Evaluating demand response opportunities for data centers

Klingert, Sonja

PDF
Dissertation_SonjaKlingert.pdf - Veröffentlichte Version
Download (16MB)

URL:	https://madoc.bib.uni-mannheim.de/58236
URN:	urn:nbn:de:bsz:180-madoc-582366
Dokumenttyp:	Dissertation
Erscheinungsjahr:	2020
Ort der Veröffentlichung:	Göttingen
Verlag:	Cuvillier Verlag
ISBN:	978-3-7369-7330-5 , 3-7369-7330-6 , 978-3-73696-330-6
Hochschule:	Universität Mannheim
Gutachter:	Becker, Christian
Datum der mündl. Prüfung:	11 August 2020
Sprache der Veröffentlichung:	Englisch
Einrichtung:	Fakultät für Betriebswirtschaftslehre > Wirtschaftsinformatik II (Becker 2006-2021)
Fachgebiet:	004 Informatik
Fachklassifikation:	CCS: Networks → Data center networks, Computing methodologies→ Discrete-event simulation,
Freie Schlagwörter (Deutsch):	Rechenzentrum , Lastmanagement , Smart Grid , Regelleistungsmarkt , Simulation
Freie Schlagwörter (Englisch):	data center , demand response , smart grid , regulation market , simulation
Abstract:	Data center demand response is a solution to a problem that is just recently emerging: Today's energy system is undergoing major transformations due to the increasing shares of intermittent renewable power sources as solar and wind. As the power grid physically requires balancing power feed-in and power draw at all times, traditionally, power generation plants with short ramp-up times were activated to avoid grid imbalances. Additionally, through demand response schemes power consumers can be incentivized to manipulate their planned power profile in order to activate hidden sources of flexibility. The data center industry has been identified as a suitable candidate for demand response as it is continuously growing and relies on highly automated processes. Technically, data centers can provide flexibility by, amongst others, temporally or geographically shifting their workload or shutting down servers. There is a large body of work that analyses the potential of data center demand response. Most of these, however, deal with very specific data center set-ups in very specific power flexibility markets, so that the external validity is limited. The presented thesis exceeds the related work creating a framework for modeling data center demand response on a high level of abstraction that allows subsuming a great variety of specific models in the area: Based on a generic architecture of demand response enabled data centers this is formalized through a micro-economics inspired optimization framework by generating technical power flex functions and an associated cost and market skeleton. As part of a two-step-evaluation an architectural framework for simulating demand response is created. Subsequently, a simulation instance of this high-level architecture is developed for a specific HPC data center in Germany implementing two power management strategies, namely temporally shifting workload and manipulating CPU frequency. The flexibility extracted is then monetized on the secondary reserve market and on the EPEX day ahead market in Germany. As a result, in 2014 this data center might have achieved the largest benefit gain by changing from static electricity pricing to dynamic EPEX prices without changing their power profile. Through demand response they might have created an additional gross benefit of 4 of the power bill on the secondary reserve market. In a sensitivity analysis, however, it could be shown that these results are largely dependent on specific parameters as service level agreements and job heterogeneity. The results show that even though concrete simulations help at understanding demand response with individual data centers, the modeling framework is needed to understand their relevance from a system-wide viewpoint.