CABaRet: Leveraging Recommendation Systems for Mobile Edge Caching
07 Aug 2018-pp 19-24
TL;DR: The results show that significant caching gains can be achieved in practice; 8 to 10 times increase in the cache hit ratio from cache-aware recommendations, and an extra 2 times increase from caching optimization.
Abstract: Joint caching and recommendation has been recently proposed for increasing the efficiency of mobile edge caching. While previous works assume collaboration between mobile network operators and content providers (who control the recommendation systems), this might be challenging in today's economic ecosystem, with existing protocols and architectures. In this paper, we propose an approach that enables cache-aware recommendations without requiring a network and content provider collaboration. We leverage information provided publicly by the recommendation system, and build a system that provides cache-friendly and high-quality recommendations. We apply our approach to the YouTube service, and conduct measurements on YouTube video recommendations and experiments with video requests, to evaluate the potential gains in the cache hit ratio. Finally, we analytically study the problem of caching optimization under our approach. Our results show that significant caching gains can be achieved in practice; 8 to 10 times increase in the cache hit ratio from cache-aware recommendations, and an extra 2 times increase from caching optimization.
Citations
More filters
TL;DR: This work quantitatively investigates how recommendation can be applied to enhance the caching efficiency of D2D enabled wireless content caching networks and shows that the optimal recommendation and caching policies which jointly maximize the cache efficiency is NP-hard to compute.
Abstract: Edge side caching assisted device-to-device (D2D) communication has been acknowledged as a promising technique to alleviate the heavy burden of backhaul transmission link and to reduce the network latency. However, the effectiveness of caching strategies at the network edge is highly dependent on the distribution of individual user’s content preference. To fully attain the benefits of edge caching, some proactive mechanisms shall be considered. Among which, recommendation performs noticeably well due to its capability of reshaping the content request probabilities of different users, which in turn affects the cache decision significantly. In this work, we quantitatively investigate how recommendation can be applied to enhance the caching efficiency of D2D enabled wireless content caching networks. And for that, the cache hit ratio maximization problem for a generic network model is formulated taking into account the requirements of each user’s personalized recommendation quality, recommendation quantity and cache capacity. Then, we show that the optimal recommendation and caching policies which jointly maximize the cache efficiency is NP-hard to compute. Further, a time-efficient sub-optimal algorithm is designed, which works in an iterative manner and has provable convergence guarantee as well as polynomial time complexity. Monte-Carlo simulation results demonstrate the convergence performance of our proposed joint decision algorithm and its cache efficiency improvements compared to extensive benchmarks.
31 citations
25 Jan 2019
TL;DR: This paper establishes a framework for the joint user association, content caching and recommendations problem, and proposes a heuristic that tackles the joint problem when the objective is to maximize the total hit ratio over all caches.
Abstract: In this paper, we investigate the performance gains that are achievable when jointly controlling (i) in which Small-cell Base Stations (SBSs) mobile users are associated to, (ii) which content items are stored at SBS co-located caches and (iii) which content items are recommended to the mobile users who are associated to different SBSs. We first establish a framework for the joint user association, content caching and recommendations problem, by specifying a set of necessary conditions for all three component functions of the system. Then, we provide a concrete formulation of the joint problem when the objective is to maximize the total hit ratio over all caches. We analyze the problems that emerge as special cases of the joint problem, when one of the three functions is carried out independently, and use them to characterize its complexity. Finally, we propose a heuristic that tackles the joint problem. Proof-of-concept simulations demonstrate that even this simple heuristic outperforms an optimal algorithm that takes only caching and recommendation decisions into account and provide evidence of the achievable performance gains when decisions over all three functions are jointly optimized.
21 citations
TL;DR: JPR schemes are presented for multiuser multiple-input single-output (MISO) systems, in which content items are pushed over MISO downlinks with multicast beamforming with results that show that presented JPR policies are capable of attaining significant effective throughput gains.
Abstract: Proactive caching is recognized as a promising approach to handle the rapid growth of data traffic, thereby attracting much attention recently. As a key performance metric of caching, the hit ratio is determined by demand probabilities of users for content items and caching decisions. Because the recommendation system is capable of shaping user demands, the joint caching and recommendation holds the potential of improving the hit ratio substantially. In this paper, joint pushing and recommendation (JPR) schemes are presented for multiuser multiple-input single-output (MISO) systems, in which content items are pushed over MISO downlinks with multicast beamforming. Aiming at maximizing the effective throughput, we formulate a multi-stage stochastic programming problem under the constraints of transmit power and quality of experience (QoE). Since the formulated problem is intractable, suboptimal online JPR policies are presented based on the convex–concave procedure and branch-and-bound methods. Simulations show that presented JPR policies are capable of attaining significant effective throughput gains.
14 citations
Cites background from "CABaRet: Leveraging Recommendation ..."
...In [26], Kastanakis and Sermpezis proposed a joint caching and recommendation approach without tight cooperation between network operations and content providers....
[...]
TL;DR: In this paper, an intelligent caching strategy for MEC based on machine learning has been proposed, which considers the user's mobility and interest preferences, and it could predict user mobility using historical trajectory based on LSTM algorithm, and predict interest using Gradient Boosting Decision Tree (GBDT) method, to obtain the content of interest in advance, and then cache the content in advance on the neighboring edge node where the user is likely to go.
Abstract: With the rapid development of 5G mobile networks, data traffic has increased dramatically and is putting tremendous pressure on the backhaul link. In 5G-based mobile edge computing (MEC) environment, efficient caching at the edge of the network provides a solution for satisfying the quality of experience (QoE) requirements for lower latency. An intelligent caching strategy for MEC based on machine learning has been proposed, namely EICache, which considers the user’s mobility and interest preferences. It could predict user’s mobility using historical trajectory based on Long Short-Term Memory (LSTM) algorithm, and predict interest using Gradient Boosting Decision Tree (GBDT) method, to obtain the content of interest in advance, and then cache the content in advance on the neighboring edge node where the user is likely to go. Performance evaluations have been conducted using public YouTube trending video datasets from Kaggle and real trajectory datasets, compared with different cache replacement methods. The metrics of the cache hit rate, and the overall request latency are used for evaluation. By training the datasets first and then predicting, the accuracy of LSTM-based location prediction is about 80%, and the accuracy of GBDT-based interest prediction reaches about 25.4%. The hit rate of the edge caching strategy is increased by 40.5% compared with the strategy of random caching without any predictions. The results have proved the efficiency of EICache, which could meet the user’s QoE requirements of low request latency.
6 citations
10 May 2021
TL;DR: In this paper, the authors propose a Markov Decision Process (MDP) formulation to model a user with random session length and provide flexibility to model users who are reactive to the quality of the received recommendations.
Abstract: Caching content over CDNs or at the network edge has been solidified as a means to improve network cost and offer better streaming experience to users. Furthermore, nudging the users towards low-cost content has recently gained momentum as a strategy to boost network performance. We focus on the problem of optimal policy design for Network Friendly Recommendations (NFR). We depart from recent modeling attempts, and propose a Markov Decision Process (MDP) formulation. MDPs offer a unified framework that can model a user with random session length. As it turns out, many state-of-the-art approaches can be cast as subcases of our MDP formulation. Moreover, the approach offers flexibility to model users who are reactive to the quality of the received recommendations. In terms of performance, for users consuming an arbitrary number of contents in sequence, we show theoretically and using extensive validation over real traces that the MDP approach outperforms myopic algorithms both in session cost as well as in offered recommendation quality. Finally, even compared to optimal state-of-art algorithms targeting specific subcases, our MDP framework is significantly more efficient, speeding the execution time by a factor of 10, and enjoying better scaling with the content catalog and recommendation batch sizes.
4 citations
References
More filters
26 Mar 2018
TL;DR: This work uses traceroute-based measurement methods to infer techniques for assigning YouTube requests to specific Google video content caches, including the interconnection links between the access providers and Google, and finds that the selection of video cache has little impact on BGP selection of interdomain links.
Abstract: For the last decade, YouTube has consistently been a dominant source of traffic on the Internet. To improve the quality of experience (QoE) for YouTube users, broadband access providers and Google apply techniques to load balance the extraordinary volume of web requests and traffic. We use traceroute-based measurement methods to infer these techniques for assigning YouTube requests to specific Google video content caches, including the interconnection links between the access providers and Google. We then use a year of measurements (mid-2016 to mid-2017) collected from SamKnows probes hosted by broadband customers spanning a major ISP in the U.S. and three ISPs in Europe. We investigate two possible causes of different interdomain link usage behavior. We also compare the YouTube video cache hostnames and IPs observed by the probes, and find that the selection of video cache has little impact on BGP selection of interdomain links.
18 citations
"CABaRet: Leveraging Recommendation ..." refers background in this paper
...Finally, research on the performance of the YouTubeCDN in relation to network parameters [1, 9, 12], indicates that our approach can be relevant to wireline networks as well....
[...]
..., taking into account routing policies [1, 12], inter-domain load balancing [12], cache locations [1]) can be used to improve the user QoE (e....
[...]
21 May 2017
TL;DR: This paper shows that servers can reuse encrypted content between sessions, thereby rejuvenating caching and provides an extension that prevents client linkability, i.e., ensuring it is impossible to tell if two clients are viewing the same content.
Abstract: End-to-end encryption seemingly signifies the death of caching, because current methods ensure that no two sessions are alike. In this paper, we show that servers can reuse encrypted content between sessions, thereby rejuvenating caching. The main idea of our technique is to allow interim nodes to cache content based on pseudo-identifiers instead of real file identities. This enables caching of reusable pseudo-identifiers, whilst maintaining content confidentiality, i.e., ensuring that only the client and the server know the actual identity of the requested file. Furthermore, we provide an extension that prevents client linkability, i.e., ensuring it is impossible to tell if two clients are viewing the same content. Finally, we formally analyse the balance between security and the hit probability performance of the cache.
17 citations
"CABaRet: Leveraging Recommendation ..." refers background in this paper
..., https) and do not typically share user-related information [11]....
[...]
Posted Content•
TL;DR: In this paper, the authors proposed a Markovian model for recommendation-driven user requests and formulated the problem of biasing the recommendation algorithm to minimize access cost, while maintaining acceptable recommendation quality.
Abstract: Caching has been successfully applied in wired networks, in the context of Content Distribution Networks (CDNs), and is quickly gaining ground for wireless systems. Storing popular content at the edge of the network (e.g. at small cells) is seen as a `win-win' for both the user (reduced access latency) and the operator (reduced load on the transport network and core servers). Nevertheless, the much smaller size of such edge caches, and the volatility of user preferences suggest that standard caching methods do not suffice in this context. What is more, simple popularity-based models commonly used (e.g. IRM) are becoming outdated, as users often consume multiple contents in sequence (e.g. YouTube, Spotify), and this consumption is driven by recommendation systems. The latter presents a great opportunity to bias the recommender to minimize content access cost (e.g. maximizing cache hit rates). To this end, in this paper we first propose a Markovian model for recommendation-driven user requests. We then formulate the problem of biasing the recommendation algorithm to minimize access cost, while maintaining acceptable recommendation quality. We show that the problem is non-convex, and propose an iterative ADMM-based algorithm that outperforms existing schemes, and shows significant potential for performance improvement on real content datasets.
5 citations