Offloading lte data traffic with named-Data networking integration - Ong Mau Dung

Journal of Science and Technology 54 (3A) (2016) 12-22 OFFLOADING LTE DATA TRAFFIC WITH NAMED-DATA NETWORKING INTEGRATION Ong Mau Dung Industrial University of Ho Chi Minh City, 12 Nguyen Van Bao Street, Ward 4, Go Vap District, Ho Chi Minh City Email: ongmaudung@iuh.edu.vn Received: 14 Mar 2016; Accepted for publication: 26 July 2016 ABSTRACT Nowadays, mobile Internet becomes increasingly popular and the number of mobile users is growing exponentially. With a traditional clients-server connection, servers are usually in overload state by a huge number of users accessing the service at the same time. Moreover, in a century of “green” Internet technology, it should be more effective in content distribution for a huge number of users. Named-Data Networking (NDN) had been proposed as the promising solution for above problem, which is a content name-oriented approach to disseminate content to edge gateways/routers. In NDN, popular contents are cached at routers for a certain time, and the previously queried one can be reused for multiple times to save bandwidth. In this paper, we propose a solution for Long Term Evolution (LTE) mobile network based on the concept of NDN. By OPNET Modeler simulation, we carry out the evaluation in realistic mobile network with a huge number of mobility LTE mobile stations access to a single server, where content names and content sizes are obtained from real trace of Internet traffic. The obtained results show that Evolved Packet Core (EPC) caching scheme can helps to further increases the quality of service as well as offloads server traffic significantly. Keywords: LTE mobile network, Pareto distribution, caching, data traffic offloading. 1. INTRODUCTION While the requirements of rich multimedia contents using smart phones and tablets continue increasing over time [1], the current capacity mobile backhaul networks as well as the bandwidth of Internet have a lot of challenges to cope with the practical growing traffic due to the centralized architecture [2]. Related with the exponent growth of mobile traffic, a skewness of popularity content characteristic was found. That is to say, a few most popularity contents are often queried by the huge number of end users [3, 4]. Therefore, it is critical to design an effective way to minimize the duplicate contents transmission to save bandwidth and offloading server traffic. In order to alleviate this problem, named data networking (NDN) is proposed to effectively distribute popular data content to a huge number of users [5]. Furthermore, reducing traffic load Offloading LTE data traffic with Named-Data Networking Integration 13 by in-network caching can enhance the mobile network with higher energy efficiency and toward the evolution of “green” mobile network. Figure 1. EPC caching scheme. Figure 1 illustrates the caching of popular contents at Evolved Packet Core (EPC) in Long Term Evolution (LTE) network. For instance, four LTE mobile stations (MSs) request the same content from Service Provider (SP). Without in-network caching, SP must send four copies redundantly via the Internet and LTE network. If we deploy EPC caching, the traffic from SP to EPC is reduced by facilitating the requests from MSs to be satisfied with high probability. As shown in Fig. 1, the mobile traffic often encounters with the issue of bandwidth bottleneck, and it may take a long duration to complete transmits content through the Internet. By caching contents at EPC, the bandwidth bottleneck is reduced, end-to-end delay is also decreased, and server traffic is offloaded. The traffic from EPC to MSs is not reduced. However, contrary to the best-effort characteristic of the Internet, operation and maintenance of the mobile backhaul network are deployed by mobile network operator (MNO), which guarantees Quality of Service (QoS) and Quality of Experience (QoE) for MSs [6]. From the above discussion, we can observe that caching decision and replacement policies are key points to determine the performance of NDN. Least Recently Used (LRU) and Least Frequently Used (LFU) replacement policies are suggested in original NDN. Motivated by above background, in this work, we propose a novel caching policy, dubbed Fine-Grained Popularity-based Caching (FGPC) [7]. The different between using FGPC in this paper and in [7] as follows. In [7], we assume users request contents from a server following a Pareto distribution: 80 % traffic request popular contents while 20 % remaining traffic request unpopular items. Furthermore, we consider that all contents have the same size fixed at the beginning of the simulation, and a popularity threshold is five. However, in the real-world implementation, the popular content distribution is more spreading than Pareto distribution, different contents should have different sizes, and undesirable popularity threshold may cause ineffective caching. For this reason, we use Internet data trace file to analyze FGPC performance in this paper. We also analyze the trace file to consider skewness popular contents, and then pre- determine the popularity threshold value. From the background of NDN, we import NDN strategy to all network elements on top of Internet Protocol (IP) layer in OPNET simulator (www.opnet.com). The existing LRU, LFU and our proposal FGPC policies have been successful constructed in NDN nodes. Our simulation results prove that NDN is a good revolution for existing challenges of traditional IP network. And FGPC outperforms than LRU, LFU with highly effective caching. Ong Mau Dung 14 Table 1. Used notations. Symbol Definition |F| Total number of file iF Size of the i th file 1 F i i F Catalog size sizeC Cache size 1 .100%R sizesize F i i C C F Relative cache size C-seg Sequence of contents (or requested file) For the sake of better readability, Table 1 lists up the notation used in this paper. The remainder of this paper is structured as follows. Section 2 highlights some recent research work pertaining to in-network caching and mobile traffic offloading techniques. In Section 3, we present the LTE interworking with NDN mechanism. Section 4 describes our network architecture simulation and verifies simulation results. Finally, the paper concludes in Section 5. 2. RELATED WORKS Recently, NDN becomes a hot topic, and lots of projects and prototypes have been applied NDN [8, 9]. Similar with content-centric networking (CCN), NDN came to play to organize such an efficient streaming based on smart caching of popular content nearby the requesting users. A typical NDN framework was proposed in [5], which presents a simple and effective communication model. In-network caching is essential and important in NDN [10]. The highest benefits of NDN is found by caching at edge routers/gateways and fast decrease when far beyond the traditional edge caching [11]. However, global NDN caching and collaborative between caches can offer benefits well beyond only edge caching in sub network [12]. Variety cache sizes and popularity skewness are often considered to have an effect on performance of different replacement policies. In [13, 14], the simulation results show that either cache size or the skewness factor increases, the performance of network improves but nonlinear growth with the factors. Content popularity-based smart caching is also considered in 5G systems; placing popular content nearby the Radio Access Network (RAN) of mobile networks [15]. Last but not least, NDN architecture and offloading mobile network traffic by in-network caching were discussed in previous works of [2, 15]. Different from them, in this paper, the NDN and LTE are newly designed in OPNET, and real trace Internet traffic is utilized in the simulation. In the next section, we introduce the network topology of LTE with NDN caching and summary the FGPC replacement policy. Offloading LTE data traffic with Named-Data Networking Integration 15 3. INTEGRATING NDN WITH LTE We now focus on how NDN can be integrated with LTE to further improve mobile network performance achieving a further efficient utilization of Internet resources. Figure 2. Interworking of LTE with NDN caching. Figure 2 illustrates the interworking of LTE with NDN caching. Roots name element is called globally-routable name broadcasted by server [5]. When NDN nodes receive the roots name, they add to the Forwarding Information Base (FIB) and forward roots name to their neighbors. Server will up-to-date the roots name by sending in periodically. EPC in the LTE model is slightly modified to support interworking with NDN mechanism. Any Interest packet (IntPk) and Data packet (DataPk) are come in and come out via EPC and NDN node. Figure 3. Interest packet flow and Data packet flow inside the NDN node. In general, there are two flows of packet: the IntPk flow from LTE MSs to server and the DataPk from server to MSs as shown in Fig. 3. When NDN node receives the IntPk, it lookups for data content in Content Store (CS). DataPk is replied if found content, otherwise IntPk is checked in Pending Interest Table (PIT). The PIT keeps track on list of users for unsatisfied IntPk. The FIB is a table of outbound faces for IntPks, and unsatisfied IntPks are forwarded up- stream(s) toward potential content sources. Then returned DataPk will be sent to down-stream(s) and CS keeps received content for the next responding. With the “best effort” service provided by Internet, all entries in the PIT must maintain timeout rather than being hold indefinitely. To maximize the probability of sharing Ong Mau Dung 16 with minimal upstream bandwidth demand and lowest downstream latency, CS should keep all arrived DataPks as long as possible. To prevent buffer from overflowing, CS maintains a timer to count “lifetime” for each content and using replacement policy rather than holds it permanently. In order to fully exhibit the potential performance gains of NDN, an efficient replacement policy for CS should be designed. The FGPC replacement policy is used as it outperforms LRU, LFU and ensures highly effective caching. In FGPC, each NDN node adds one more popularity table to maintain statistic information in terms of content name, content counter and time stamp. Based on a historical table, FGPC keeps track of popular contents by counting locally the frequency of appearance of each content name. Figure 3 is presented for how to integrate FGPC into NDN algorithm. Through FGPC policy, it requires more resources such as memory and computation than LRU and LFU. For more detail on the FGPC policy, please refer to [7]. 4. SIMULATION AND RESULTS 4.1. Network architecture We implement NDN protocol stack and perform simulation using OPNET Modeler 16.0. In our scenario, NDN is overlaid over IP layer. Indeed, NDN processing module is integrated into all network elements such as router/gateway, EPC, and modified other network elements, such as LTE MSs, server and IP Cloud. To the best of our knowledge, this NDN simulation model is the most realistic one based on OPNET simulator. Figure 4. Network topology utilized in the simulation. Offloading LTE data traffic with Named-Data Networking Integration 17 By generalizing LTE network topology [6, 16], we use a topology showing in Fig. 4. The simulated network includes three LTE cells and each cell is handled by eNodeB with ten MSs. Totally thirty MSs move and handover around three eNodeBs with random waypoints and random speeds configuration. The BW bottleneck of the Internet backbone is deployed by given random latency value once a packet passed through the Internet backbone. In Fig. 4, popular content are cached at edge gateways as known as EPC. It should be noted that, the caching is performed at the data object level, e.g. Data packet (DataPk). In general, all NDN routers/gateway located along the reverse path from server to MSs can cache data objects. However, a lot of papers showed that the highest benefits of NDN is achieved at the edge routers/gateways and fast decrease when far beyond the traditional edge caching [12, 13]. Intelligent caching at edge routers and gateways is of vital importance for the efficiency of NDN [9]. For this reason, we only focus on EPC caching. However, global CCN caching, supported with Self-Organized Networking (SON) functions among caches can offer benefits well beyond only edge caching in sub-networks [14]. In order to ensure the realistic of our simulation, we download the Internet traffic trace file from an ircache.net website. Content name and content size are two fields in a data format used for the simulation. We use 400000 lines of requested content from the trace file. The trace file is imported to both clients (MSs) and server. Then clients obtain name of contents by sequence reading from top to down until end of the trace file. Server plays a role that reply contents once it received IntPks from clients (or NDN nodes). Before running the simulation, we analyze the Internet traffic trace file by using C language programming to determine the skewness popularity level of contents. Table 2. The skewness popularity content of trace file. Popularity Level Counter Number of Contents Ratio % Number of Requests Ratio % 1 112081 61.71 112081 28.02 2 40549 22.33 81098 20.27 3 15815 8.71 47445 11.86 4 6142 3.38 24568 6.14 5 2456 1.35 12280 3.07 6 1240 0.68 7440 1.86 7 620 0.34 4340 1.09 8 458 0.25 3664 0.92 9 290 0.16 2610 0.65 10i i 1979 1.09 104474 26.12 SUM 181630 100 400000 100 Table 2 shows out the popularity levels of contents with increasing arrangement. As shown in Table 2, the number of contents and the ratio decrease when the popularity level counter increases. Assume content is called popular content when a counter is greater than or equal three, then Table 2 indicates that 15.96 % popular contents are demanded by 51.71 % requests. Ong Mau Dung 18 From this observation, the hitting rate of NDN node will be high when the cache recognizes and stores 15.96 % popular contents. For this reason, the threshold value for FGPC policy is three for the simulation. It should be noted that the FGPC is not described in detail in this paper because of limited space. For more information about FGPC policy, please refer to [7]. Table 3 sums up important parameters for the simulation. Table 3. Sum up important parameters for the simulation. Element Attribute Value Server Total number of files 181630 files Catalog size 7119.338 MB Link 1000 BaseX MS Start time 100 + random(10) seconds Stop time End of the trace file IntPk inter-arrival time 1 second DataPk time-out 2 second IntPk packet size 1 Kb DataPk packet size 1 Mb Number of requested contents 400000 Mobility configuration 3 km 2 with random waypoints and speeds NDN node Cache size (Csize) 50/ 100/ 200/ 300/ 400/ 500/ 600/ 700 and 800 MB Relative cache size ( R sizeC ) 0.7/ 1.4/ 2.8/ 4.2/ 5.6/ 7/ 8.4/ 9.8 and 11.2% Replacement policy LFU/ LRU/ FGPC The popularity threshold (Pth) 3 Link 1000 BaseX Internet BW bottleneck Latency per packet Normal(0.05, 0.02) distribution 4.2. Results Analysis 4.2.1 Impact of replacement policy on hitting rate Figure 5 compares the performance of different replacement policies when we fix the relative cache size at 11.2 %. As shown in Fig. 5, FGPC has better performance with the highest hitting ratio, following by LRU and LFU, respectively. Figure 5 also presents for characteristics of replacement policies as follows. (i) LRU (LFU): A cache always stores new coming content. Thus, hitting rate quickly increases when the cache is filled from zero to full. After the cache was Offloading LTE data traffic with Named-Data Networking Integration 19 overflowed, drawback of LRU (LFU) happens and causes the hitting rate increasing slowly. (ii) FGPC: Similarly, FGPC always stores new content when it has available memory, hitting rate of FGPC quickly increases too. After the cache was overflowed, FGPC has enough information about popularity level of contents. Thus, hitting rate continues going up to the final state. Figure 5. Impact of C-Seg on hitting rate with R sizeC equal to 11.2 %. Figure 6. Impact of R sizeC on final state hitting rate by LRU, LFU and FGPC. Ong Mau Dung 20 Figure 6 gives us an overview about final states of multiple replacement policies with multiple relative cache sizes. FGPC always outperforms with higher hitting rate than LRU and LFU in all situations. In Fig. 6, the relative cache size ( R sizeC ) at NDN node is increased in turn at 0.7%, 1.4%, 2.8%, 4.2%, 5.6%, 7%, 8.4%, 9.8% and 11.2%, respectively. Simulation results of all the replacement policies show that, when the cache volume increases, higher hitting rate can be achieved. However, a growth of hitting rate does not linear increase with the cache volume. As shown in Fig. 6, when R sizeC is 8.4%, the cache can handle all most of popular contents. If we continue to enlarge R sizeC up to 11.2%, the algorithm performance gain decreases. Thus, there is a tradeoff between cache volume (cost) and algorithm performance, and we need to make a balance by choosing a suitable cache size. 4.2.2 Average end-to-end delay at MSs Figure 7 shows the average end-to-end delay for a second unit of time. When the scheme of “Without Cache” is employed, all DataPks are affected by the Internet BW bottleneck and the delay is always staying around 0.226 s. With caching schemes, the effective BW bottleneck is reduced. The higher hitting rate, the lower delay for MSs. For this reason, the end-to-end delay is reduced step by step from 0.162 s to below 0.1 s, and FGPC helps MSs to achieve lower delay than LRU (LFU). Figure 7. Average end-to-end delay at MSs. 4.2.3 Impact of caching on the server load In NDN strategy, contents are cached in NDN nodes located near end users. Instead of sending all the requests to the origin sever, NDN nodes act as surrogates to original server and the cached contents are responded to end users, then limiting the backbone traffic. If the NDN node performs with higher hitting ratio, lower requested traffic is fetched to server, and then higher percent offloading traffic is achieved. Offloading LTE data traffic with Named-Data Networking Integration 21 Figure 8. Accumulated size of data sent by server. Figure 8 shows the amount of byte responding by server under various caching schemes. Without caching, all requests are fetched to server which poses high redundancy content replied by server. With EPC caching, a number of IntPks are satisfied at EPC node, which reduces the request bit rate received by server significantly. 5. CONCLUSION In this paper, we discuss a new framework that not only offloads Internet mobile traffic but also increases QoS by proposing a new type of cache decision and replacement policy FGPC. In our algorithm, it engages NDN nodes with higher hitting rate by effective caching and increases offloading server significantly. Furthermore, we carry out simulation reflecting the real world implementation by using the real Internet traffic trace file, and the results validate the effective and efficiency of our proposed scheme. In the near future, we will publish the NDN simulation module to the OPNET community. REFERENCES 1. Cisco. - Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 2014-2019, white paper, (2015) 1-42. 2. Yousaf F. Z., Liebsch M., Maeder A., et al. - Mobile CDN Enhancements for QoE- Improved Content Delivery in Mobile Operator Networks, IEEE Network 27 (2) (2013) 14-21. 3. Cha M., Kwak H., Rodriguez P., et al. - Analyzing the Video Popularity Characteristics of Large-Scale User Generated Content Systems, IEEE/ACM Transaction on Networking 17 (5) (2009) 1357-1370. Ong Mau Dung 22 4. Li G., Wang M., Feng J., et al. - Understanding User Generated Content Characteristics: A Hot-Event Perspective, IEEE Communications Conference (IEEE ICC2011), Kyoto, (2011) 1-5. 5. Jacobson V., Smetters D., Thornton J., et al. - Networking named content, Communication of the ACM 55 (1) (2012) 117-124. 6. Alcatel Lucent. - Introduction to Evolved Packet Core, strategic white paper, (2009) 1-7. 7. Mau D.O., Chen M., Taleb T., et al. - FGPC: Fine-Grained Popularity-based Caching Design for Content Centric Networking, Procedure ACM MSWIM'14, USA, (2014) 295- 302. 8. Han B., Wang X., Kwon T., et al. - AMVS-NDN : Adaptive Mobile Video Streaming with Offloading and Sharing in Wireless Named Data Networking, IEEE INFOCOM, (2013) 375 - 380. 9. Amadeo M., Campolo C., and Molinaro A. - Enhancing content-centric networking for vehicular environments, Journal of Computer Networks 57 (10) (2013) 3222-3234. 10. Zhang G., Li Y., and Lin T. - Caching in information centric networking: A survey, Journal of Computer Networks 57 (10) (2013) 3128-3141. 11. Tyson G., Kauney S., and Miles S. - A Trace-Driven Analysis of Caching in Content- Centric Networks, IEEE Conference on Computer Communications and Networks (ICCCN), Munich, (2012) 1-7. 12. Ghodsi A., Koponen T., and Raghavan B. - Information-Centric Networking: Seeing the Forest for the Trees, Procedure of 10th ACM Workshop Hot Topics in Networks, USA, (2011) 1-6. 13. Li J., Wu H., Liu B., et al. - Popularity-driven Coordinated Caching in Named Data Networking, ACM/IEEE symposium on Architectures for networking and communications systems (ANCS), USA, (2012) 15-26. 14. Ming Z., Xu M., and Wangy D. - Age-based Cooperative Caching in Information-Centric Networks, IEEE Conference on Computer Communication Workshops (INFOCOM) (2012) 268-273. 15. Wang X., Chen M., Taleb T., et al. - Cache in the air: exploiting content caching and delivery techniques for 5G systems, IEEE Communications Magazine (2014) 131-139. 16. ETSI. - LTE, Evolved Universal Terrestrial Radio Access Network (E-UTRAN), Architecture description, 3GPP TS 36.401 version 11.2.0 Release 11, 2013.