation with zero accesses
Experimental Results (1/3)
Cache Associativity (2/3)
Cache Associativity (3/3)
Conclusions Significant power savings from software and hardware optimizations Loop Unrolling Max = 17% reduction Median = 15.9% reduction Mean = 15.9% reduction Cache Associativity Max = 12.5% reduction Median = 4% reduction Mean = 5% reduction
Future Work & Directions Study combined effects of optimizations Apply these optimizations for new microprocessor configurations Apply these optimizations to a larger test suite
References David Brooks, Vivek Tiwari, and Margaret Martonosi, Wattch: A Framework for Architectural-Level Power Analysis and Optimizations, 27 th International Symposium on Computer Architecture (ISCA), June Doug Burger and Todd M. Austin, The SimpleScalar Tool set, Version 2.0, Computer Architecture News, pages 13-25, June Sohail Hirani, Energy Consumption of Encryption Schemes in Wireless Devices, Master’s Dissertation, University of Pittsburgh, Kenneth Keniston, Grassroots ICT projects in India: Some Preliminary Hypotheses, ASCI Journal of Management 31(1&2), C.K. Prahalad and Allen Hammon, Serving the World's Poor, Profitably, Harvard Business Review, September C.K. Prahalad and Stuart L. Hart, The Fortune at the Bottom of the Pyramid, strategy+business, issue 26, SimpleScalar toolset. Wattch toolset.
Questions Any Questions?