An efficient and social-aware distributed in-network caching scheme in named data networks using matching theory
Introduction
While the Internet was originally designed to provide communications between two end machines, currently it is mainly exploited to massively diffuse content. This structural mismatch in design and usage profile makes the Internet an inefficient architecture in terms of resource utilization, users’ experienced latency and costs. A large range of solutions, e.g., Peer-to-Peer (P2P) networks [1] and Content Delivery Networks (CDNs) [2] were proposed, which both deploy distributed caches across the network. However, these solutions were not able to overcome the existing shortcomings due to the limited storage and high costs. Information-Centric-Networking (ICN) has been proposed as an alternative paradigm to the Internet [3]. In ICN, content objects are the main entity of the network and are identified by their names instead of their physical locations [4]. One of the promising architectures among different ICN architectures is Named Data Networking (NDN) [5]. A major characteristic of this novel architecture is the universal in-network caching [6]. Due to the finite NDN caching capacity and the infinite number of objects, caching scheme in this architecture is of great importance [7]. Caching solutions in NDN are based on caching along the delivery path referred to as on-path caching or caching in off-path nodes of the network namely off-path caching [8]. In on-path caching approach, caching decisions are only limited to the data objects and nodes along the data delivery path. On-path caching solutions fail to improve the cache space efficiency of the network as the total available cache space of the network is not considered as a whole. On the other hand, while off-path caching solutions are able to optimize the cache space efficiency of the network, they are not practical due to the need to aggregate great amount of information from the network and their centralized nature. While several existing works discussed caching solutions in NDN, most of them have concentrated on improving the bandwidth utilization and content retrieval latency of the network [9]. Thus in these works the aspect of cache space efficiency which determines how the total network-wide available cache space is utilized, is almost ignored. In our previous work [10], we focused on cache utilization efficiency in order to increase the hit ratio of the network. We proved that the outcome of the proposed distributed algorithm is stable and showed by simulation that the social welfare of our method reaches the optimal social welfare when the price step number is sufficiently small. In this paper we extend our social-aware distributed in-network caching scheme based on matching theory in NDN [11] and demonstrate its new properties. We analytically prove that the output of our proposed method is in Competitive Equilibrium which is an extension of Nash Equilibrium in game theory. The main concentration of this paper is on the cost of retrieving data objects for ISPs as it is a crucial practical challenge. We prove that our proposed algorithm minimizes the Internet Service Providers’ (ISPs) costs of retrieving data objects from outside the network. The proposed matching algorithm is based on many to one matching where each node matches a number of objects proportional to its caching capacity and each object matches at most one node. Thus, many to one matching guarantees that there exists at most one copy of each object inside the network while the caching capacity of the network is optimally utilized. Our contributions are summarized as follows:
- •
We formulate the distributed in-network caching scheme as a many-to-one objects to node matching game in order to maximize the social welfare of the network. We are the first to propose a principled solution based on matching theory to address the caching problem in NDN.
- •
In the proposed distributed caching scheme, each node needs locally available information to make caching decisions considering object’s local popularity.
- •
It is proved that the proposed algorithm is stable and the outcome is in competitive equilibrium. Further, we show analytically that the social welfare of our proposed algorithm converges to the optimal centralized approach after a finite number of iterations with a significantly lower overhead.
- •
We prove that the cost of retrieving required objects for an ISP is the minimum possible amount, which is beneficial for an ISP in order to reduce its costs.
- •
Simulation results show that our distributed proposed scheme converges to the optimal centralized solution, outperforming the default caching approach in NDN in terms of both network’s social welfare and hit ratio.
The rest of this paper is organized as follows. First, the related work is discussed in Section 2. In Section 3, the system model is described. The problem of objects to nodes allocation is formulated in Section 4. In Section 5, we propose our many-to-one objects to nodes matching algorithm to optimize the social welfare of the network considering the utility functions of the players. Analysis of the proposed caching scheme is presented in Section 6. Moreover, simulation results are depicted in this section. Finally, the paper is concluded in Section 7.
Section snippets
Related work
Off-path caching solutions in NDN are similar to the content replication problem in CDN [12] and web cache placement [13]. A centralized off-path caching solution to the content replication problem in NDN is studied in [14] which allocates the most global popular objects to the nodes with the lowest access cost in order to reduce the inter-domain and intra-domain traffic load. On the other hand, a number of probabilistic on-path caching schemes are proposed in [15], [16]. In the Random Caching
Scenario description
A multi-ISP cache-enabled network is considered, comprising of K nodes inside an ISP, denoted by the set . Each Nodej has a limited cache space of Cj in order to keep popular objects, which is referred to as the quota of the node. Furthermore, there are M objects each with a different size, denoted by with sizes located inside the content provider. We denote the index set of nodes by and the index set of objects by . Note that at the beginning,
Matching function definitions
Before formulating the optimization problem, the definitions and notations of many-to-one matching are first introduced as follows: Definition 1 Given two disjoint finite sets of players: and and the index sets and definition of the matching function is as follows:Where denotes the quota of Γ members and
Proposed matching algorithm
This section elaborates on the many-to-one matching algorithm to solve the optimization problem (10) and acquire the optimal allocation matrix in a distributed fashion. To proceed with designing a matching algorithm for solving the problem where objects and nodes maximize their utility functions, we focus on the preference lists of the players. Utility function of a node increases with the popularity to size ratio of objects that are cached inside the node and decreases with the amount of money
Stability
Definition 2 A one-to-one matching Ψ is defined as stable if it is not blocked by any individual or pair. Consider and where . The matching Ψ is blocked by an individual, if or prefers not to be matched, than being matched with its current partner under the matching function Ψ. For this implies that and for this implies that . The matching Ψ is blocked by a pair if (1) it is
Conclusion
The object allocation problem in NDN to enhance the cache utilization efficiency of the network was investigated. We formulated the object allocation problem as a distributed many-to-one matching game in order to maximize the social welfare of the network. We proved that the produced matching is stable and the outcome of the proposed algorithm is in competitive equilibrium. We also analytically showed that the proposed algorithm converges to the centralized optimal approach for sufficiently
Conflict of interest
None.
Mahsa Ehsanpour received the B.Sc. degree in Computer Engineering-Software in 2015 from Khajeh Nasir Toosi University of Technology, Tehran, Iran and the M.Sc. degree in Information Technology-Computer Networks in 2017 from Sharif University of Technology, Tehran, Iran. She is currently a Ph.D. candidate at the University of Adelaide, Australia. Her research interests include Named Data Networks, Computer Vision, Machine Learning and Deep Learning.
References (35)
- et al.
A distributed in-network caching scheme for P2P-like content chunk delivery
Comput. Netw.
(2015) - et al.
Data popularity measurements in distributed systems: survey and design directions
J. Netw. Comput. Appl.
(2016) - et al.
Performance analysis of in-network caching for content-centric networking
Comput. Netw.
(2013) - et al.
Efficient cache availability management in information-centric networks
Comput. Netw.
(2015) - et al.
Survey on caching approaches in information centric networking
J. Netw. Comput. Appl.
(2015) - et al.
Joint forwarding and caching with latency awareness in information-centric networking
Comput. Netw.
(2016) - et al.
Two-sided Matching
Handbook of Game Theory with Economic Applications
(1992) - et al.
Storage planning and replica assignment in content-centric publish/subscribe networks
Comput. Netw.
(2011) - et al.
The LCD interconnection of LRU caches and its analysis
Performance Eval.
(2006) - et al.
Greedy caching: an optimized content placement strategy for information-centric networks
Comput. Netw.
(2018)
An energy-efficient distributed in-network caching scheme for green content-centric networks
Comput. Netw.
A survey of information-centric networking
IEEE Commun. Mag.
Naming in content-oriented architectures
Proceedings of the 11th ACM SIGCOMM Workshop on Information-centric Networking
Named data networking (NDN) project
PARC, Tech. Rep. NDN-0001
On efficient and social-aware object allocation in named data networks using matching theory
Proceedings of IEEE Symposium on Computers and Communications
Distributed caching algorithms for content distribution networks
Proceedings of IEEE International Conference on Computer Communications
A novel caching scheme for the backbone of named data networking
Proceedings of IEEE International Conference on Communications
Cited by (6)
An efficient adaptive cache management scheme for named data networks
2023, Future Generation Computer SystemsDesign of Multi-Armed Bandit-Based Routing for in-Network Caching
2023, IEEE AccessA two-sided matching model for data stream processing in the cloud - Fog continuum
2021, Proceedings - 21st IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing, CCGrid 2021Popularity-Aware In-Network Caching for Edge Named Data Network
2021, Wireless Communications and Mobile ComputingImproved Merkle Hash Tree-Based One-Time Signature Scheme for Capability-Enhanced Security Enforcing Architecture for Named Data Networking
2020, Wireless Personal Communications
Mahsa Ehsanpour received the B.Sc. degree in Computer Engineering-Software in 2015 from Khajeh Nasir Toosi University of Technology, Tehran, Iran and the M.Sc. degree in Information Technology-Computer Networks in 2017 from Sharif University of Technology, Tehran, Iran. She is currently a Ph.D. candidate at the University of Adelaide, Australia. Her research interests include Named Data Networks, Computer Vision, Machine Learning and Deep Learning.
Siavash Bayat (S’10â;;M’13) received the B.Sc. and M.Sc. degrees in electrical engineering in 2002 and 2005, respectively, and the Ph.D. degree in electrical engineering from the University of Sydney, Australia, in 2013. He was then a Postdoctoral Research Fellow at the University of Sydney, prior to joining the Sharif University of Technology in 2014, where he is currently an Assistant Professor. From 2007 to 2009, he was affiliated with Sharif University of Technology, where he held a position of faculty member with responsibility for the research in wireless communication networks system design. His research interests include wireless resource management, wireless communications, Internet of things, signal processing, heterogeneous networks, cognitive radio, game theory, and physical layer security.
Ali Mohammad Afshin Hemmatyar received B.Sc., M.Sc. and Ph.D. degrees in Electrical Engineering from Sharif University of Technology, Tehran, Iran in 1988, 1991 and 2007, respectively. Since 1991, he has joined Department of Computer Engineering in Sharif University of Technology, where he is currently an assistant professor. His research interests are Vehicular Ad-hoc Networks, Wireless Sensor Networks, Cognitive Radio Networks, and Social Networks.