Yujin Li

Can Mobile Cloudlets Support Mobile Applications

A mobile cloudlet is a set of resource-rich mobile devices - referred to as cloudlet nodes - that an initiator mobile device can connect to for services. In this paper, we examine the fundamental mobile cloudlet properties that unfold whether and when a mobile cloudlet can provide mobile application service. Specifically, we investigate the cloudlet size, cloudlet nodes lifetime and reachable time. Traces and mathematical analysis demonstrate that 1) the more frequently mobile devices meet, the larger the pool of computing resources an initiator can access; 2) intermittent connection between devices has little adverse effect on the optimal computing performance of mobile cloudlet in the long run; 3) the ratio E(TC)/[E(TI)+E(TC)] indicates the connection likelihood of an initiator and a cloudlet node (i.e., reachability of the cloudlet node), where TC and TI are their contact and inter-contact time. We further derive upper and lower bounds on computing capacity and computing speed of a mobile cloudlet. An initiator can use both bounds to decide whether to offload its task to remote clouds or local mobile cloudlets for better mobile application services.

Modeling and Analytical Study of Link Properties in Multihop Wireless Networks

The radio link between a pair of wireless nodes is determined by radio channels, transmission range, node mobility and node-pair distance, which form a set of random factors in multihop wireless networks. The properties of such radio links can be characterized by link lifetime, residual link lifetime and link change rate, which, in fact, have been widely used for network design and performance evaluation. In this paper, we take a new modeling approach that captures the dynamics of radio channels and node movements in small-scale. More specifically, distance transition probability matrix is designed in order to describe the joint effects of dynamic transmission range due to radio channel fading and relative distance of a node-pair resulting from random movements. We find that the PDF of link lifetime can be approximated by an exponential distribution with parameter characterized by the ratio of average node speed \bar{V} to effective transmission range R_e. To further understand the implication of link properties, analytical results are used to investigate the upper bound of network connectivity and the associated network performance is evaluated by extensive simulations.

The unheralded power of cloudlet computing in the vicinity of mobile devices

With the popularity of smartphones and explosion of mobile applications, mobile devices are becoming the prevalent computing platform for convenient communication and rich entertainment. Because mobile devices still have limited processor power, computing-intensive applications need to be offloaded to either remote clouds or nearby cloudlets for processing. But, remote cloud computing is hindered by the long latency and expensive roaming charges of cellular radio access. Therefore, cloudlet computing becomes appealing to provide instant and low-cost service through resource-rich devices (e.g., desktops) in the vicinity of mobile devices. It is evident that cloudlet computing is challenged by the intermittent connection between cloudlets and mobile devices due to user mobility. The question is how to evaluate the impact of user mobility on cloudlet computing performance. In this paper, we examine the cloudlet access probability, task success rate, and task execution speed to measure the impact of mobility. We discover that the cloudlet access probability is μTc /(μT1 + μTc) determined by mean connection time μTc and mean inter-connection time μT1 between the mobile device and the cloudlet. Furthermore, we find that the task success rate and execution speed depend on not only task computation demand and cloudlet computing speed but also cloudlet access probability. Our findings reveal that the ratio μTc /(μT1 + μT c) quantifies the impact of node mobility on both cloudlet access probability and cloudlet computing performance.

Exploring device-to-device communication for mobile cloud computing

With the popularity of smartphones and explosion of mobile applications, mobile devices become the prevalent computing platform for convenient communication and rich entertainment. Mobile cloud computing (MCC) is proposed to overcome the limited resources of mobile systems. However, when users access MCC through wireless networks, cellular network is likely to be overloaded and Wi-Fi connectivity is intermittent. Therefore, device-to-device (D2D) communication is exploited as an alternative for MCC. An important issue in exploring D2D communication for MCC is how users can detect and utilize the computing resources on other mobile devices. In this paper, we propose two mobile cloud access schemes: optimal and periodic access schemes, and study the corresponding performance of mobile cloud computing (i.e., mobile cloud size, nodes serviceable time percentage, and task success rate). We find that optimally, nodes serviceable time percentage and task success rate approach 1. Using more practical periodic access scheme, nodes serviceable time percentage and task success rate are determined by the ratio of contact and inter-contact time between two nodes.

Horizon on the move: Geocast in intermittently connected vehicular ad hoc networks

Vehicular ad hoc network (VANET) is one of the most promising large-scale applications of mobile ad hoc networks. VANET applications are rooted in advanced understanding of communication networks because both control messages and data information need to be disseminated in geographic regions (i.e., Geocast). The challenges come from highly dynamic environments in VANET. Destination nodes in geocast are dynamic over time due to vehicle mobility, which undermines existing results on dissemination latency and information propagation speed with pre-determined destinations. Moreover, the affected area by the dissemination, which is referred to as horizon of message (HOM), is critical in geocast as it determines the latency for the message reaching nodes inside the area of interest (AOI), in which the message is relevant to drivers. Therefore, we characterize the HOM in geocast by how far the message can reach within time t (referred as dissemination distance) and how long the message takes to inform nodes at certain locations (referred as hitting time). Analytic bounds of dissemination distance and hitting time are derived under four types of dissemination mechanisms, which provide insights into the spatial and temporal limits of HOM as well as how the numbers of disseminators and geographic information exchanges affect them. Applying analytic and simulation results to two real applications, we observe that geocast with AOI near the source or high reliability requirement should recruit multiple disseminators while geocast with AOI far from the source need to utilize geographic information for fast message propagation.

Internode Mobility Correlation for Group Detection and Analysis in VANETs

Recent studies on mobility-assisted schemes for routing and topology control and on mobility-induced link dynamics have presented significant findings on the properties of a pair of nodes (e.g., the intermeeting time and link life time) or a group of nodes (e.g., network connectivity and partitions). In contrast to the study on the properties of a set of nodes rather than individuals, many works share a common ground with respect to node mobility, i.e., independent mobility in multihop wireless networks. Nonetheless, in vehicular ad hoc networks (VANETs), mobile devices installed on vehicles or held by humans are not isolated; however, they are dependent on each other. For example, the speed of a vehicle is influenced by its close-by vehicles, and vehicles on the same road move at similar speeds. Therefore, the gap between our understanding of the impact of independent mobility and our interest in the properties of correlated mobility in VANETs, along with the real systems altogether, declare an interesting question. How can we measure the internode mobility correlation, such as to uncover the node groups and network components, and explore their impact on link dynamics and network connectivity? Bearing this question in mind, we first examine several traces and find that node mobility exhibits spatial locality and temporal locality correlations, which are closely related to node grouping. To study the properties of these groups on the fly, we introduce a new metric, i.e., dual-locality ratio (DLR), which quantifies mobility correlation of nodes. In light of taking spatial and temporal locality dimensions into account, the DLR can be used to effectively identify stable user groups, which in turn can be used for network performance enhancement.

Message Dissemination in Intermittently Connected D2D Communication Networks

Device-to-device (D2D) communications enable direct communications and information distribution among closely located devices in wireless networks. Many applications of D2D communications require message dissemination to a group of mobile users at certain locations. The challenges of message dissemination come from highly dynamic network environments due to movements of devices. Existing studies on message dissemination have focused on information propagation speed and latency, but the size of the area affected by message dissemination at time t is also critical to D2D communication applications that heavily depend on message dissemination among users in a geographic region. In this paper, we study the fundamental issues in D2D communications: how far the message dissemination can reach by time t (referred as dissemination distance) and how long the dissemination takes to inform nodes located at distanced (referred to as hitting time), especially in dynamic, intermittently connected networks. We first derive analytic bounds of dissemination distance and hitting time under different dissemination mechanisms, providing the spatial and temporal limits of message dissemination. Analytic results are further validated by simulation results of several corresponding dissemination algorithms. Finally, two application scenarios are provided to illustrate how our results serve as guidelines to choose or design appropriate dissemination methods for different D2D communication applications.

Intermittently Connected Vehicle-to-Vehicle Networks: Detection and Analysis

Vehicular Adhoc Networks (VANETs) are dedicated to improve the safety and efficiency of transportation systems through vehicle to vehicle or vehicle to road side communications. VANETs exhibit dynamic topology and intermittent connectivity due to high vehicle mobility. These distinguished features declare a challenging question: how to detect on the fly vehicular networks such that we can explore mobility-assisted message dissemination and topology control in VANETs. As being closely related to network dynamics, vehicle mobility could be explored to uncover network structure. In this paper, we have observed that mobility of vehicle, rather than being random, shows \emph{temporal locality} (i.e., frequently visiting several communities like home and office), and \emph{spatial locality} (i.e., velocity constrained by road layout and nearby vehicles). We first examine temporal locality using a campus trace, then measure temporal locality similarity between two vehicles based on the relative entropy of their location preferences. By further incorporating spatial locality similarity, we introduce a new metric, namely \emph{dual locality ratio} (DLR), which represents the mobility correlation of vehicles. Simulation results show that DLR can effectively identify dynamic vehicular network structures. We also demonstrate applications of DLR for improving performances of data forwarding and clustering in vehicle-to-vehicle networks.

Geo-dissemination in vehicular ad hoc networks

Vehicular Adhoc Networks (VANETs) aim to improve road safety and convenience through vehicle-to-vehicle and vehicle-to-roadside communications. Traffic information and accident warnings are often disseminated to vehicles in certain areas where driving could be affected by hazardous situations. Such message dissemination with destinations confined in specific geographic regions is referred to as Geo-Dissemination. In this paper, we analyze how far a geo-dissemination can possibly reach over a period of time t (denoted as dissemination distance D(t)), and what is the latency for a message to reach locations that are d distance far from the source (denoted as the stopping time τ). Simulations results of two dissemination methods (stateless opportunistic forwarding and GPS-based message broadcasting) are compared with our analytic results.

The Impact of Network Size and Mobility on Information Delivery in Cognitive Radio Networks

There have been extensive works on the design of opportunistic spectrum access and routing schemes to improve spectrum efficiency in cognitive radio networks (CRNs), which becomes an integral component in the future communication regime. Nonetheless, the potentials of CRNs in boosting network performance yet remain to be explored to reach the full benefits of such a phenomenal technique. In this paper, we study the end-to-end latency in CRNs in order to find the sufficient and necessary conditions for real-time applications in finite networks and large-scale deployments. We first provide a general mobility framework which captures most characteristics of the existing mobility models and takes spatial heterogeneity into account. Under this general mobility framework, secondary users are mobile with an mobility radius a, which indicates how far a mobile node can reach in spatial domain. We find that there exists a cutoff point on a, below which the latency has a heavy tail and above which the tail of the latency is bounded by some Gamma distributions. As the network grows large, the latency is asymptotically scalable (linear) with respect to the dissemination distance (e.g., the number of hops or euclidean distance). An interesting observation is that although the density of primary users adversely impacts the expected latency, it makes no influence on the dichotomy of the latency tail in finite networks and the linearity of latency in large networks. Our results encourage CRN deployment for real-time and large applications, when the mobility radius of secondary users is large enough.