Sustainable IT and IT for Sustainability

Liu, 2014

Category: Systems

Overall Rating

3.7/5 (26/35 pts)

This thesis provides rigorous analytical frameworks, notably in Chapter 5, for understanding system efficiency in resource allocation and market design under conditions of uncertainty, strategic behavior, and physical constraints.
While the specific models and empirical results are tied to 2014 data centers and market structures, the analytical methodology for quantifying trade-offs (e.g., prediction error vs. market power) offers a valuable blueprint for analyzing complex, coupled, distributed systems like modern DERs in microgrids or peer-to-peer energy markets.
It is this reusable analytical approach, rather than the specific solutions presented, that holds the most actionable potential for modern research.

The rigorous analytical treatment of distributed optimization for geographical load balancing with theoretical guarantees (Chapter 2), the holistic integrated optimization framework for IT, cooling, and power (Chapter 3), and especially the analytical characterization of prediction-based pricing efficiency under market power and network constraints (Chapter 5) provide foundational insights and mathematical tools.
The mathematical and algorithmic techniques presented are highly transferable.
Modern advancements are perfectly positioned to unlock significant value from this research.
Specifically, the analytical framework developed in Chapter 5 for understanding the efficiency of prediction-based pricing in the presence of participant market power, prediction errors, and network constraints is highly relevant to designing incentive mechanisms for modern peer-to-peer energy trading platforms or community microgrids.

The core models, particularly for IT workloads, cooling infrastructure, and electricity markets, exhibit significant decay in relevance.
Crucially, the electricity market focus, particularly the deep dive into 2011/2012 Fort Collins Utilities CPP data, is highly specific and does not generalize to the dynamic, granular... and often location-dependent pricing and ancillary service markets prevalent today across many grids.
The paper likely faded because its theoretical contributions, while mathematically sound within their simplified models, offered limited practical leverage for the rapidly evolving real-world systems.
The empirical validation, tied to scaled traces and a small testbed (4 servers), represents a significant gap between proof-of-concept and deployable solution.

Watch