QoS prediction for service recommendations in mobile edge computing
Introduction
Mobile edge computing is an emerging technology that provides Web and cloud services within the close proximity of mobile subscribers. Traditional telecom network operators perform traffic control flow (forwarding and filtering of packets), but in mobile edge computing, edge servers are also deployed in each edge server. It also enables application developers and content providers to serve QoS-aware service recommendation based user context information by using real time radio access network information [[1], [10]]. In Mobile edge computing environment, edge server is deployed in between the mobile client and server near mobile proximity. For example, when a mobile web browser sends a request for a URL page, the response from the server is first intercepted at the edge server, since it can device information and analyze users behavior to improve services [1]. Based on the growing popularity of mobile devices, a large number of mobile services have been developed that run on mobile devices and often are invoked by people accessing edge servers in mobile edge computing [25]. Thus, it is important to know which mobile services have better QoS values for performance optimization. Hence, how to predict the QoS values accurately before services are invoked is a very important issue for service recommendation in mobile edge computing.
As it is well known that service QoS data are notably more volatile, and mobile devices often roam in mobile environment [[30], [2]]. Due to the mobility of mobile devices, history QoS data of mobile services in an edge server will fail when mobile devices move in another edge server and the services QoS data in the new edge server is empty. To explain changes of edge server for mobile users easily, we present two types of edge server definition:
Definition 1 Old Edge Server This refers to the edge server from which the active user adopted services before he moved out of its radio coverage.
Definition 2 New Edge Server This refers to the current edge server after the active user moved from the radio coverage of the old edge server. The active user adopts services by accessing this new edge server.
Although many QoS-aware service recommendation approaches [[40], [13], [15]] have been proposed in traditional Internet environments, they often fail to make accurate service recommendation in mobile edge computing because two problems exist that decrease service recommendation accuracy:
(1) Volatility of QoS data. One active user invokes the same service many times, and QoS value is different each time. For example, one active user named Sam watches a movie on his mobile phone; the movie can be smooth one time but freeze the next time because of volatile QoS data. The above phenomenon is common in real life.
(2) Mobility of active users. An active user often moves around, and edge servers change according to the location of the active user [11] in mobile edge computing. Suppose that Sam often uses service from an old edge server. When using one video service on his mobile phone, its response time is 100 ms on average when the host server running the service is deployed in the old edge server. When Sam roams in a new edge server, if the video service remains invoked, traditional service recommendation approaches often monitor its historical QoS data in the old edge server, and obtained response time remains 100 ms. However, its real response time will be different because Sam is located change.
Based on research and experiments with existing service recommendation approaches such as [[40], [24], [15]], and [[8], [28]], we found that these approaches caused large errors in mobile edge computing because of user mobility. User mobility results in changing user locations and data volatility. These large errors are introduced in detail as follows:
- (1)
Mobility of user locations. In mobile edge computing, users invoke services by accessing different edge servers based on their dynamically changing locations. Because of user mobility, edge server handoff will be frequent [[30], [12]]. Therefore, history QoS data of users in the old edge server will likely be invalid when user the location changes significantly, and the QoS data of users in the new edge server are absent. Therefore, we should consider the mobility of users and it is important to learn how to predict the user QoS data from new edge servers.
- (2)
Volatility of mobile networks. Because of the volatility of mobile environment, if you use QoS data for invoking the same service one time, the QoS prediction value cannot reflect the real situation of the QoS. Therefore, QoS prediction values for services will cause larger errors based on one-time QoS data.
- (3)
Volatility of the same services at different invoked times. The QoS data for invoking the same services at a different time by one user are volatile. Calculating similarity between users based on the original QoS data is not reliable. If we do not preprocess the original QoS data, they will cause larger errors when calculating similarity between users.
Different from traditional service recommendation approaches, we first predict QoS values by reducing the influence of the above three factors. We then perform QoS prediction based on collaborative filtering and make service recommendations based on user mobility. Our approach was inspired by the following two cases; i.e., when users roam in a new edge server, if there are users in the new edge server and they invoke the service, then we can predict the QoS value based on their historical data. Otherwise, we use other user historical data from other edge servers at which they invoke the service. Finally, we conduct several experiments to verify our prediction accuracy based on the real-world environments.
The remainder of this paper is organized as follows: Section 2 shows our related work, and Section 3 introduces a motivation scenario. Section 4 presents our service recommendation approach based on user similarity and edge server similarity. Section 5 describes the implementation of our experiments and performance comparisons. Section 6 draws conclusion for our paper.
Section snippets
Related work
We have reviewed many Web service recommendation studies based on collaborative filtering algorithms, such as [[16], [15], [38], [7], [13]], and [32]. A few classic studies on the subject exist, including [[24], [39]]. For example, Shao et al. [24] proposed an approach based on collaborative filtering to perform similarity mining and make predictions for users based on their experiences. The approach contains two steps. First, they calculate the similarity between each two consumers with their
Motivation
Suppose that Sam often use one service on his mobile phone by accessing an edge server with response time (e.g., one QoS property) of 100 ms on average. As Fig. 1 depicts, Sam now travels to another edge server , and he want to use the same service. Then how to predict the QoS value of the service become an important issue by accessing the edge server . If the predicted value is less than 100 ms, this means Sam can still use the service; otherwise the service will be migrated to the edge
Our approach
Motivated by the above analysis, we propose an approach based on the CF algorithm to predict user QoS data by weakening the volatility of QoS data and considering the mobility of users. In our approach, based on the QoS data after normalization, we initially calculate user or edge server similarity. If the service invoked by an active user exists in the QoS data of new edge server, we calculate the similarity between users. If not, we should find other similar edge servers for the active user;
Experiments
In this section, we perform experiments to verify the performance of our approach and compare the results with other CF methods. Our experiments are intended to (1) validate the rationality of our proposed approach; (2) compare our approach with other CF methods; and (3) analyze parameters of our approach to achieve optimum performance.
Conclusion and future work
Different from traditional service recommendations base on QoS prediction, our approach considers user mobility and data volatility to adapt to mobile edge computing environments. Based on a real-world hybrid dataset, our experimental results show that prediction accuracy outperforms other approaches in mobile edge computing environments. In this paper, our approach initially calculates user or edge server similarity depending upon users’ changing locations, selects the Top-K most-similar
Acknowledgments
This work was supported in part by the National Science Foundation of China (Grant No. 61472047). The helpful suggestions from the anonymous reviewers are gratefully acknowledged, and I also offer thanks to Lubao Wang whose discussion and comment helped greatly.
Shangguang Wang is an associate professor at the State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications (BUPT). He received his Ph.D. degree at BUPT in 2011. He has co-authored more than 100 papers, and played a key role at many international conferences and workshops, such as General Chair and TPC Chair. His research interests include Service Computing and Cloud Computing. He is a Senior Member of the IEEE.
References (42)
- et al.
Multisensor data fusion: A review of the state-of-the-art
Inf. Fusion
(2013) - A. Ahmed, E. Ahmed, A survey on mobile edge computing, in: 10th International Conference on Intelligent Systems and...
- R. Arshad, H. ElSawy, S. Sorour, T.Y. Al-Naffouri, M.-S. Alouini, Handover management in dense cellular networks: A...
- et al.
Tracking and Data Fusion
(2011) - et al.
Pearson correlation coefficient
- et al.
Empirical analysis of predictive algorithms for collaborative filtering
A review of data fusion techniques
Sci. World J.
(2013)- X. Chen, X. Liu, Z. Huang, H. Sun, Regionknn: A scalable hybrid collaborative filtering algorithm for personalized web...
- et al.
Web service recommendation via exploiting location and QoS information
IEEE Trans. Parallel Distrib. Syst.
(2014) - et al.
A reality check of base station spatial distribution in mobile networks
An overview of handoff techniques in cellular networks
Int. J. Inf. Technol.
Latent semantic models for collaborative filtering
ACM Trans. Inf. Syst.
A highly accurate prediction algorithm for unknown web service qos values
IEEE Trans. Serv. Comput.
Support from the relationship of genetic and geographic distance in human populations for a serial founder effect originating in Africa
Proc. Natl. Acad. Sci. USA
Throughput prediction in cellular networks: experiments and preliminary results
Cited by (143)
Server placement in mobile cloud computing: A comprehensive survey for edge computing, fog computing and cloudlet
2024, Computer Science ReviewA distributed streaming framework for edge–cloud triangle counting in graph streams
2023, Knowledge-Based SystemsA federated collaborative recommendation model for privacy-preserving distributed recommender applications based on microservice framework
2023, Journal of Parallel and Distributed ComputingCitation Excerpt :Such centralized data processing manner results in availability and security issues [14] [31] [49]. For example, for real-time recommendation application, it is common that some of the distributed systems or devices malfunction and fail to deliver its local data to the centralized cloud server, making the recommendation service unavailable [52] [43]. Moreover, some owners of the distributed systems or devices may refused to upload the raw data to the centralized cloud server due to potential attackers leading to possible privacy data leakage [24] [36].
Towards machine-learning-driven effective mashup recommendations from big data in mobile networks and the Internet-of-Things
2023, Digital Communications and NetworksMobility-Aware and Privacy-Protecting QoS Optimization in Mobile Edge Networks
2024, IEEE Transactions on Mobile Computing
Shangguang Wang is an associate professor at the State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications (BUPT). He received his Ph.D. degree at BUPT in 2011. He has co-authored more than 100 papers, and played a key role at many international conferences and workshops, such as General Chair and TPC Chair. His research interests include Service Computing and Cloud Computing. He is a Senior Member of the IEEE.
Yali Zhao received bachelor’s degree in computer science and technology from Shandong University, in 2013. Currently, she is a Master Degree Candidate at the State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications. Her research interests include service computing and edge computing.
Lin Huang received the M.E. degree in computer science and technology from the Institute of Network Technology, Beijing University of Posts and Telecommunications, in 2012. Currently, she is a Ph.D. candidate at the State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications. Her research interests include Reputation measurement, Web service selection.
Jinliang Xu received the bachelor’s degree in electronic information science and technology from Beijing University of Posts and Telecommunications in 2014. Currently, he is a Ph.D. candidate in computer science at the State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications. His research interests include Service Computing, Information Retrieval, and Crowdsourcing.
Ching-Hsien Hsu is a professor in the department of information engineering and computer science at Feng Chia University, Taiwan. His research includes high performance computing, cloud computing, parallel and distributed systems, and ubiquitous/pervasive computing and intelligence. He has been involved in more than 100 conferences and workshops as various chairs and more than 200 conferences/workshops as a program committee member. He is the editor-in-chief of an international journal on Grid and High Performance Computing and has served on the editorial board for approximately 20 international journals.