INTRODUCTION
Research background
Since the early 21st century, the IT industry has emerged as a major economic driver globally, with over seven billion devices connected to the internet, making data a valuable commodity that requires effective management Businesses increasingly prefer outsourcing data management rather than maintaining internal server systems According to Artizon's report, "Data Center Market - Global Outlook & Forecast 2022-2027," the data center market was valued at $215 billion in 2021 and is projected to grow to $288.3 billion by 2027 In 2021 alone, 400 new data centers were established, particularly in countries like the US, China, Japan, Australia, the UK, Germany, India, Saudi Arabia, South Africa, and several Southeast Asian nations, driven by incentives such as tax exemptions and favorable energy policies.
Hyperscale data centers are data centers that are significantly larger than enterprise data centers with over 5,000 servers, and 10,000 square feet (Vertiv, 2021)
The rise of hyperscale data centers by major tech companies like Google, Facebook, AWS, Alibaba, and Microsoft has led to significant challenges, particularly concerning electricity consumption These facilities require near-constant operation, with the Uptime Institute mandating a minimum uptime of 99.67% for Tier 1 data centers, equating to 28.8 hours of annual downtime In contrast, Tier 4 data centers are expected to maintain an impressive 99.99% uptime, allowing for only 0.8 hours of downtime each year Currently, data centers account for approximately 1% of global electricity usage, with the EU reporting that these facilities consumed 2.7% of the region's total electricity in 2018, projected to rise to 3.2% by 2030 Notably, Singapore's data centers utilize 7% of the nation's electricity.
In 2020, a ban on new data centers was lifted, leading to a new requirement in 2022 that mandates increased energy efficiency for newly constructed data centers (Mah, 2022).
To address the challenges of sustainability, numerous global initiatives are promoting greener data centers, including the Science Based Targets initiative (SBTi), the Climate Neutral Data Center Pact, the Long Duration Energy Storage (LDES) Council, and RE100 The green data center market, valued at $35.58 billion in 2021, is projected to grow to $55.18 billion by 2027, as reported by Arizton’s “Green Data Center Market - Global Outlook & Forecast 2022-2027.”
Vietnam, classified as a Lower Middle-Income Country, aims to achieve developed nation status by 2045, prompting the government to prioritize the growth of fintech, AI, e-commerce, software outsourcing, and educational technology With a strong emphasis on 4.0 technology and digitalization, trends such as remote work, electronic administration, and social media are emerging, highlighting the importance of data centers as a crucial industry that will serve as the backbone of the Internet.
As of 2021, Vietnam's data center industry remains relatively small, primarily controlled by four key players: FPT, VNTP, Viettel, and CMC In recent years, there has been a noticeable emergence of smaller-scale cloud-based service operations The unexpected impact of Covid-19 has not only accelerated the need for digitization but has also fostered its continued development.
The data center market in Vietnam is currently small, with most facilities situated in urban centers like Ho Chi Minh City, Hanoi, Danang, Binh Duong, and Can Tho In contrast, international trends show a shift towards relocating data centers to rural areas, as evidenced by one-third of such facilities in the US being located outside city centers (Isberto, 2021) These rural data centers benefit from lower land prices, reduced energy costs, and the ability to build one or two-story structures, allowing them to leverage economies of scale for competitive advantage Additionally, environmental concerns regarding data centers in Vietnam remain minimal, with a lack of clear government strategies or corporate initiatives aimed at promoting greener practices in the industry.
Research problem formulation
The location problem for hyperscale data centers is primarily a cost minimization challenge, as predicting revenue based on location is complex due to their reliance on online business rather than local competition Unlike traditional facilities, hyperscale data centers function as both data storage and cloud computing processing plants, necessitating 24/7 operation to accommodate users across different time zones This unique requirement positions the location issue similarly to a facility location problem, prompting critical questions regarding optimal site selection.
What are the key criteria affecting the selection of hyperscale data centers’ placement in Vietnam?
How are these criteria ranks based on their importance to the selection decision?
Research objectives
This research aims to identify key criteria for selecting data center locations in Vietnam and develop a Multi-Criteria Decision-Making (MCDM) model through literature review and expert interviews Utilizing fuzzy Analytic Hierarchy Process (AHP) to analyze expert opinions, the study calculates weight scores for each criterion It further highlights the characteristics of data center placement by comparing expert and non-expert models To illustrate the application of this selection process for managers and stakeholders, a simulation was conducted.
Research scope
This study, conducted in Vietnam between October 2021 and May 2022, examines the decision-making process involved in selecting locations for hyperscale data centers, drawing insights from industry experts in the region.
Research paper structure
The chapter introduces overview of the research background of the data center industry in Vietnam and in the world, research problem formulation, research objectives, research scope and research structure
This chapter contains the review of existing research in decision making, multicriteria analysis, and facilities location selection problem
This chapter discusses the research design, the process of data collection and the AHP methodology and fuzzy set methodology used
Chapter 4: Data analysis and discussion
This chapter presents the research findings, which are analyzed to highlight their theoretical and practical implications for decision-making literature and the data center industry in Vietnam.
This chapter concludes the research paper and provide implication for stakeholders in the data center industry in Vietnam as well as research contribution and limitation
LITERATURE REVIEW
Facilities location problem
Site selection is a critical challenge for businesses, influenced by location theory, which examines the geographic factors affecting economic decisions and the behaviors of buyers and sellers (Gorter & Nijkamp, 2001) The optimal location theory identifies three key approaches: cost minimization, revenue maximization, and net benefit maximization Introduced by Weber in 1909, the Theory of the Location of Industry emphasizes minimizing transportation costs through a locational triangle Modern economic growth theory, as discussed by Krugman (1991), highlights the interconnection between trade and location Facility location decisions are strategic and have lasting effects on operating costs, delivery speed, and competitiveness, particularly in Just-In-Time and flexible distribution systems These decisions also impact finance, marketing, human resources, and production goals (Yang & Lee, 1997) Furthermore, location factors vary based on industry characteristics, facility types, and product life cycles, with labor-intensive sectors like fast fashion prioritizing labor costs, while high-tech industries focus on quality of life to attract skilled labor (HBS, 1999).
Numerous studies have explored the site selection problem through various methodologies, employing mathematical models and scientific approaches within operations research Kochetov (2013) identifies four fundamental discrete facility location models: the uncapacitated facility location problem, which aims to minimize transportation costs for all users; the multi-stage facility location model, focused on determining facilities and their connecting paths to reduce opening, production, and transportation costs; and the facility location problem that incorporates user preferences.
In decision-making, multiple stakeholders are involved, allowing for diverse priorities beyond just minimizing production and transportation costs Factors such as travel time and competitive location also play crucial roles in the decision process Notably, in scenarios where two firms compete, the first mover aims for profit maximization while the follower must adapt to the established market dynamics.
Data center location selection
The TIA Standard, developed by the Telecommunication Industry Association (TIA), serves as an international guideline for data center infrastructure, aiding managers in making informed decisions about ICT infrastructure Specifically, the TIA 942 Standard is tailored for data centers and complies with ISO/IEC 17020/17021 While it does not impose strict requirements on data center locations, Annex F offers valuable recommendations for selecting suitable sites for data centers (TIA, 2005).
Table 2.1 Summary of site selection considerations from TIA 942 Standard Annex F
General Follow national, state, and local codes
Follow local, state, and federal accessibility guidelines and standards
Follow seismic standards of the International Building Code Seismic Zone
Free of asbestos, lead-containing paint, PCB’s, and other environmental hazards
Below 3050m elevation for proper cooling Architectural site selection considerations
Be one or two story buildings dedicated to data center
If not dedicated to data center, other tenants need to be non-industrial
If located on upper floor of multi-tenants building, need to have adequate infrastructure as required Electrical site selection considerations
Local utility needs to be able to supply current and future needs for power
If not, site needs to support self-generation, cogeneration or distributed generation equipment Mechanical site selection considerations
Multi-tenant building needs to have air conditioning heat rejection equipment
Need to have a pre-action sprinkler system dedicated to the data center
At least two diversely routed optical fiber entrance rooms with different local access provider offices
Have dedicated access provider equipment located in the data center space in multi-tenants building
Access 24 hrs/day, 7 days/week
When selecting a site for development, it is crucial to avoid locations that pose significant risks or hazards These include areas directly above parking garages, within 100-year flood plains, near earthquake faults, or on hillsides prone to landslides Additionally, sites should not be downstream from dams or rivers, within the flight path of airports, or within 0.8 km of railroads or major interstate highways Proximity to airports, research labs, chemical plants, landfills, rivers, coastlines, and military bases should also be carefully considered, ensuring a minimum distance of 0.4 km and 0.8 km respectively Furthermore, sites should be at least 1.6 km away from nuclear, munitions, or defense plants, adjacent to foreign embassies, and located outside high-crime areas to ensure safety and security.
proximity of police stations, fire stations, and hospitals
alternate uses of the building after it is no longer needed as a data center
Multicriteria decision making
Decision making is a fundamental aspect of human life, attracting research interest for centuries The complexities of making decisions with multiple stakeholders and conflicting opinions have been addressed through methods like the Borda count and the Condorcet principle since the 1700s, and further explored in Kenneth Arrow's social choice theory in 1951 To streamline decision-making processes, it is essential to analyze the alternatives and the criteria used for their evaluation, which is a key component of Multi-Criteria Decision Making (MCDM) within the field of operations research.
The Analytic Hierarchy Process (AHP), introduced by Saaty in 1980, is a widely utilized multi-objective pair-wise comparison method that effectively computes the weight for each criterion in performance-related problems.
8 resource management, corporate policy and strategy, public policy, political strategy, and planning The steps of AHP with a decision maker committee:
(1) Form a committee of decision maker
(4) Collect data and rank each potential location
FAHP, which integrates fuzzy set theory with the AHP method, effectively analyzes expert opinions while addressing uncertainty and the challenges of assigning precise numerical evaluations Initially introduced by van Laarhoven and Pedrycs in 1983 through a logarithmic least squares approach, this method faced time efficiency issues as the number of criteria increased To mitigate this, Chang proposed an extent analysis method in 1996, offering simpler calculations and reduced time consumption Further advancements were made by Hue et al in 2022, enhancing the technique as discussed in Chapter 3.
Multicriteria decision making criteria for facilities location problem
According to Yang & Lee (1997), selecting a location is a dynamic challenge influenced by changing critical factors, often resulting in the absence of an optimal solution; therefore, compromises are necessary in the decision-making process This complexity makes the Analytic Hierarchy Process (AHP) model particularly applicable for addressing facility location issues.
Table 2.2 Selected literature review for criteria for facilities location problem
Criteria for facilities location problem Source
Market o Market growth potential o Proximity to market o Proximity to raw materials
Transportation o Land transportation o Water transportation o Air transportation
Labor o Cost of labor o Availability of skilled workers o Availability of semi-skilled workers
Community o Housing o Education o Business climate
Climate o Temperature o Rain o Humidity o Sunshine
Geological o Earthquake intensity o Flood history o Interruption of earth
Military o Active defensive o Non-active defensive o Frontier threats o Internal threats o Access to supported echelons o Density of supported echelons
Economical o Native expert labors o Economic activities
Access to essential infrastructures is crucial for efficient transportation and resource management This includes well-connected roads, rail systems, airports, and harbors, which facilitate the movement of goods and people Additionally, access to vital water resources and power lines supports various industries, while proximity to dispatching centers and fuel stations enhances operational effectiveness.
Qualitative factors for plant location
Infrastructure availability (roads, sewer system, municipality services)
Housing and residence availability for workers
Qualitative factors for distribution center location
Provincial finance subsidies o Costs associated with logistics
Coordination among supply chain members
Transportation accessibility o Production and operation costs
Tax structure and tax incentives
Environmental o Ash management o Energy-saving o Effect on resources and nature reserves
Soil o Distance from historical-tourist area o Greenhouse gas emission
Social o Policy and legal support
Changes in the energy policy o Work force
Percentage of highly qualified people
Minimum Wage o Impact on Society
New business unit in ICT industry
Quantitative factors o Capital investment costs o Transportation costs o Operational costs
Qualitative factors o Political and economic environment o Legal framework o Competition o Suppliers
ICT specific factors o Human resources availability o Infrastructure availability o Cultural compatibility
METHODOLOGY
Research Process
The research design is a combination of the MCDM problem design and initial interviews to assess research necessity and ending interviews to assess research validity
Review literature to hypothesize the research problem
Conduct interviews with experts to assess research neccesity
Analyse the data from the interviews
Build a research model to conduct MCDM
Collect responses from experts and non-experts for the model
Analyze and interpret result of data analysis
Interview with experts to assess model clarity and usefulness
Simulation using updated expert model
Data collection for research necessity assessment
Two semi-structured interviews were conducted with a Vietnamese 4.0 technology public policy expert and a DC management specialist with 14 years of experience in the IT/ICT industry Each interview consisted of 10 prepared questions, with additional inquiries made for clarification During the discussions, Expert 2 highlighted the importance of uptime requirements and TIA standards.
942 for location selection The interviews confirmed the necessity of the research and both raised the following issues:
(1) There is a lack of interest in environmental consequences regarding DC
(2) The site selection for DC are mainly made by firms and not by the Central Government There are no zoning requirements or incentives at National level
(3) There is a lack of qualified personnel to run DC
(4) There are difficulties in connecting with the national electrical grid Vietnamese electrical grid might not be stable in more rural areas.
MCDM model
In selecting locations for data centers, various criteria from MCDM models, such as environmental conditions, costs, labor availability, government policies, and infrastructure, play a crucial role, as highlighted in Table 2 Unlike traditional businesses, data centers do not compete with one another when situated close together, allowing for the exclusion of market competition from the decision-making process The TIA 942 Standard and insights from expert interviews underscore the importance of these factors in determining optimal site selection for data centers.
Data centers are essential for business operations, and to maintain their stability, TIA 942 recommends avoiding locations within 100-year flood plains due to the severe water damage that can affect electronic equipment Additionally, data centers consume significant amounts of electricity, generating considerable heat that necessitates effective cooling systems Therefore, selecting a location with lower temperatures is a key advantage for optimal data center performance.
When selecting a site for data centers, elevation plays a crucial role, as higher altitudes are associated with lower temperatures and decreased flooding risk, although excessively high altitudes can hinder cooling machinery performance Data centers, which are essential ICT facilities, require significant electricity and high-speed internet bandwidth, alongside water for their cooling systems Cost considerations are vital and encompass land use, construction, transportation, and operational expenses Additionally, labor factors include both the cost of labor and the availability of specialized data center professionals in the region Government support is also a key criterion, focusing on incentives and energy policies that can influence site selection.
Research Methodology
The criteria in this research will be evaluated with Chang (1996) extensive analysis and Hue et al (2022) approach for Fuzzy AHP method for expert model
Fuzzy sets, introduced by Zadeh in 1965, provide a mathematical framework for representing uncertainty and imprecision in natural phenomena These sets are widely used in various fields, particularly in decision-making processes Jain's 1977 work established a decision-making procedure that translates qualitative terms into quantitative ones using fuzzy sets Additionally, Fuzzy AHP integrates pairwise comparisons to enhance decision-making effectiveness.
15 method and fuzzy set theory is superior in capturing experts’ subjective opinions, especially for qualitative criteria which does not have a quantifiable method to evaluate
According to Dubois and Prade (1978), we have the following definition, the fuzzy number T ( ,a a a a w 1 2 , 3 , 4 ; ) is a trapezoidal fuzzy number if its membership function is defined as:
where f T L x and f T R x are respectively the left and right membership functions of T
If a 1 a 2 a 3 a 4 , T becomes a generalized triangular fuzzy number, and can be denoted by T ( ,a a a w 1 2 , 4 ; ) If w1, then T is a normal fuzzy number
The operations for fuzzy numbers are as follow:
(l1,m1,u1).k=(l1.k,m1.k,u1.k) where k is a positive real number
The combination of fuzzy concept and AHP method was extensively by Chang in 1996, converting the fuzzy sets into crisp sets The method can be described as follow:
Let T = {t1, t2, , tn} represent a set of objects, and G = {g1, g2, , gm} denote a set of goals Utilizing Chang's approach, an extent analysis is conducted for each goal (gi) corresponding to every object Consequently, this process yields m extent analysis values for each object, indexed as g1, g2, , gm for i = 1, 2, , n.
M g (j1, 2, , )n are triangular fuzzy numbers (TFNs)
Assuming that ( , , ) i j g ij ij ij
M l m u are the values of the extent analysis of the ith object for m goals, the value of the fuzzy synthetic extent, S i can be defined as:
, , , , 1, 2, i i n n n j j i g g j i j n n n ij ij ij j j j n n n n n n ij ij ij i j i j i j
Letting S 1 ( ,l m u 1 1 , ) 1 and S 2 ( ,l m u 2 2 , 2 ) be two TFNs, the degree of possibility of S 1 S 2 is defined as follows:
The membership degree of possibility can be described as:
where d is the ordinate of the highest intersection point of two membership functions
Figure 3.3 The comparison two fuzzy numbers
The degree of possibility for a convex fuzzy number to be greater than k convex fuzzy numbers S i I ( 1, 2, , )k can be defined as:
The weight vector is expressed by:
Using normalization, the weight vectors, where W is a non-fuzzy number, can be calculated as:
Wang et al (2008) highlighted that Chang's approach could irrationally assign zero weight to valuable decision criteria and alternatives, which may lead to their exclusion from decision analysis, as further illustrated in Chapter 4 Additionally, Liu et al (2020) and Hue et al (2022) noted that Chang's method is ineffective for certain triangular fuzzy numbers, functioning only with normal fuzzy numbers Hue et al (2022) subsequently proposed a new approach to address these limitations.
The generalized triangular fuzzy comparison matrix is expressed by:
where x ij (a b c w ij , ij , ij ; ij ),
1 (1/ ,1/ ,1/ ; ) ij ij ij ij ij x c b a w for ,i j1, ,n and i j.The fuzzy synthetic extents, S i are defined using the correct normalization formula presented by Wang et al (2008) as follow:
, , ; min( ) i i n n n j j i i i i ij g g j i j n n n ij ij ij j j j n n n n ij n n n n i j ij ij k k i j kj ij k k i j kj j j
, , ; min( ) , i n n n n j g ij ij ij ij j j j j
The centroid indices of fuzzy synthetic extents, denoted as S i, can be computed using the method proposed by Dat et al (2012) Given a set of fuzzy synthetic extent values S S 1, 2, ,S n, the centroid point for all fuzzy numbers is represented as C i = (x S i, y S i), where i ranges from 1 to n.
The distance between the centroid point ( i , ), 1, 2, , i S S i
C x y i n and the minimum point G(x min,y min), is determined by:
D S G x x y y where x min min( ), g i y min min( w ij ) l 1 m 2 d u 2 m 1 u 1
Figure 3.4 The distance between the centroid point ( i , ) i S S i
C x y and the minimum pointG(x min,y min)
The weight vector W ( ,w 1 ,w n ) T of the fuzzy comparison matrix can be defined as:
Table 3.1 Saaty's preferences and triangular fuzzy conversion in the pair-wise comparison process
Verbal judgements of preferences between alternative i and alternative j
Hue et al fuzzy triangular sets
Ci is equally important to Cj 1 (1,1,1) (1,1,1,1.0)
Ci is slightly more important than Cj
Ci is strongly more important than Cj
Ci is very strongly more important than Cj
Ci is extremely more important than Cj
Table 3.2 Reciprocal for Saaty's preferences and triangular fuzzy conversion in the pair-wise comparison process
Verbal judgements of preferences between criteria I and criteria j
Hue et al fuzzy triangular sets
Ci is equally important to
Ci is slightly more important than Cj
Ci is strongly more important than Cj
Ci is very strongly more important than Cj
Ci is extremely more important than Cj
Data Collection for MCDM
A questionnaire was distributed to 25 experienced experts in the Vietnamese data center industry, specifically targeting those with a minimum of 5 years in managerial roles Out of the 25 experts, 9 responses were received, with 5 deemed valid and 4 invalid due to participants selecting the same numbers multiple times Upon follow-up for confirmation, only one expert responded and adjusted their answers, resulting in a total of 6 valid responses.
Expert Position Company Experience in
Operation Team Leader CMC Telecom More than 5 years
M&E & Infrastructure Manager HTC-ITC More than 5 years
Technical Manager Viettel IDC More than 10 years
Data Center Manager True IDC
CEO DCServices More than 20 years
A second questionnaire was conducted in Vietnamese and distributed to IT/ICT professionals in Vietnam with a minimum of two years of experience Out of 34 participants, 14 responded, but one did not work in the IT/ICT industry, resulting in 13 valid answers Among the respondents, 31% had 2-5 years of experience, another 31% had 6-10 years, and 7% had 11 or more years of experience.
With 15 years of experience, a significant 31% of participants have over 15 years in their field They have collectively worked for 11 companies, predominantly in the IT/ICT sector, with 10 out of 11 being in this industry and only 1 in banking To simplify the evaluation process for non-expert participants, a modified 5-point scale was utilized instead of Saaty's traditional 9-point scale.
Table 3.4 Saaty's preferences and triangular fuzzy conversion in the pair-wise comparison process for non-expert model
Verbal judgements of preferences between criteria i and criteria j
Hue et al fuzzy triangular sets
Ci is equally important to Cj
Ci is more important than Cj
Ci is strongly more important than Cj
Following the weight calculations for the expert models, two interviews were held with specialists to evaluate the model's validity and effectiveness The first expert, a public policy specialist, had previously participated in the assessment interview, while the second expert is a technical manager with over a decade of experience in the IT/ICT sector, currently engaged in the data center industry.
Following the data analysis of the Fuzzy AHP and the calculation of weights, two expert interviews were conducted to validate the model's effectiveness One interview was with the same policy expert previously consulted for necessity assessment, while the other was with Expert 3, who also contributed to the MCDM data collection.
The data is then cleaned and analyzed using the calculation in part 3.4 on Excel
ANALYSIS RESULTS AND DISCUSSION
Expert model analysis and discussion
The expert model will be analyzed using both Chang (1996) and Hue et al (2022) methods, with calculation detailed in Chapter 3
Table 4.1 Code for criteria and sub-criteria
Criteria & sub-criteria name Code
Availability of DC specialists in the area L2
In Chapter 3, Chang's calculations present fuzzy matrices and priority vectors for two levels of comparison, as shown in Tables 8 to 13, leading to the final weights summarized in Table 14.
Table 4.2 Fuzzy comparison matrix and its priority vector for the first level’s criteria of expert model using Chang approach
Table 4.3 Fuzzy comparison matrix and its priority vector for the second level’s criteria of expert model (Environment) using Chang approach
Table 4.4 Fuzzy comparison matrix and its priority vector for the second level’s criteria of expert model (Accessibility to resources) using Chang approach
Table 4.5 Fuzzy comparison matrix and its priority vector for the second level’s criteria of expert model (Cost) using Chang approach
Co1 Co2 Co3 Co4 Synthetic extent
Table 4.6 Fuzzy comparison matrix and its priority vector for the second level’s criteria of expert model (Labor) using Chang approach
Table 4.7 Fuzzy comparison matrix and its priority vector for the second level’s criteria of expert model (Local Government support) using Chang approach
Table 4.8 Final weight for criteria and sub-criteria of expert model using Chang approach
Natural disaster history 1 Accessibility to resources
Availability of DC specialists in the area 0.819
Wang et al (2008) highlighted that the Chang method resulted in numerous zero weights for sub-criteria, with nine sub-criteria being eliminated entirely While this underscores the significance of factors such as natural disaster history, land use cost, operational cost, availability of data center specialists, and energy policy, it limits the model's applicability due to the exclusion of many criteria Therefore, this research adopts the Hue et al approach in the subsequent section to address the issue of fuzziness.
4.1.2 Using Hue et al method
According to the calculations by Hue et al outlined in Chapter 3, Tables 15 to 26 present the fuzzy matrices and priority vectors derived from two levels of comparison, leading to the final weights displayed in Table 27.
Table 4.9 Aggregated pair wise comparison matrices from experts for the first level’s criteria using Hue et al approach
Table 4.10 Fuzzy comparison matrix and its priority vector for the first level’s criteria of expert model using Hue et al approach
Criteria Synthetic extent Centroid point
Table 4.11 Aggregated pair wise comparison matrices from experts for the second level’s criteria (Environment) using Hue et al approach
Table 4.12 Fuzzy comparison matrix and its priority vector for the second level’s criteria of expert model (Environment) using Hue et al approach
Table 4.13 Aggregated pair wise comparison matrices from experts for the second level’s criteria (Accessibility to resources) using Hue et al approach
Table 4.14 Fuzzy comparison matrix and its priority vector for the second level’s criteria of expert model (Accessibility to resources) using Hue et al approach
Table 4.15 Aggregated pair wise comparison matrices from experts for the second level’s criteria (Cost) using Hue et al approach
Table 4.16 Fuzzy comparison matrix and its priority vector for the second level’s criteria of expert model (Cost) using Hue et al approach
Table 4.17 Aggregated pair wise comparison matrices from experts for the second level’s criteria (Labor) using Hue et al approach
Table 4.18 Fuzzy comparison matrix and its priority vector for the second level’s criteria of expert model (Labor) using Hue et al approach
Table 4.19 Aggregated pair wise comparison matrices from experts for the second level’s criteria (Local Government support) using Hue et al approach
Table 4.20 Fuzzy comparison matrix and its priority vector for the second level’s criteria of expert model (Local Government support) using Hue et al approach
Table 4.21 Final weight for criteria and sub-criteria for expert model using Hue et al approach
Criteria Weight score Sub-criteria Weight score
Natural disaster history 0.548 Accessibility to resources
Availability of DC specialists in the area 0.719 Local Government support
In comparison to Chang's method outlined in section 4.1.1, Hue et al.'s approach yielded no zero values, while still maintaining the relative importance of criteria for effective comparison and ranking Consistent with the findings in section 4.1.1, key sub-criteria such as Natural Disaster History, Land Use Cost, Operational Cost, Availability of DC Specialists, and Energy Policy emerged as dominant factors Furthermore, within the primary criteria, the Environment category received a significantly higher weight, nearly matching that of Accessibility to Resources and Labor This indicates that employing Chang's method may result in the loss of critical significance in certain criteria.
Non-expert model analysis and discussion
The non-expert model is synthesized using the method developed by Hue et al The fuzzy matrices and priority vectors for two levels of comparison are presented in Tables 28 to 39 The final weights derived from this analysis can be found in Table 40.
Table 4.22 Aggregated pair wise comparison matrices from non-experts for the first level’s criteria using Hue et al approach
Table 4.23 Fuzzy comparison matrix and its priority vector for the first level’s criteria of non-expert model using Hue et al approach
Criteria Synthetic extent Centroid point
Table 4.24 Aggregated pair wise comparison matrices from non-experts for the second level’s criteria (Environment) using Hue et al approach
Table 4.25 Fuzzy comparison matrix and its priority vector for the second level’s criteria of non-expert model (Environment) using Hue et al approach
Table 4.26 Aggregated pair wise comparison matrices from non-experts for the second level’s criteria (Accessibility to resources) using Hue et al approach
Table 4.27 Fuzzy comparison matrix and its priority vector for the second level’s criteria of non-expert model (Accessibility to resources) using Hue et al approach
Table 4.28 Aggregated pair wise comparison matrices from non-experts for the second level’s criteria (Cost) using Hue et al approach
Table 4.29 Fuzzy comparison matrix and its priority vector for the second level’s criteria of non-expert model (Cost) using Hue et al approach
Table 4.30 Aggregated pair wise comparison matrices from non-experts for the second level’s criteria (Labor) using Hue et al approach
Table 4.31 Fuzzy comparison matrix and its priority vector for the second level’s criteria of non-expert model (Labor) using Hue et al approach
Table 4.32 Aggregated pair wise comparison matrices from non-experts for the second level’s criteria (Local Government support) using Hue et al approach
Table 4.33 Fuzzy comparison matrix and its priority vector for the second level’s criteria of non-expert model (Local Government support) using Hue et al approach
Table 4.34 Final weight for criteria and sub-criteria for non-expert model using Hue et al approach
Criteria Weight score Sub-criteria Weight score
Natural disaster history 0.245 Accessibility to resources
Availability of DC specialists in the area 0.501 Local Government support
The expert and non-expert models exhibit significant differences in evaluating hyperscale data center locations While both prioritize the same top two criteria, experts rank Cost as the most critical factor, whereas non-experts place Local Government support at the forefront Additionally, non-experts consider Environmental criteria as their third most important factor, in contrast to experts, who view it as the least significant These disparities highlight the varying levels of understanding between experts and non-experts regarding the complexities of data center location decisions.
Data centers must maintain constant uptime, making the potential for service disruption a significant flaw in their infrastructure According to the Uptime Institute (2009) and expert assessments, the history of natural disasters is a crucial factor, although non-experts tend to rank it lower than other considerations Similarly, while experts prioritize stable electricity as essential for uninterrupted data center operations, non-experts do not place as much emphasis on it Furthermore, operational costs are deemed highly important by experts, whereas non-experts often consider transportation costs to be more significant.
In the labor criteria for data centers, a notable inconsistency arises between experts and non-experts, with experts prioritizing the availability of DC specialists as crucial Initial interviews reveal that Vietnam faces a significant shortage of highly skilled professionals in the data center industry, complicating the placement of hyperscale data centers in rural areas This challenge is exacerbated by the preference of available DC specialists to remain in urban centers, where the quality of life is significantly better.
Experts prioritize energy policy when evaluating local government strategies, while non-experts emphasize the significance of incentives in influencing businesses' decisions on data center locations.
Additional comments collected from experts
Expert 1 questioned the inclusion of Labor cost in Operational cost and that commonly data centers are to be located within 30km of city centers as requirements
Experts highlighted the necessity of relocating to rural areas to attract higher-paying talent, while emphasizing the importance of analyzing output costs Clarification was sought regarding a scale that fluctuated between 9 to 1 and back to 9, prompting the provision of an alternative Excel file for user convenience Additionally, considerations were raised by Expert 6 regarding essential factors such as optical fiber networks, electricity infrastructure, oil transportation, security, costs, and environmental impact, along with compliance with Uptime and TIA 942 requirements.
Validity and usefulness assessment interviews
According to a public policy expert, there are currently no local government policies or tax incentives for data centers in Vietnam, making these factors insignificant in location selection The expert also suggested that environmental considerations may be intuitively influencing decisions, as the geography of Vietnam leads to the automatic dismissal of areas like Central Vietnam and Upper Northern Vietnam (mountainous regions) before proper evaluation, contributing to their low rankings.
Both experts agreed that Operational costs might already include Labor cost, and this might confuse survey participants The importance of electricity and DC specialists were agreed by both experts
Ranking of sub-criteria
The composite weights for each sub-criteria are outlined in Table 41 for the original expert model and Table 42 for the revised model, which excludes the Local Government support criteria for current location selection The most significant weight is attributed to the Availability of DC specialists in the area, followed by Operational cost and Construction cost as the second and third most important factors, respectively.
Table 4.35 Composite weights of all sub-criteria from expert model and their rankings against each other
Criteria Sub-criteria Composite weight
Availability of DC specialists in the area 0.122 2
Table 4.36 Composite weights of expert model and their rankings for selecting current locations (Expert model minus Local government support sub-criteria)
Criteria Sub-criteria Composite weight
Availability of DC specialists in the area 0.164 1
Simulation example
Due to time constraints and the challenge of reaching a consensus among experts from a single company, a comparative analysis of locations could not be performed Given the significant differences in strategies across companies, it was deemed inappropriate to involve experts from various organizations in the comparison Instead, a simulation was carried out to demonstrate managerial steps, utilizing data collected from three non-experts through a survey (see Appendix 3).
This analysis utilizes the sub-criteria and weights from Table 42 to compare three distinct locations: Bac Ninh, Quang Ninh, and Son La Son La, a mountainous region in Northern Vietnam, offers significantly lower land costs but is located farther from Hanoi, the northern hub In contrast, Bac Ninh, situated closest to Hanoi, features flat terrain and has seen considerable industrial park development aimed at relocating factories from the capital Quang Ninh stands out as a coastal province, providing convenient access to ports, the sea, and a high-speed optical fiber network.
Table 4.37 Overall rating of 3 sites Sub-criteria Weight Son La Bac Ninh Quang Ninh
DC specialists in the area 0.164 0.070 0.470 0.459
In this simulation, the suitable location is Son La
CONCLUSION AND DICUSSION
Conclusion
The research considered both Chang and Hue et al method for solving fuzziness in experts’ opinions and agreed that Hue et al is the superior method for comparison
A comparison of two weight sets, one comprising experts and the other non-experts, reveals significant differences in their opinions This highlights the importance of recognizing location issues as expert decision-making challenges.
Cost is the primary concern for businesses in the data center industry when selecting a location for a new facility, with operational costs being particularly significant Other key factors include labor availability and local government support, while environmental considerations and resource accessibility are less critical However, stable electricity remains a crucial requirement, as it is essential for data center operations The history of natural disasters is a major concern in Vietnam, given its vulnerability to floods, storms, and landslides due to its extensive coastline and river networks Additionally, the need for more trained data center specialists in Vietnam is highlighted as vital for the industry's growth.
Research implications
In selecting real locations for hyperscale data centers, it is crucial to focus on four key criteria due to the absence of zoning policies, incentive programs, and energy regulations from both local and central governments in Vietnam Energy policies play a vital role for businesses, highlighting the need for collaboration with local authorities to ensure the sustainability of the data center industry, particularly given its significant electricity consumption and environmental impact This underscores a policy gap in Vietnam's data center sector, which is expected to greatly influence the country's future energy usage Therefore, businesses should actively engage with local governments to secure their support, especially in alignment with national digitization efforts.
Secondly, there is a clear opportunity for businesses capable of training DC specialists as the demands are still very high
Vietnam's abundant locations for wind and solar energy present a significant opportunity for hyperscale data centers relocating to rural areas to harness green technology Although the Vietnamese data center industry is still in its infancy, establishing pioneering green data centers could offer long-term advantages for businesses.
Finally, since AHP application is a subjective initial assessment based on subjective judgments of experts, further studies using more rigorous and more objective methods should be applied for this problem.
Research contribution
The research focused on the Vietnamese data center industry, providing valuable insights into the challenges faced by Vietnam as a transitioning economy, where cost remains a primary concern Additionally, it contributes to the body of knowledge on facility location problems by examining a specific type of facility that operates without market competition criteria.
Limitation
The primary limitation of this research is the insufficient time to gather data from industry experts Targeting technical and operations managers at data centers with over five years of experience, the study incorporates insights from various companies Consequently, the rankings do not reflect a singular business strategy but rather provide a general assessment, making it challenging to identify optimal data center locations To address this, a simulation is presented, illustrating how businesses can replicate the process to determine suitable locations by collaborating with experts from the same company who share a unified business strategy.
MCDM is a subjective method that leverages expert opinions to assess location problems To enhance the evaluation process, incorporating rigorous techniques like computational simulation modeling may prove beneficial in future applications.
Arizton, (2022) Green Data Center Market - Global Outlook & Forecast 2022-
2027 Retrieved from https://www.arizton.com/market-reports/green-data-center- market
Chang, D.-Y (1996) Applications of the extent analysis method on fuzzy AHP
European Journal of Operational Research, 95(3), 649–655 doi:10.1016/0377-
Dat, Luu & Yu, Vincent & Chou, Shuo-Yan (2012) An improved ranking method for fuzzy numbers based on the centroid-index International Journal of Fuzzy
Dubois, D., & Prade, H (1978) Operations on fuzzy numbers International Journal of Systems Science, 9(6), 613–626 doi:10.1080/00207727808941724
Ertuğrul, İ., & Karakaşoğlu, N (2007) Comparison of fuzzy AHP and fuzzy TOPSIS methods for facility location selection The International Journal of Advanced
European Digital Strategy (2021) Energy-efficient Cloud Computing Technologies and Policies for an Eco-friendly Cloud Market Retrieved from https://digital-strategy.ec.europa.eu/en/library/energy-efficient-cloud-computing- technologies-and-policies-eco-friendly-cloud-market
Farahani, R Z., & Asgari, N (2007) Combination of MCDM and covering techniques in a hierarchical model for facility location: A case study European Journal of Operational Research, 176(3), 1839–1858 doi:10.1016/j.ejor.2005.10.039
Gorter, C., & Nijkamp, P (2001) Location Theory International Encyclopedia of the Social & Behavioral Sciences, 9013–9019 doi:10.1016/b0-08-043076-7/02490-
HBS (Harvard Business School) (1989), Note on Facility Location, Publishing Division, Harvard Business School, Boston, MA
In their 2022 study published in Sustainability, Hue et al utilized an enhanced Fuzzy Analytic Hierarchy Process to prioritize factors influencing lecturer research productivity The research highlights the critical elements that affect academic output, providing valuable insights for educational institutions aiming to enhance research effectiveness The findings contribute to a deeper understanding of productivity dynamics in higher education, emphasizing the importance of strategic prioritization in fostering academic excellence.
Isberto, M., (2021) Are There Benefits of a Rural Data Center? Colocation America Retrieved from https://www.colocationamerica.com/blog/rural-data-center- benefits
Jain, R (1977) A procedure for multiple-aspect decision making using fuzzy sets
International Journal of Systems Science, 8(1), 1–7 doi:10.1080/00207727708942017
Kahraman, C., Ruan, D., & Doǧan, I (2003) Fuzzy group decision-making for facility location selection Information Sciences, 157, 135–153 doi:10.1016/s0020-
Kochetov, Yury (2013) Facility Location: Discrete Models and Local Search Methods doi:10.3233/978-1-60750-718-5-97
Krugman, P (1991) Increasing Returns and Economic Geography Journal of Political Economy, 99(3), 483–499 doi:10.1086/261763
Liu, Y., Eckert, C M., & Earl, C (2020) A review of fuzzy AHP methods for decision-making with subjective judgements Expert Systems with Applications, 113738 doi:10.1016/j.eswa.2020.113738
Ly, T., (2021) Vietnam in the Post-COVID Era: Realizing a ‘Digital Country’
Asia Pacific Foundation of Canada Retrieved from https://www.asiapacific.ca/publication/vietnam-post-covid-era-realizing-digital- country
Mah, P (2022) Cracking the green conundrum in Singapore, amid a data center moratorium Retrieved from https://www.datacenterdynamics.com/en/analysis/cracking-green-conundrum- singapore-amid-data-center-moratorium/
Marinkovic, S & Nikolić, I & Rakicevic, J., (2018) Selecting location for a new business unit in ICT industry Zbornik Radova Ekonomskog Fakultet au Rijeci 36 801-
Ozgen, D., & Gulsun, B (2014) Combining possibilistic linear programming and fuzzy AHP for solving the multi-objective capacitated multi-facility location problem Information Sciences, 268, 185–201 doi:10.1016/j.ins.2014.01.024
Rooks, T., (2022) Big data centers are power-hungry, but increasingly efficient
| DW | 24.01.2022 Retrieved from https://www.dw.com/en/data-centers-energy- consumption-steady-despite-big-growth-because-of-increasing-efficiency/a-60444548/
TIA, 2005 TIA-942 Telecommunications Infrastructure Standard for Data Center Arlington: Telecommunication Industry Association Retrieved from https://tiaonline.org/products-and-services/tia942certification/ansi-tia-942-standard/
Uptime Institute (2009) Tier Standard: Topology Data Center Site Infrastructure Santa Fe
Wang, Y.-M., Luo, Y., & Hua, Z (2008) On the extent analysis method for fuzzy AHP and its applications European Journal of Operational Research, 186(2), 735–747 doi:10.1016/j.ejor.2007.01.050
Vertiv, (2022) What Is a Hyperscale Data Center? Retrieved from https://www.vertiv.com/en-asia/about/news-and-insights/articles/educational- articles/what-is-a-hyperscale-data-center/
Yang, J., & Lee, H (1997) An AHP decision model for facility location selection
Zadeh, L A (1965) Fuzzy sets Information and Control, 8(3), 338–353 doi:10.1016/s0019-9958(65)90241-x
QUESTIONNAIRE FOR EXPERTS
You are invited to take part in a research study focused on the criteria for locating newly built hyperscale data centers in Vietnam This research is led by Investigator Quynh Pham, with the support of co-supervisors Prof Dr Tanabu Monotari from Yokohama National University in Japan and Dr Luu Quoc Dat from the University of Economics and Business at Vietnam National University in Hanoi.
You are invited to participate in this questionnaire, since you are an expert in the data center industry Please take your time with the questionnaire
The questionnaire is expected to take approximately 15-20 minutes to complete
Participant information will remain confidential, as personal details such as names and addresses are not requested If any personal information is provided, it will be removed from the questionnaire to ensure anonymity Responses will solely be utilized for research purposes and to compile a report.
The findings gathered will fulfill the requirements for the MBA program at Vietnam Japan University, Vietnam National University, Hanoi Additionally, these insights may be utilized in seminars, conference presentations, and academic publications.
I understand that I may refuse to participate in this study freely
I also understand that if, for any reason, I wish to stop participating, I will be free to do so I agree to participate in this study
What is your current job title?
Where are you currently working at?
How many years have you worked in the IT or ICT industry?
How many years have you worked in the data center industry?
Have you been consulted on the problem of data center location in your job before?
Have you participated in the decision making of data center location in your job before?
The data center industry in Vietnam is currently small and dominated by a few key players, but the emergence of smaller cloud-based operations over the past three years signals growth With government initiatives promoting 4.0 technology and digitalization, data centers are poised to become a critical industry supporting the Internet infrastructure The COVID-19 pandemic has accelerated the need for digitization, highlighting the importance of robust data centers amidst the rise of remote work, electronic administration, and social media As a result, the data center sector is expected to become increasingly competitive.
The data center market in Vietnam is currently limited, with most hyperscale facilities situated in urban centers like Hanoi, Ho Chi Minh City, Danang, and Can Tho, except for one in Binh Duong Province near Ho Chi Minh City In contrast, global trends show a shift towards relocating data centers to rural areas, particularly in the US, where one-third are found outside city limits This relocation can enhance energy efficiency by utilizing green technologies and natural water resources, although the significant initial investment poses a challenge This research seeks to identify the key criteria and their relative importance for selecting suitable locations for new hyperscale data centers in Vietnam.
Below is the hierarchy for decision making
Goal: Locate a new hyperscale data center
Criteria: Five criteria are chosen in the evaluation
1 Environment: It refers to environmental conditions of the location, including temperature, elevation and history of natural disasters
2 Accessibility to resources: It refers to the distance from required resources to operate hyperscale data centers
3 Cost: It refers to the cost of building, operating and transporting required by locating data center in the location
4 Labor: It refers to the human resources necessity for a new data center
5 Local Government support: It refers to the support given by local Government including tax and other incentives and energy policies
When evaluating the ENVIRONMENT, please rank the significance of the options listed in the left column against those in the right column on a scale from 1 to 9, where 9 signifies extreme importance and 1 indicates equal importance.
Temperature Natural disaster history Natural disaster history
With respect to ACCESSIBILITY TO RESOURCES, using the scale from 1 to 9 (where
9 is Extremely important and 1 is Equally important), please indicate the relative importance of the option in the left column to the option in the right column
With respect to COST, using the scale from 1 to 9 (where 9 is Extremely important and
1 is Equally important), please indicate the relative importance of the option in the left column to the option in the right column
Land use cost Construction cost
Land use cost Transportation cost
Land use cost Operational cost
When evaluating LABOUR, please rate the importance of the options in the left column compared to those in the right column on a scale of 1 to 9, with 9 representing Extremely important and 1 indicating Equally important.
DC specialists in the area
When evaluating LOCAL GOVERNMENT SUPPORT, please rate the importance of the options listed in the left column compared to those in the right column on a scale from 1 to 9, with 9 representing Extremely important and 1 indicating Equally important.
When evaluating potential locations for a hyperscale data center, please rate the importance of the options listed in the left column compared to those in the right column on a scale from 1 to 9, with 9 being extremely important and 1 being equally important.
Government support Accessibility to resources
Government support SECTION D: Additional comment
Please share your thoughts and recommendation to the research
Thank you for your participation
Any inquiries can be addressed at 20117014@st.vju.ac.vn or quynhpham42@gmail.com
Bạn được mời tham gia nghiên cứu về tiêu chí chọn địa điểm cho trung tâm dữ liệu siêu cường tại Việt Nam, do Phạm Trúc Quỳnh, học viên thạc sĩ tại Đại học Việt Nhật, thực hiện Nghiên cứu được hướng dẫn bởi Giáo sư Tiến sĩ Tanabu Monotari từ Đại học Quốc gia Yokohama và Tiến sĩ Lưu Quốc Đạt từ Đại học Kinh tế, Đại học Quốc gia Hà Nội Sự tham gia của bạn là quan trọng vì bạn là chuyên gia trong lĩnh vực trung tâm dữ liệu Xin vui lòng đọc kỹ nội dung khảo sát.
Khảo sát sẽ kéo dài khoảng 15-20 phút để hoàn thiện bao gồm phần giới thiệu dự án và nội dung
Thông tin cá nhân của người tham gia sẽ được bảo mật và không chia sẻ cho bên thứ ba Tên, tuổi, địa chỉ và các thông tin cá nhân khác sẽ không được yêu cầu, và nếu có, chúng sẽ bị loại bỏ khỏi khảo sát Các câu trả lời chỉ được sử dụng cho mục đích nghiên cứu và viết báo cáo Kết quả nghiên cứu sẽ phục vụ cho chương trình MBA tại Đại học Việt Nhật, Đại học Quốc gia Hà Nội, và có thể được sử dụng trong các buổi thuyết trình tại hội thảo hoặc trong các nghiên cứu xuất bản.
Phần 1: Thông tin cho người tham gia
Tôi hiểu rằng tôi có thể từ chối tham gia vào nghiên cứu này
Tôi cũng hiểu rằng nếu vì bất kỳ lý do gì, tôi muốn ngừng tham gia, tôi sẽ được tự do làm như vậy
Tôi đồng ý tham gia nghiên cứu này
Phần 2: Thông tin về người tham gia
Xin vui lòng cung cấp thông tin về lịch sử công tác và chuyên môn của Quý anh/chị, bao gồm chức vụ hiện tại mà Quý anh/chị đang đảm nhiệm.
Quý anh/chị hiện đang công tác ở đâu?
Quý anh/chị có kinh nghiệm bao nhiêu năm trong ngành IT/ICT?
Quý anh/chị có kinh nghiệm bao nhiêu năm trong mảng trung tâm dữ liệu?
Quý anh/chị đã từng được tham vấn về vấn đề lựa chọn địa điểm cho trung tâm dữ liệu trong công việc chưa?
Quý anh/chị đã từng được tham gia vào quá trình lựa chọn địa điểm cho trung tâm dữ liệu trong công việc chưa?
Phân tích đa tiêu chí cho vị trí đặt trung tâm dữ liệu hyperscale tại Việt Nam cho thấy rằng ngành trung tâm dữ liệu hiện nay chủ yếu do một số công ty lớn như FPT, VNPT, Viettel và CMC chi phối Trong ba năm qua, các dịch vụ điện toán đám mây quy mô nhỏ đã gia tăng đáng kể Sự thúc đẩy của Chính phủ về công nghệ 4.0 và số hóa khẳng định vai trò quan trọng của trung tâm dữ liệu như xương sống của Internet Dù Covid-19 gây ra nhiều thách thức, nó lại thúc đẩy nhanh chóng nhu cầu về số hóa, đặc biệt trong bối cảnh làm việc từ xa và quản trị điện tử, tạo ra một môi trường cạnh tranh mạnh mẽ cho ngành trung tâm dữ liệu trong tương lai gần.
Thị trường trung tâm dữ liệu tại Việt Nam hiện còn nhỏ, với hầu hết các trung tâm hyperscale tập trung ở các thành phố lớn như Hà Nội, TP HCM, Đà Nẵng và gần đây là Cần Thơ, ngoại trừ một trung tâm ở Bình Dương Trái ngược với xu hướng toàn cầu, nơi nhiều trung tâm dữ liệu được chuyển ra vùng nông thôn, đặc biệt là ở Mỹ, nơi một phần ba trung tâm nằm ở khu vực này Việc di chuyển các trung tâm dữ liệu ra khỏi thành phố không chỉ giúp tối ưu hóa tiêu thụ năng lượng bằng công nghệ xanh mà còn tận dụng tài nguyên nước tự nhiên cần không gian lớn hơn.
Nghiên cứu này tập trung vào việc xác định các tiêu chí quyết định và trọng số của chúng, dựa trên ý kiến của các chuyên gia, nhằm lựa chọn vị trí phù hợp cho các trung tâm dữ liệu hyperscale mới tại Việt Nam.
Dưới đây là bảng phân cấp cho việc ra quyết định
Mục tiêu: Xác định vị trí trung tâm dữ liệu siêu cấp mới
Tiêu chí: Năm tiêu chí được chọn trong đánh giá
1 Môi trường: là điều kiện môi trường của địa điểm, bao gồm nhiệt độ, độ cao và lịch sử thiên tai