Green Server Design: Beyond Operational Energy to Sustainability

1. Green Server Design: Beyond Operational Energy to Sustainability Wei Wu

2. Introduction • Environmental sustainability • minimize the environmental impact • Importance to design “green” IT systems • Priority work • large work for reducing the operational electricity consumption • do not address environmental impact of a system across all the stages of its lifecycle • Contribution • a methodology: reason about sustainability form a system architecture perspective • a systematic analysis of the environmental impact of current designs across the entire lifecycle, and the tradeoffs with state-of-the-art energy-efficiency techniques

3. Measuring Sustainability: Using Energy for Architecture Studies • Life-cycle assessment(LCA) • Thermodynamic metric of energy consumption • several previous studies: how the consumption of the irreversibility associated with various processes and the environmental sustainability • Models for specific IT systems • optimization to reduce lifecycle energy consumption • well mapped to the optimization based on other environmental criteria

4. Previous lifetime energy characterization • Estimation of environmental impact • mapping of system mass or material flows to per-unit estimates of the environmental impact burden • Process-based breakdown of total energy • not very useful: system architecture choice is not clear • architecture perspective is required

5. Our Approach • Architecture-centric approach • aggregate raw materials at the component level • succinct in terms of system architectural choice • Decomposition • embedded energy • energy used to “make” a system component • extraction, manufacturing, transportation, recycling • appropriate energy destruction values: aggregate them discern the overall energy consumption related to each architectural component • Operational energy • operational energy is equivalent to electricity consume • use maximum power rating, and model how power varies with utilization • Infrastructure energy • cooling and power delivery infrastructure • handle the maximum power rating, use power usage effectiveness (PUE) metric

6. Energy Breakdown for our approach • The embedded energy contributes a sizeable amount • Dominant components are from silicon-based processes and PCB design

7. Evaluatingthe State-of-the-art • Design space exploration • three broad categories of optimizations • Energy proportionality (EP) • be proportional to the activity in the system • Consolidation (Con) • multiple virtual machines on different servers • be consolidated onto a single server, raising its utilization and reducing the requested server count • Low-power server solutions (LP) • based on power-efficiency, but low-power processors • solution: better match the processor architecture to the workload characteristics to leverage significantly better performance/watt

8. Tradeoffs between different Energy-efficiency optimizations • The benefits from EP are primarily a function of workload average utilization • Total energy based exploration • For a given average workload utilization, different tradeoffs for the LP designs is using performance/watt multiplier on the X-axis • For web workloads, prior studies have found LP to yield better multipliers ranging from 2 to 5 • Consolidation: in turn is a function of the peak-of-sum utilization specific to the workload (Y-axis) • lower values indicate the peaks are completely non-synchronized and consolidation can more readily be leveraged

9. Tradeoffs between different Energy-efficiency optimizations (cont.) • The division of the heat map into various regimes shows the technique that achieve the best energy for that region of workload/system configuration • Operational energy based exploration (similar picture) • EP: ideal proportionality • Con: perfect bin-packing • Observations • the figures individually show best and the cross-over points, and relative magnitude of the benefits • comparing the two figures allows us to examine the changes to these design tradeoffs when optimizing for just operational energy vs. considering total energy • total energy is minimized when going towards the bottom left region • EP outperforms Con when workloads are not bursty and don’t lend themselves to packing • Different than when just focusing on operational energy: always better • LP designs are always better than both EP and Con • the optimal choice between LP and EP is dependent on their relative energy efficiencies for the type of workload • the tradeoff between fewer machines and higher utilization is shown as the angled line dividing LP and Con

10. Summary • Focusing on the most efficient system design for operational energy does not always produce the most sustainable solution • The best way to optimize for sustainability • use power-efficient and material-efficient systems • scale power with resource usage and are utilized fully • As the ratio of embedded energy to total energy grows, new optimizations will be needed that explicitly target embedded energy • It is important to notice that embedded energy, operational energy, infrastructure energy and performance are not independent variables • Sustainability becomes a more important design consideration for future system, design methodologies and system optimization need to correspondingly change to address these merging challenge

11. Thank you！ Questions？