Abstract
In cloud, service providers and consumers are primary stakeholders that maintain a business liaison. Cloud service providers (CSPs) offer the services, and consumer uses the services on a payment basis. From a business perspective, the selection of a service based on mutual evaluation benefits both the CSPs and consumers. This paper presents an efficient CloudConsumerism model where the multicriteria decision-making method (MCDM) method, TOPSIS, is used for evaluating the performance of CSPs and consumers. For performance evaluation of CSPs, the performance attributes defined by Cloud Service Measurement Initiative Consortium (CSMIC) are exploited. For evaluating the consumers, this paper is the first approach towards identifying the behavioral attributes for evaluating the cloud consumers analogous to the business models. A service mapping algorithm is proposed for efficient (less overhead and higher robustness) mapping. Extensive simulation experiments are conducted; the results show that the proposed framework can be used for the online cloud-based platform due to limited overhead and high robustness.
1. Introduction
Cloud computing is a business model [1] where provider-consumer liaison is formed. The consumer goes through the services from the catalog of the multiple cloud service providers (CSPs) and selects the services from the CSP, which can provide higher performance at the lowest rate. The services offered by the CSPs fall in one of three categories, i.e., infrastructure as a service (IaaS), platform as a service (PaaS), and software as a service (SaaS) [2]. Manually selecting a service having multiple attributes offered from different CSPs is very tedious for the consumers. A new entity, broker, is evolved, which selects the services from the different CSPs on behalf of the consumer. The broker organization can be a third party or a part of the participating organizations. Some well-known organizations, namely, AWS Service Broker (https://http://aws.amazon.com/partners/servicebroker/), IBM Multicloud Management Services (https://www.ibm.com/in-en/services/cloud/multicloud), Cloudmore (https://web.cloudmore.com/cloud-broker-cloudmore), Cloud Services Brokerage (https://www.jamcracker.com/cloud-service-brokerage) etc., are playing the role of cloud broker.
The demand for cloud-based services is increasing day by day. COVID-19 disruption has also accelerated the usage of cloud-based services [3]. To fulfill the ever-increasing demand for cloud-based services and generate more revenue, many big IT organizations are playing the role of CSPs, namely, Amazon, Microsoft, Google, IBM, etc. [4]. Due to enormous expansion in the number of CSPs and consumers, their efficient selection while mapping has become complicated. In literature, quality of service (QoS)-based selection of a CSP for a consumer is recommended for the efficient service mapping [5–7]. The selection of a CSP for the consumer is a unidirectional approach where the selection of the service is based on the CSP’s evaluation in the context of the QoS requirements of a consumer. For making the cloud business successful, the bidirectional evaluation of the participating entities (CSPs and consumers) involved in the mapping process is necessary for increasing their satisfaction.
For evaluating the CSPs, Cloud Service Measurement Initiative Consortium (CSMIC) has introduced performance metrics that are combined in the form of Service Measurement Index (SMI) [8]. In contrast, no standard parameters are defined for evaluating the behavior of consumers for the CSPs. Consumer behavior is about the approach of how people buy and use merchandise and services [9]. Consumer behavior helps in identifying whom to target, how to target when to reach them, and what message is to be given to them to reach the target consumers to buy the product. Analyzing consumer behavior helps the CSPs in improving brand equity and boosting sales. For evaluating consumer behavior, some consumer behavior analytics tools such as Mixpanel (https://mixpanel.com/), Google Analytics (https://support.google.com/analytics/answer/7126596?hl=en), and Kissmetrics (https://www.kissmetrics.io/) are exploited. These tools maintain the complete profile of consumers from website usage to social media engagement. In marketing research, RFM (recency, frequency, and monetary) model [10] [11, 12] is also proposed for evaluating consumer behavior.
Consumer evaluation plays an important role to make the business successful. Earlier frameworks are proposed for the evaluation of CSPs and consumers. The TRCSM framework is presented for evaluating the CSPs and consumers during service mapping [7]. The proposed framework uses prepurchase transaction attributes turnover, duration, and transaction for evaluating the cloud consumers. Another framework, MECSM [12], is proposed which uses a standard RFM model for evaluating the consumers parallel to the CSPs. RFM model uses prepurchase behavior attributes, recency, frequency, and monetary attributes for evaluating the consumers. The prepurchase behavior shows their behavior in past transactions. In the cloud, the same service is offered from different geographical regions, and sometimes service is available in some geographical regions, and it is not available in another geographical region. Observe the consumer behavior during the owning of the service (in-purchase behavior) while the change in service, location, price, and quality is also important. To improve the service quality and increase the consumer base, it is also necessary to observe consumer postpurchase behaviors.
Due to the dynamic need of the current IT industry, the demand of IT resources varies frequently [6, 13]. For example, for a large computing task, efficiency is the main deciding factor. For time-critical tasks, distance is a high priority. Further, a server’s distance is different from different consumers. Also, a consumer may have multiple demands with different priorities. On the other hand, service providers may also have multiple different criteria for evaluating a consumer, e.g., consumers involved for a long duration, consumers with a high transaction amount, and a consumer with less retention rate. Evaluating an entity based on multiple attributes is a multicriteria decision-making problem (MCDM) [14–17]. The glaringly used MCDM methods in the cloud environment are AHP [18], TOPSIS [19], VIKOR [20], and PROMETHEE II [21]. These MCDM methods can be used for ranking both the provider as well as the consumer. Since cloud computing is an internet-based technology, therefore, the execution time of the MCDM method will greatly impact the service performance. In our previous work [6], the TOPSIS method is suggested in terms of lower execution time and higher robustness.
In this paper, CloudConsumerism model is proposed which evaluates the CSPs by exploiting the QoS attributes defined by CSMIC framework 2.0 [8], and consumer behavior attributes are defined for evaluating the consumers in the cloud environment. The service mapping algorithm is developed for efficient service mapping covering all the mapping scenarios (one to one, one to many, many to one, and many to many). A case study is presented for showing the process of service mapping. The experimental analysis is performed on large-scale synthetic dataset for showing its applicability in online cloud-based service selection.
1.1. Motivation and Contribution
Cloud-based services are geographically distributed to support the service consumer organizations in terms of lower upfront cost and dynamic change of hardware/software needs. The COVID-19 disruption has also accelerated the usage of cloud-based services [3]. According to a Gartner report [22], end-user spending on CSPs is forecast to grow 23.1% in 2021 to total $332.3 billion, up from $270 billion that was in 2020. The extensive adoption of cloud-based services in various emerging domains is imposing new challenges and forcing researchers to think about new strategies. To the best of our knowledge, a few research [7, 12] has been conducted where cloud consumers are analyzed parallel to the CSP. The main aim of this paper is to identify the consumer behavior attributes for evaluating the cloud consumer parallel to the CSPs. After the performance evaluation of both the CSPs and consumers, a ranking-based service mapping algorithm is designed which maps a large-scale CSPs and consumers within the limited time overhead of 10 seconds.
The next section discusses the performance evaluation attributes of cloud consumers. Section 3 presents a system model. In Section 4, a case study is presented to show the process of service mapping and the satisfaction of mapped CSPs, and consumers are also evaluated. The experimental analysis is presented in Section 5. The last section concludes the research paper with some future directions.
2. Consumerism in Cloud
In the cloud, consumers can be an individual or an organization. In this work, small organizations are considered as consumers. Organizational buying behavior is mostly a group process. In an organization, a single person does not typically make a buying decision. The service provider must understand group behavior. This group of people may be actual buyers, the people who impact directly or indirectly on the buying behavior, and the employees of the organization who are going to use these services. For evaluating the group behavior, the attributes are defined in three categories as presented in Figure 1.
2.1. Prepurchase Behavior
The prepurchase behavior of the consumer demonstrates the previous transactions with the provider organization and the complete history of the consumer buying behavior. The buying behavior of the consumer can be predicted by using the following attributes. (i)Recency. Recency refers how recently a consumer had made a purchase. It is believed that the more recently a consumer has purchased with the CSP, the more likely he or she will continue to keep the service provider and their services in mind for consequent purchases. Compared with consumers who have not acquired from the CSP for a longer period, the probability of engaging in future transactions with recent consumers is arguably higher. Such information can be used to hark back recent consumers to review the business soon to continue meeting their purchase needs. The tendency of recency attribute is considered as negative (lower value is better)(ii)Frequency. Frequency refers that within a given time, how many times the business deal has been made. The frequency of a consumer’s transactions may be affected by factors such as the type of product, the price point for the purchase, and the need for replenishment or replacement. If the purchase cycle can be predicted, for example, when a customer needs to buy new services, marketing efforts could be directed towards reminding them to visit the business. The tendency of frequency attribute is considered as negative (lower value is better)(iii)Monetary. Monetary value stems from the lucrativeness of expenditures the consumer makes with the business during their transactions. A natural tendency is to put more prominence on encouraging consumers who spend the most money to continue to do so. While this can produce a better return on investment in marketing and consumer service, it also helps in separating those consumers who have been consistent but have not spent as much with each transaction. The tendency of monetary attribute is considered as positive (higher value is better)(iv)Duration. Duration shows the time the consumer is engaged with the CSP. According to some theories, a consistent consumer buys the services slowly and returns small profit regularly. The tendency of duration attribute is considered as negative (lower value is better)(v)Churn rate. It is defined as the percentage rate of consumers who cancel or do not subscribe to a service or consumer cancelation over time, usually on an annual basis. The consumer organization which has a lower churn rate will have a constant or higher demand for services. Also, the probability that a single consumer will cancel during a specific time is less. The tendency of churn rate attribute is considered as negative (lower value is better).where denotes the number of consumers, is the amount of elapsed time, and is the number of consumers canceling in time (vi)Consumer acquisition cost. It is defined as the total marketing cost of acquiring a new consumer. Total marketing expenses contain advertising costs, commissions and bonuses paid, and salaries of marketers and sales managers. The tendency of customer acquisition cost attribute is considered as negative (lower value is better). The consumer acquisition cost is defined as(vii)Retention cost. Consumer retention cost (CRC) is defined as the total cost of retaining a consumer. It is assumed that a 5% increase in retention rate boosts the organization’s profitability by 25% to 95%. A strategic organization manager should not only focus on retaining the consumers but also focused on the CRC used for retaining the consumer. The tendency of retention cost attribute is considered as negative (lower value is better)(viii)Recurring revenue. Recurring revenue is the economy metric of subscription of the service over a fixed time. The consumer organization which has higher recurring revenue will have a higher monetary value. The tendency of recurring revenue attribute is considered as positive (higher value is better). The amount of subscription revenue owed by a consumer over a fixed time, usually measured monthly (MRR), quarterly (QRR), or annually (ARR).where denotes the subscription revenue owed during the amount of elapsed time
2.2. In-Purchase Behavior
In-purchase behavior of the consumer describes the behavior of the consumer during the purchase of services. The behavior of consumers during the purchase can be estimated by the following attributes. (i)Flexibility. Flexibility is the inherent ability of the consumer to change or adapt or react to a decision-making environment with little change in cloud service cost and performance. The tendency of attribute is considered as positive (higher value is better)(ii)Persistent. Despite a little difficulty in in-service performance, the consumer is continuing firmly with the existing CSP. The tendency of the attribute is considered as positive (higher value is better)(iii)Legal ethics. Legal ethics are principles and values which, together with rules of conduct and laws, regulate a profession, such as the legal profession. A consumer, who is aware of the legal ethics of the geographical region from where the services are provisioned, does not blame the CSP regarding privacy issues. The tendency of attribute is considered as positive (higher value is better)(iv)Willingness. The different consumers have different willingness to purchase a service and to pay for the service. The willingness of the consumer for the modified old services is lesser than the new technology-based services. A consumer who is willing to adopt the changes in technology is considered good for the perspective of the CSP. The tendency of attribute is considered as positive (higher value is better)(v)Price sensitivity. Many consumers are price-centric if they do not see any deal; they are not going to make a transaction. The consumers who are focused on the performance of the service without the rigid boundary cost are considered good consumers. The tendency of attribute is considered as negative (lower value is better)(vi)Brand consciousness. Consumers who are purchasing the services from the nationally renowned, expensive, and best-selling brands come in this category. These consumers are also called “price equals quality.” Consumers assume that a high price tag is an indicator of a product of higher quality(vii)Confusion from over choice. Consumers are very quick in deciding on the service selection among the available services, but due to largely available competitors, sometimes consumers get confused. The consumers who are clear about their demand and goals are beneficial for the CSP
2.3. Postpurchase Behavior
Postpurchase behavior describes the behavior of the consumer, such as the way of consumer thinks, feels, and act after provisioning the service. (i)Viral growth. Viral growth refers to the number of new consumers an existing consumer brings to your services/product in a defined period. These consumers are also called an advocate of the service(ii)Feedback. Consumer feedback is the information provided by the consumer about the experience of usage of the service. Positive feedback on the service motivates other consumers to make a purchase(iii)Satisfaction rate. The repurchase of the services by the consumer is a positive sign of satisfaction rate. The repurchase of the services from the same brand also helps the CSPs in generating more revenue(iv)Maintenance. Maintenance cost refers to any cost incurred by an organization to keep its services in operational condition. These costs may be for general maintenance like running antivirus software, or they may be fixing some hardware/software problems. A consumer who requires low maintenance of services is beneficial for the service provider(v)Usage cost. Cloud services have usage-based pricing. Consumers whose usage of the services after provisioning is more is considered beneficial for the service providers(vi)Cash flow problems. A cash flow problem arises when an organization is unable to pay its debt due to late and slow-paying consumers. A consumer is considered valuable in terms of cash flow if he pays fast
3. CloudConsumerism: A Consumer-Centric Ranking Model for Efficient Service Mapping
This section presents the hypothesis and components such as possible service mapping scenarios, MCDM method TOPSIS used for ranking the CSPs as well as consumers, and Algorithm 1: service mapping algorithm for efficiently mapping the CSPs to the consumer in all the possible scenarios.
3.1. Service Mapping Scenarios
In the brokerage-based service selection, the number of cloud entities that needs to be evaluated is two; therefore, the degree of mapping is two, and the number of possible scenarios is four. The proposed CloudConsumerism model deals efficiently with all possible scenarios.
Case 1. One to one mapping. In this case the single cloud broker provides service that is provisioned by one cloud service provider to a consumer as shown in Figure 2.
Case 2. One to many mapping. In this case a cloud broker provides service that is provisioned by a cloud service provider to many consumers as shown in Figure 3.
Case 3. Many to one mapping. In this case many consumers request a service to the cloud broker, and only one cloud service provider is available for the service provisioning as shown in Figure 4.
Case 4. Many to many mapping. In this case many consumers are requesting a service to the cloud brokers, and many cloud service providers are available for the service provisioning as shown in Figure 5.
3.2. MCDM Method TOPSIS
CloudConsumerism model uses TOPSIS method [23] for evaluating the stakeholders. The method is based on the idea that the selected CSP must be closest to the positive ideal solution and farthest from the negative ideal solution. The steps of TOPSIS methods are the following:
Step 1. Normalize the decision matrix , where for CSP for attributes are given by
Step 2. Calculate the weighted normalized decision matrix , where for CSP for attribute is given by
Step 3. Calculate the positive ideal solution and the negative ideal solution .
where
is associated to the attribute which is beneficial attribute.
is associated to the attribute which is cost attribute.
Step 4. Evaluate the separation distance from positive ideal solution and negative ideal solution . where and are L2-norm distance of the target CSP from the positive and negative ideal solution, respectively.
Step 5. Calculate the relative closeness.
Step 6. Rank the CSP based on relative closeness.
3.3. Service Mapping Algorithm
This subsection presents the service mapping algorithm. The Algorithm 1 is used by the proposed CloudConsumerism model during the service mapping process.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
4. Case Study
In the presented case study, the offered QoS data is collected from three IaaS public CSPs: Amazon EC2, Rackspace, and Windows Azure. The SMI attributes, namely, accountability, agility, assurance, cost, performance, and security, are considered for estimating the ranking the CSPs. The QoS data offered by CSP1 (Amazon EC2), CSP2 (Rackspace), CSP3 (Microsoft), and relative weights of consumer’s requirement is presented in Table 1. The QoS data is collected from the paper [5]. Due to unavailability of data of QoS attributes accountability and security, the values are randomly assigned. QoS attributes comparison rules and scaling are adopted from the paper [24]. For the ranking estimation of consumers, the transaction attributes recency, frequency, monetary, duration, recurring revenue, churn rate, and acquisition cost are considered, and data values are assigned randomly, as presented in Table 2. The preferences of CSPs for selecting the valuable consumers is presented in Table 3.
The TOPSIS method is applied to rank the CSPs and consumers based on the preferences of their counterparts. The ranking of CSPs is shown in Figure 6. In business perspective, every consumer follows a cost-quality trade-off. Thus, cost-quality-based service ranking is considered as the final stage of selection of CSPs. Since the offered cost of the CSP for the services is different, the ranking score of CSPs per unit cost is estimated as presented in Figure 7. From the figure, it is seen that CSP1 is ranked as the first choice by all the consumers. The CSP2 is ranked as the second choice except for consumer6, and CSP3 is ranked as the last choice by all consumers except consumer6. Thus, for CSP1, all the consumers are in competition. Therefore, consumers are ranked based on the consumers transaction behavior attribute value and desired consumer weighting by the CSPs. Ranking results of consumers are shown in Figure 8. The duration of the transactions differs from consumer to consumer. The ranking score per unit time is generated for the healthy comparison as shown in Figure 9, which is considered at the final stage of ranking. From the results, it is observed that consumer10 is ranked at first choice by the CSP1 and others also. Considering the mutual choice, the consumer10 is mapped to the CSP1. CSP2 is preferred at second choice only by consumer6; therefore, consumer6 is directly mapped (no need of ranking the consumer because of no competition) to the CSP2. CSP3 is preferred as the second choice by the rest consumers. The rest of consumers are ranked based on the requirement of CSP3, and it is found that CSP3 prefers consumer1 as their second choice (first choice is consumer10 which is already mapped), so consumer1 (second choice) is mapped with the CSP3. The mapping of consumers and CSPs after considering mutual evaluation is presented in Figure 10.
4.1. Satisfaction of Mapped Consumers and CSPs
After mapping the of consumers and CSPs, the service satisfaction of consumers and CSPs are computed [12]. For computing the satisfaction, the ranking score of the CSPs/consumers is considered. The satisfaction of the CSP/consumer is calculated using the offered and received performance. For the satisfaction of consumers, the ranking score of CSP, which is at the first choice based on the requirement of a consumer, is considered as offered performance for the consumer. The ranking score of actual mapped CSP after considering the ranking of consumers based on the requirement of CSP is considered as received performance and vice versa. The case study results show that consumer1 is mapped to CSP3, consumer6 is mapped to CSP2, and consumer10 is mapped to CSP1.
The CSP1 is ranked as the first choice by consumer1. The ranking score of CSP1 (1.4577) is considered as the desired/offered performance for the consumer1, and at the final stage of mapping, CSP3 (0.7592) is mapped to the consumer1, which is considered as actual received performance; therefore, the satisfaction of consumer1 is evaluated as:
The CSP1 is ranked as the first choice by the consumer6. The ranking score of CSP1 (0.4605) is considered as the desired/offered performance for the consumer6, and at the final stage of mapping, CSP6 (0.3040) is mapped to the consumer6, which is considered as received performance for the consumer6; therefore, the satisfaction of consumer6 is evaluated as
The CSP1 is ranked as the first choice by the consumer10. The ranking score of CSP1 (1.0058) is considered as the desired/offered performance for the consumer10, and at the final stage of mapping, CSP1 (1.0058) is mapped to the consumer10, which is considered as the received performance. For consumer10, the offered performance and the received is equal; therefore, the satisfaction of consumer10 is 100%.
5. Experimental Results and Discussion
For experimental analysis, the MCDM methods are implemented in JAVA, running on Windows 10 Home Single Language with Intel Core i7-4510U @ 2.00GHz 2.60 GHz having 8 GB RAM. The ranking overhead is evaluated with variations in the number of CSPs and the number of consumers. The number of attributes for CSPs as well as consumers is fixed. The benchmark for ranking overhead is considered 10 seconds. The mapping overhead of three different scenarios (by varying the number of CSPs and consumers) is recorded. In the first scenario, the number of consumers and CSPs are considered equal, i.e., 500 each and the ranking overhead is recorded. The results show that within a limited time overhead of 10 milliseconds, the mapping overhead increases linearly and scales mapping up to 2500 consumers and 2500 CSPs, as shown in Figure 11. In the second scenario, the number of CSPs is fixed to 2500 (considering as a scaling limit as obtained from the first scenario), and the number of consumers is varied from 500 to 2500 with a step size of 500. The results show that the ranking overhead increases linearly, as shown in Figure 12, and within a time overhead of 8 milliseconds, the 2500 consumers and 2500 provides are mapped. In the third scenario, the number of consumers is fixed to 2500, and the number of CSPs is varied from 500 to 2500 with a step size of 500. The ranking overhead increases linearly as the number of CSPs increases, and within 8 milliseconds, 2500 consumers and 2500 CSPs are mapped as shown in Figure 13. For each data value, the results are taken 40 times, and the average execution time is considered the final ranking overhead in each scenario.
6. Conclusion and Future Work
In this paper, we have addressed a consumer-centric quality of service-based service mapping model, CloudConsumerism. In the proposed model, for evaluating the consumers, the behavioral attributes of consumers are defined, and while mapping, the consumers and CSPs both are evaluated. The number of consumers and service providers is scaleup up to 2500 consumers and 2500 providers. We demonstrate the applicability of our work in the online cloud service selection model by increasing the scaleup of the number of consumer and the number of providers within the limited time overhead of 10 seconds. By applying the requirement filtering from both the side, i.e., cloud consumer as well as CSP, it is shown that the valuable consumers are mapped to the high-performance service provider and hence the satisfaction of the cloud consumer and CSP is increased. The comparative analysis with existing frameworks has been presented in Table 4. In the future, the separate broker for each consumer and provider will be involved so that the third-party broker can also be evaluated from both the side consumer as well as provider. For evaluating the broker, the new attribute will also be defined. From Table 4, it is concluded that the modified MECSM is more efficient in terms of overhead and robustness.
Data Availability
The data used to support the findings of this study are currently under embargo, while the research findings are commercialized. Requests for data, 12 months after the publication of this article, will be considered by the corresponding author.
Conflicts of Interest
There is no conflicts of interest of any authors.