تحلیلی از برنامه های مدیریت تحرک توزیع شده با یک مدل مبتنی بر مدت زمان جریان

It is believed that traditional centralized mobility management cannot satisfy the demands of mobile data traffic which has dramatically expanded for years and will remain at a high growth rate in the long term. And now many efforts are putting emphasis on developing distributed mobility management schemes. This paper discusses the advantages of multiple Home Address (HoA) allocating, which is considered as an important character of distributed mobility schemes, by a flow duration based model for the first time as we know. We think that most Internet flows are very short and the best way of forwarding them is using local HoA instead of tunneling to a home link as described in the context of centralized mobility management schemes. We summarize the two cases of distributed mobility management schemes, called the multiple-level tunnel and the direct tunnel and propose a new flow duration based model to evaluate their performance. For the purpose of analysis, we study the cumulative distribution function of Internet flow duration based on our campus real data. Numerical results show that the relatively low velocity and short online time scenario is more suitable for distributed mobility management, and the direct tunnel scheme can always get better performance than the multiple-level tunnel case and the centralized mobility management scheme.

In the context of traditional mobility management schemes such as Mobile IPv6 (MIPv6) and Proxy Mobile IPv6 (PMIPv6), Mobile node (MN) always uses the same Home Address (HoA) or the address generated by Home Network Prefix (HNP) for communication (Perkins et al., 2011 and Bernardos et al., 2010). It is noted that in this paper we do not distinguish these two definitions. All of the Corresponding Nodes (CNs) are not aware of the movement of Mobile Node (MN) which is considered to remain reachable just as it is on the home link all the time. To achieve this purpose, data plane traffic from MN would be forwarded to the mobility anchor such as Home Agent (HA) and Local Mobility Anchor (LMA) over the tunnel to the home link if a route optimization mechanism is not introduced (Liebsch et al., 2011). We consider these traditional schemes as centralized mobility management schemes because the unique mobility anchor not only takes charge in forwarding all data plane traffic but also has to process mobility related signaling messages. In recent years, with the rapid development of mobile Internet related technology, wireless handheld devices such as Smartphone and Tablet, which has more powerful processing capability and longer battery life than ever before, becomes more and more popular. People can access the Internet by higher wireless bandwidth and obtain the ubiquitous mobile service. According to a survey the global mobile data, which had reached 885 peta bytes per month by the end of 2012, is predicted to remain at a high growth speed, nearly doubling for the next few years in a row (Cisco Visual Networking Index, 2017). With the consideration of the system scalability, overall reliability and especially the dramatically increased traffic volume in recent years, the above centralized mobility management scheme can hardly be considered as efficient. A new trend of developing mobility support is to design distributed mobility management (DMM) schemes, which is aimed at solving the above issues. DMM is to distribute the traffic in an optimal way and not to rely on any centralized mobility anchor (DMM, 2012). The requirements and the approaches to achieve DMM are described in Chan et al. (2013), Liu et al. (2013) and Chan et al. (2011). And our earlier work studied the three main goals of DMM (Yi et al., 2013), in which a PMIPv6 based DMM scheme supported by data and control plane separation is proposed. In the context of DMM, as MN always locates on the nearest mobility anchor, apparently the best way to communicate with CNs for MN is using a local HoA or HNP. However, when handover occurs, the traffic with a previous prefix should still be tunneled to a newly accessed mobility anchor as shown in Fig. 1. Thus it is easy to figure out that long flows would survive through several handovers. Here we propose the following interesting questions: How long is the Internet flow duration? How to evaluate the influence of flow duration on the performance of mobility management scheme? This paper will answer these questions hereafter. This paper is based on our earlier work (Yi et al., 2012), which only considers the multiple-level tunnel case of distributed mobility management as we defined in this paper. However, in this paper we conclude two cases of distributed mobility management which are illustrated in Section 2 for comprehensive analytical study, and a more detailed performance analysis is given in Section 5. The remainder of this paper is organized as follows. Section 2 summarizes the main goals and several current efforts on distributed mobility management. Section 3 describes the flow duration based model and analyzes the distributed mobility management scheme׳s performance with such model. Section 4 reviews the previous research work of Internet flow duration and summarizes its probability density distribution characters based on our campus real data analysis. Numerical results are illustrated in Section 5. And finally Section 6 concludes the paper.

This paper summarizes two cases of DMM schemes, called the multiple-level tunnel and the direct tunnel and analyzes the performance of DMM schemes based on flow duration for the first time as we know. We propose a newly defined parameter R to express the ratio of flows that should be tunneled between handovers. Also we study the delivery cost comparison of three kinds of schemes. Based on the analytical model and real data analysis, the value of R shows that most of the Internet flows, which are very short, need not be tunneled when MN moves at a relatively low velocity. Delivery cost analysis shows that the advantage of both DMM schemes is obvious when MN moves at a relatively low velocity and online time is not very long. As the MN online time gets longer, the delivery cost of DMM case one becomes worse. But the performance of DMM case two remains the same. In a high velocity scenario, the performance of DMM case one becomes very bad, but DMM case two can still get a better performance compared to centralized mobility management schemes. Truly, most of Internet flows are very short, but long flows (for example video stream and online game) still exist. Though such flows are very few, multiple-level tunnels will still be set up in DMM case one. To be more efficient, our advice for developing DMM schemes is to follow the architecture of DMM case two.