Multi-Criteria Assessment of Data Centers Environmental Sustainability

Abstract

The size and capacity of Data Centers (DCs) is growing at a rapid pace to meet the increased demand of data processing and storage capacity requested by a digital information society. Since DCs are infrastructures that have large energy consumption, there is a need to change their design approach to make them more efficient and more environment friendly. This research was motivated by the planning of a new DC in Portugal. It proposes a multi-criteria framework to assess the sustainability of a DC, which includes a new metric to evaluate the DC efficiency taking into account the environmental conditions of the DC location. ELECTRE TRI was chosen for aggregating different metrics concerning the environmental sustainability of a DC into sustainability categories. The evaluation methodology allows some freedom for each DC to place more weight on the aspects in which it is stronger, an analysis facilitated by the IRIS decision support system.

Editors’ Comments on “Multi-Criteria Assessment of Data Centers’ Environmental Sustainability”

The main contribution of Chapter “Multi-Criteria Assessment of Data Centers’ Environmental Sustainability”, by M.T. Covas, C.A. Silva and L.C. Dias is the assessment of the environmental sustainability performance of Data Centers (DC) that underlie most common Internet and telecommunications services available today, proposing a framework for that purpose. This case study is hence related to the energy/environment field, a most popular application area for MCDA and a topic that is well covered in this book (see Chaps. 8–14).

The main objective of this application is to help a telecommunications company assess the sustainability of its planned new data center (different variants) in a simple way that could be used as a standard by this industry. Other objectives are to propose: (1) a general framework for assessing the environmental sustainability of data centers, and (2) a new metric for measuring the efficiency of such data centers.

The client of the decision aid application was a telecommunications operator in Portugal. Three experts from the client organization participated in model building (definition of criteria, judging the adequacy of the method, setting preference-related parameters). The three authors of the chapter acted as analysts. Their role was, in particular, to suggest the MCDA methodology and to provide guidance in its use. Experts and analysts collaborated for elaborating a DC assessment framework. One of the authors is a member of the client organization and therefore also acted as a Data Center expert.

Identified phases in the process concern: (1) building a set of criteria, (2) elaborating the scales of the criteria and determining how to asses the DC’s on these scales, and (3) selecting a model for aggregating the evaluations of the DC’s on the criteria and determining the model’s parameters. The duration of the process was about 3 months, from after the new Data Center location was announced. Hence the main goal was not to assess this particular DC. The planned DC was more used as a reference alternative; the set of alternatives used to elaborate or validate the evaluation model were either variants pertaining to choices the client has to make (building a wind farm, reuse of heat) or fictitious (assessing the results if the Data Center was built elsewhere).

The decision problem statement in this case study was ordinal sorting, interpreted as an evaluation method. Based on a literature review on Data Center metrics and incorporating other sustainability concerns of the client the analysts propose five performance criteria, four quantitative and one qualitative. No uncertainties were considered for the alternatives’ performances and the model parameters, except the criteria weights. Each weight is allowed to vary in an interval. The handling of uncertainties shares characteristics with Data Envelopment Analysis: the evaluation methodology allows, for each DC, to put more weight on the criteria on which it is stronger.

Due to, first, the type of result sought (a level on an ordered scale), secondly, the wish to avoid making explicit trade-offs between the criteria, and, thirdly, the wish to deny that a very good performance on one criterion can compensate a poor performance on another criterion (by using veto) a valued outranking model of preference aggregation is chosen. And, considering chosen the decision problem statement (ordinal sorting) the ELECTRE TRI method with five sorting categories is applied in this case study. The parameter setting, i.e., the category limiting profiles, the discrimination thresholds for indifference, preference and veto situations, as well as criteria weights constraints were directly given by the client. Divergence among actors is addressed by agreement between the experts, and a sensitivity analysis of the effect of considering veto situations or not is performed.

Among the tangible results and artifacts achieved in this case study are:

• 1. The criticism of the PUE metric, generally agreed upon as the measure of DC’s energy efficiency, and the proposal of an alternative metric (TRUE) taking into account the temperature of the region, which has an impact on the efficiency of the cooling system,

• 2. The proposal of a set of criteria and the implementation of an evaluation model, including category definitions, for assessing the sustainability of data centers,

• 3. An agreement about the parameter values and weight constraints to be used,

• 4. A tool for disclosing Data Centers’ sustainability performance, suitable for organizations of any size or type, and from any geographic region, and allowing comparability between Data Centers that tackle sustainability with different strategies,

• 5. A (imprecise) classification of the alternatives in ordered categories (more precisely, the alternatives are assigned to an interval of possible categories).

As intangible results we may list in this study, first, a knowledge transfer about MCDA from the analysts to the client organization. The decision aid also helped the company to better define the vision or goals what should be a sustainable DC. Establishing the framework helped to identify what must be assessed to evaluate a DC sustainability performance. The Client wishes to use this approach for helping making architecture choices, in the planning phase for a new data center or a major renovation of an existing building. It can also help Data Center decision making processes related to site selection, e.g. looking for sites where can be increased the use of the heat recovered from the Data Center (e.g. swimming pools; greenhouses, etc.). This framework may furthermore help the industry to have a common understanding of Data Center Sustainability measurement, and can generate dialogue to improve it. It can also foster the promotion of sustainable Data Centers industry internationally. It facilitates transparency and accountability by organizations and provides to stakeholders a universally applicable and comparable framework, from which they can understand disclosed information. Finally, this environmental sustainability assessment framework could be a tool to communicate with customers, to help them to buy services from more eco-friendly Data Centers.

The impact of the decision aiding is to help the client to understand which options would be important to obtain a good classification if this type of categorization is adopted by the industry. It helped the client to extend its concerns beyond the PUE metric, in order to be as good as possible in terms of the other criteria. The tool helped to make architecture choices, in the planning phase for a new data center in order to improve the heat recovering from the DC. Despite the company’s effort to have a strategy focused on energy efficiency (an excellent PUE performance) and an IT resources optimization (e.g. increasing the virtualization and server utilization levels), this would not be sufficient to achieve the best category.