In this paper, we propose SPARCLE, a novel scheduling system offering network-aware polynomial-time task assignment and resource allocation algorithms for stream processing applications in dispersed computing networks. In particular, we address two major challenges. The first one concerns the assignment of both computation and transport tasks comprising a stream processing application to computing nodes and communication links of the network, respectively, to maximize the application's processing rate. The second one concerns the resource allocation of multiple stream processing applications to satisfy their requested QoS. Our experimental results on a real image stream processing application and extensive simulations show that SPARCLE can increase the application's processing rate by 9 times and 3 times, compared to the cloud computing case and state-of-the-art algorithms, respectively.

Computer networks exist to provide a communication medium for social networks, and information from social networks can help in estimating their communication needs. Despite this, current network management ignores the information from social networks. On the other hand, due to their limited and fluctuating bandwidth, mobile ad hoc networks are inherently resource-constrained. As traffic load increases, we need to decide when and how to throttle the traffic to maximize user satisfaction while keeping the network operational. The state-of-the-art for making these decisions is based on network measurements and so employs a reactive approach to deteriorating network state by reducing the amount of traffic admitted into the network. However, a better approach is to avoid congestion before it occurs, by (a) monitoring the network for early onset signals of congestive phase transition, and (b) predicting future network traffic using user and application information from the overlaying social network. We use machine learning methods to predict the amount of traffic load that can be admitted without transitioning the network to a congestive phase and to predict the source and destination of near future traffic load. These two predictions when fed into an admission control component ensure better management of constrained network resources while maximizing the quality of user experience.

Due to their limited and fluctuating bandwidth, mobile ad hoc networks (MANETs) are inherently resource-constrained. As traffic load increases, we need to decide when to throttle the traffic to maximize user satisfaction while keeping the network operational. The state-of-the-art for making these decisions is based on network measurements and so employs a reactive approach to deteriorating network state by reducing the amount of traffic admitted into the network. However, a better approach is to avoid congestion before it occurs by predicting future network traffic using user and application information from the overlaying social network. We use machine learning methods to predict the source and destination of near future traffic load.

The capacity of mobile Infostation network can be greatly increased, if, in addition to direct short-range communications between mobile nodes and fixed infostations, nodes also communicate amongst themselves whenever they meet. However, this requires cooperation among mobile nodes that are not necessarily spontaneous for commercial applications. We propose means to create opportunistic cooperation in the context of contention distribution in selfish mobile infostation networks. First, we assume that all nodes have a common interest in all files. We stipulate a social contract such that a bilateral file exchange takes place only when either node obtains something it wants from the exchange. Resulting capacity depends on mobility, the number of files being disseminated, and node density. In addition to the existence of multiuser diversity, our results indicate the existence of data diversity-throughput increases as the number of files of interest to all nodes increases. We also consider the case where nodes have dissimilar interests. Results show that as the level of interest overlap decreases, network performance degrades dramatically. We propose an alternate user strategy in the partially overlapping-interests case and show that network throughput is significantly improved by allowing better use of multiuser diversity. We conclude that through opportunistic cooperation, both data and multiuser diversities exist in noncooperative mobile infostation networks.

This paper investigates the use of reactive cognitive radio algorithms to enable co-existence between IEEE 802.11b and 802.16a networks in the same unlicensed band. In particular, we develop a system model in which the two wireless systems share radio resources in frequency, space and time, and reactive coordination methods are used to reduce the mutual interference and improve link throughput. Reactive cognitive radio schemes utilize the available degrees of freedom in frequency, power and time, and react to observations in these dimensions to avoid interference. Dynamic frequency selection (DFS) enables radios to choose the band with the least interference. power control (PC) allows communications at the least possible transmit power. Time agility (TA) enables radios to adapt to each other's traffic patterns and avoid increasing interference in poor channel conditions. Simulation results are given for the following scenarios: (i) single 802.16a cell with single 802.11b hotspot; (ii) multiple 802.16a cells with multiple 802.11b hotspots. The results demonstrate that reactive cognitive radio schemes can provide significant improvements in 802.11b and/or 802.16a throughputs in the typical operating scenarios considered.

The paper describes the design and evaluation of ad-hoc extensions to the IEEE 802.15.3 medium access control (MAC) layer for wireless personal area networks (WPANs). The proposed protocol allows communication between ad-hoc devices without the intervention of any central entity and, at the same time, ensures bounded delays for isochronous traffic. Ad-hoc communication is made possible without the hidden terminal problem. Features from both IEEE 802.15.3 and IEEE 802.11 standards are used - in particular the TDMA structure from 802.15.3 and RTS-CTS based contention from 802.11. The protocol includes certain other ingredients, like a decentralized synchronization procedure using randomized beaconing, periodically interspersed contention periods, bit maps to convey reservation information and a mechanism to estimate and react to channel errors. The MAC has been simulated in ns-2 and simulation results are reported.

The assumption that all nodes cooperate to relay packets for each other may not be realistic for commercial wireless ad hoc networks. An autonomous (selfish) node in a wireless network has two disincentives for forwarding for others: energy expenditure (real cost) and possible delays for its own data (opportunity cost). We introduce a mechanism that "fosters cooperation through bribery" in the context of forwarding in ad hoc networks. Using a microeconomic framework based on game theory, we design and analyze a pricing algorithm that encourages forwarding among autonomous nodes by reimbursing forwarding. Taking a joint network-centric and user-centric approach, the revenue maximizing network and utility (measured in bits-per-Joule) maximizing nodes interact through prices for channel use, reimbursements for forwarding, transmitter power control, as well as forwarding and destination preferences. In a three-node (two-sources, one-access-point) network, the network converges to an architecture that induces forwarding only when the network geometries are such that forwarding is likely to increase individual benefits (network revenue and node utilities). For other geometries, the network converges to architectures that do not favor forwarding. We then generalize to a multinode network, where it is seen that the nodes' willingness to forward decrease for large ratios of the average internodal distance to the smallest distance between the access point and any source node. Pricing with reimbursement generally improves the network aggregate utility (or aggregate bits-per-Joule), as well as utilities and revenue compared with the corresponding pricing algorithm without reimbursement.

A pricing mechanism to mediate (and allocate resources) between conflicting user and network objectives has been recently proposed by the authors in a single-cell system. Here, we extend the results to a multicell system, where the autonomous base station assignment and power control are formulated as a noncooperative game among users. The network prices the resources using two strategies: global pricing that maximizes the revenue, and minimax pricing that trades off the revenue for a more even resource allocation.

Using a microeconomic framework based on game theory, we design and analyze a pricing algorithm that encourages forwarding among autonomous (selfish) nodes by reimbursing forwarding. Taking a joint network-centric and user-centric approach, the revenue maximizing network and utility maximizing nodes interact through prices for channel use, reimbursements for forwarding, transmitter power control as well as forwarding and destination preferences. In a three-node (two sources, one access point) network, the network converges to an architecture that induces forwarding only when the network geometries are such that forwarding is likely to bring higher network efficiency. For other geometries, the network converges to architectures that do not favor forwarding. We characterize the Nash equilibria for each geometry and design specific game implementations to achieve the most Pareto superior Nash equilibrium. In a general multi-node network, our results indicate that the nodes' willingness to forward decrease for large values of the ratio of the average internodal distance to the average distance between the access point and nodes with direct connections to it.

Objectives of most radio resource-management schemes can be classified as either user centric or network centric. User-centric schemes try to maximize the interests of individual users, while network-centric schemes optimize collective metrics for all users. These two types of resource management tend to result in qualitatively different resource allocations (with, sometimes, very different degrees of fairness). In this paper, we consider the joint optimization of both user-centric and network-centric metrics. Specifically, we use a utility function (measured in units of bits per Joule) as the user-centric metric, and for the network-centric counterpart, we consider a function of the sum of the throughputs of users in the network. The user-centric measure reflects the individual user's throughput, as well as the battery energy (transmit power) consumed to achieve it. The network-centric measure reflects the total revenue derived by the usage of network resources. We introduce an explicit pricing mechanism to mediate between the user-centric and network-centric resource-management problems. Users adjust their power in a distributed fashion to maximize the difference between their utilities and their payments (measured as a product of the unit price and throughput). The network adjusts the unit price in order to maximize its revenue (measured as the sum of the individual payments). We show that the distributed user-centric power control results in a unique Nash equilibrium. Our numerical results indicate that there exists a unique unit price that maximizes the revenue of the network. We also derive a semianalytical, computationally simple, and highly accurate approximation to the optimal solution. Our results show that while users with better channels receive better qualities of service, as usual (e.g., as in waterfilling), they also make proportionally higher contributions to the network revenue.

In wireless networks, it is often assumed that all nodes cooperate to relay packets for each other. Although this is a plausible model for military or mission based networks, it is unrealistic for commercial networks and future pervasive computing environments. We address the issue of noncooperation between nodes in the context of content distribution in mobile infostation networks. We assume all nodes have common interest in all files cached in the fixed infostations. In addition to downloading files from the fixed infostations, nodes act as mobile infostations and exchange files when they are in proximity. We stipulate a social contract such that an exchange occurs only when each node can obtain something it wants from the exchange. Our social contract enables much higher system efficiency compared to downloading from fixed infostations only while not requiring true cooperation among nodes. We show by analysis and simulations that network performance depends on the node density, mobility and the number of files that are being disseminated. Our results point to the existence of data diversity for mobile infostation networks. The achievable throughput increases as the number of files of interest to all users increases. We have also extended the common interest model to the case where nodes have dissimilar interests. Our simulation results show that as mobile nodes change from having identical interests to mutually exclusive interests, the network performance degrades dramatically. We propose an alternative user strategy when nodes have partially overlapping interests and show that the network capacity can be significantly improved by exploiting multiuser diversity inherent in mobile infostation networks. We conclude that data diversity and multiuser diversity exist in noncooperative mobile infostation networks and can be exploited.

In wireless networks, it is often assumed that all nodes cooperate to relay packets for each other. Although this is a plausible model for military or mission-based networks, it is unrealistic for commercial networks and future pervasive computing environments. We address the issue of noncooperation between nodes in the context of content distribution in mobile infostation networks. All nodes have common interest in all files cached in the fixed infostations. In addition to downloading files from the fixed infostations, nodes act as mobile infostations and exchange files when they are in proximity. We stipulate a social contract such that an exchange occurs only when each node can obtain something it wants from the exchange. We show by analysis and simulations that network performance depends on node density, mobility and the number of files that are being disseminated. Our results point to the existence of data diversity for mobile infostation networks. As the number of files of interest to all users increases, the achievable throughput increases. Moreover, each user has a fairer share of the total network throughput. In particular, the transmission of each channel is only limited by contention, indicating the noncooperation strategy achieves near optimum resource utilization.