We introduce Model-Distributed Inference for Large-Language Models (MDI-LLM), a novel framework designed to facilitate the deployment of state-of-the-art large-language models (LLMs) across low-power devices at the edge. This is accomplished by dividing the model into multiple partitions, which are then assigned to different devices/nodes within the network. These nodes exchange intermediate activation vectors via device-to-device links, enabling collaborative computation. To enhance the efficiency of this process, we propose the "recurrent pipeline parallelism" technique, which reduces idle time on each device and facilitates parallel inference during the generation of multiple text sequences. By leveraging the combined computational resources of multiple edge devices, MDI-LLM enables the deployment of LLMs that exceed the memory capacity of individual devices, making it possible to perform inference on low-cost hardware. Furthermore, as the number of participating devices increases, MDI-LLM boosts token generation throughput and reduces memory usage.

Distributed inference techniques can be broadly classified into data-distributed and model-distributed schemes. In data-distributed inference (DDI), each worker carries the entire deep neural network (DNN) model but processes only a subset of the data. However, feeding the data to workers results in high communication costs, especially when the data is large. An emerging paradigm is model-distributed inference (MDI), where each worker carries only a subset of DNN layers. In MDI, a source device that has data processes a few layers of DNN and sends the output to a neighboring device, i.e., offloads the rest of the layers. This process ends when all layers are processed in a distributed manner. In this paper, we investigate the design and development of MDI with early-exit, which advocates that there is no need to process all the layers of a model for some data to reach the desired accuracy, i.e., we can exit the model without processing all the layers if target accuracy is reached. We design a framework MDI-Exit that adaptively determines early-exit and offloading policies as well as data admission at the source. Experimental results on a real-life testbed of NVIDIA Nano edge devices show that MDI-Exit processes more data when accuracy is fixed and results in higher accuracy for the fixed data rate.

Multi-party computation (MPC) is promising for designing privacy-preserving machine learning algorithms at edge networks. An emerging approach is coded-MPC (CMPC), which advocates the use of coded computation to improve the performance of MPC in terms of the required number of workers involved in computations. The current approach for designing CMPC algorithms is to merely combine efficient coded computation constructions with MPC. We show that this approach fails short of being efficient; e.g., entangled polynomial codes are not necessarily better than PolyDot codes in MPC setting, while they are always better for coded computation. Motivated by this observation, we propose a new construction; Adaptive Gap Entangled (AGE) polynomial codes for MPC. We show through analysis and simulations that MPC with AGE codes always perform better than existing CMPC algorithms in terms of the required number of workers as well as computation, storage, and communication overhead.

Random walk learning (RWL) has recently gained a lot of attention thanks to its potential for reducing communication and computation over edge networks in a decentralized fashion. In RWL, each node in a graph updates a global model with its local data, selects one of its neighbors randomly, and sends the updated global model. The selected neighbor becomes a newly activated node, so it updates the global model using its local data. This continues until convergence. Despite its promise, RWL has two challenges: (i) training time is long, and (ii) nodes should have the complete model. Thus, in this paper, we design Random Walking Snakes (RWS), where a set of nodes instead of one node is activated for model update, and each node in the set trains a part of the model. Thanks to model partitioning and parallel processing in the set of activated nodes, RWS reduces both the training time and the amount of the model that needs to be stored. We also design a novel policy that determines the set of activated nodes by taking into account the computing power of nodes. Simulation results show that RWS significantly reduces the convergence time as compared to RWL.

Cooperative computation is a promising approach for localized data processing at the edge, e.g., for Internet of Things(IoT). Cooperative computation advocates that computationally intensive tasks in a device could be divided into sub-tasks, and offloaded to other devices or servers in close proximity. However, exploiting the potential of cooperative computation is challenging mainly due to the heterogeneous and time-varying nature of edge devices. Coded computation, which advocates mixing data in sub-tasks by employing erasure codes and offloading these sub-tasks to other devices for computation, is recently gaining interest, thanks to its higher reliability, smaller delay, and lower communication costs. In this paper, we develop a coded cooperative computation framework, which we name Coded Cooperative Computation Protocol (C3P), by taking into account the heterogeneous and time-varying resources of edge devices. C3P dynamically offloads coded sub-tasks to helpers and is adaptive to time-varying resources. We show that (i) task completion delay of C3P is very close to optimal coded cooperative computation solutions, (ii) the efficiency of C3P in terms of resource utilization is higher than 99%, and (iii) C3P improves task completion delay significantly as compared to baselines via both simulations and in a testbed consisting of real Android-based smartphones.

Existing deep learning-based Text-to-Speech (TTS) mechanisms are computationally intensive, which puts a strain in their practical applications especially over edge networks comprised of resource constrained devices. Our focus is on distributing TTS tasks across multiple devices (i.e., workers) at edge networks and providing TTS-aware resiliency against straggling workers. In particular, we design a REsilient DIstributed Tts (REDIT) framework by exploiting the text summarization as redundancy to provide resiliency for distributed TTS. We show analytically that REDIT improves the task completion time as compared to the distributed TTS without resiliency. We determine the optimum amount of redundancy/summary based on our task completion time analysis. We implement our REDIT framework in a real testbed consisting of NVIDIA Jetson Nano cards, and show that our REDIT algorithm improves the task completion delay as compared to baselines.

Multi-party computation (MPC) is promising for designing privacy-preserving machine learning algorithms at edge networks. An emerging approach is coded-MPC (CMPC), which advocates the use of coded computation to improve the performance of MPC in terms of the required number of workers involved in computations. The current approach for designing CMPC algorithms is to merely combine efficient coded computation constructions with MPC. Instead, we propose a new construction; Adaptive Gap Entangled polynomial (AGE) codes, where the degrees of polynomials used in computations are optimized for MPC. We show that MPC with AGE codes (AGE-CMPC) perform better than existing CMPC algorithms in terms of the required number of workers as well as storage, communication and computation load.

The traditional approach to distributed deep neural network (DNN) training is data-distributed learning, which partitions and distributes data to workers. This approach, although has good convergence properties, has high communication cost, which puts a strain especially on edge systems and increases delay. An emerging approach is model-distributed learning, where a training model is distributed across workers. Model-distributed learning is a promising approach to reduce communication and storage costs, which is crucial for edge systems. In this paper, we design ResPipe, a novel resilient model-distributed DNN training mechanism against delayed/failed workers. We analyze the communication cost of ResPipe and demonstrate the trade-off between resiliency and communication cost. We implement ResPipe in a real testbed consisting of Android-based smartphones, and show that it improves the convergence rate and accuracy of training for convolutional neural networks (CNNs).

This paper considers a broadcast scenario where a base station seeks to transmit a common content to a group of users. Due to impairments inherent to broadcast channels, transmitted content may be partially received by the users, thereby requiring retransmissions to recover missing parts of the content. The retransmission process can be optimized by leveraging instantly decodable network coding (IDNC) with the help of disjoint clusters of combinations of single-hop fully connected and/or multi-hop partially connected device-to-device (D2D) network configurations. This paper aims to find an efficient encoding strategy for the broadcasted packets that minimizes the packet completion time problem, defined as the number of transmission slots necessary to satisfy the demand of all users. To this end, we design an efficient mechanism, LoCoNCDC, which enables us to find feasible codes that can be transmitted by the base station via cellular links. Moreover, the proposed LoCoNCDC permits us to identify the set of transmitting D2D users along with the corresponding IDNC packets that the transmitting users can broadcast through D2D links. Furthermore, we theoretically characterize the performance bound of the proposed approach. Finally, the efficiency of the proposed LoCoNCDC is validated through numerical results which demonstrate substantial performance gain compared to existing schemes.

Edge computing is emerging as a new paradigm to allow processing data at the edge of the network, where data is typically generated and collected, by exploiting multiple devices at the edge collectively. However, exploiting the potential of edge computing is challenging mainly due to the heterogeneous and time-varying nature of edge devices. Coded computation, which advocates mixing data in sub-tasks by employing erasure codes and offloading these sub-tasks to other devices for computation, is recently gaining interest, thanks to its higher reliability, smaller delay, and lower communication cost. In this paper, our focus is on characterizing the cost-benefit trade-offs of coded computation for practical edge computing systems, and develop an adaptive coded computation framework. In particular, we focus on matrix multiplication as a computationally intensive task, and develop an adaptive coding for matrix multiplication (ACM2) algorithm by taking into account the heterogeneous and time varying nature of edge devices. ACM2dynamically selects the best coding policy by taking into account the computing time, storage requirements as well as successful decoding probability. We show that ACM2improves the task completion delay significantly as compared to existing coded matrix multiplication algorithms.

We consider a model-distributed learning framework in which layers of a deep learning model is distributed across multiple workers. To achieve consistent gradient updates during the training phase, model-distributed learning requires the storage of multiple versions of the layer parameters at every worker. In this paper, we design mcPipe to reduce the memory cost of model-distributed learning, which is crucial in memory-constrained edge computing devices. mcPipe uses an on-demand weight updating policy, which reduces the amount of weights that should be stored at workers. We analyze the memory cost of mcPipe and demonstrate its superior performance as compared to existing model-distributed learning mechanisms. We implement mcPipe in a real testbed and show that it improves the memory cost without hurting converge rate and computation cost.

Distributed computation plays an essential role in cloud and edge computing. Data such as images, audio, and text can be represented as matrices to facilitate efficient computation, especially in the domains of distributed machine learning, computer vision, and signal processing. Many coded computation algorithms have been proposed for big data applications to securely partition and distribute matrices to parallel worker devices. However, these proposals have yet to be adapted for mobile platforms beyond theoretical means. Mobile IoT networks can greatly benefit from secure distributed computing, however, commercial devices such as smartphones and tablets are much more limited in resources compared to platforms in data centers, requiring special design considerations. We investigate existing distribution schemes from an operational complexity and security viewpoint and study their performance in several mobile IoT networks, identifying performance bottlenecks in regards to communication and computation costs. From our findings, we propose new, scalable algorithms optimized to handle the unique constraints of mobile IoT. Extensive evaluations of our proposals on publicly available image classification datasets show how distributed learning can be specially optimized to enhance runtime and battery performance on mobile IoT by over 10×.

In this paper, we study the problem of transmittinga common content to a number of cellular users with the help of device-to-device (D2D) links via instantly decodable network coding (IDNC). In particular, a common content that is broadcast by a base station may be partially received by cellular users due to packet erasures over cellular links. D2D links among cellular users, which may be intermittently available due to mobility and link imperfections, are used to recover the missing content. In this setup, we design an IDNC mechanism to minimize packet completion time, defined as the amount of time it takes for the common content to be received by all cellular users. We develop an optimal packet completion time strategy by constructing a two-layer IDNC conflict graph where the higher layer represents all feasible network coded packets that can be transmitted over cellular links, while the lower layer consists of network coded packets (as well as corresponding cellular users that can generate and transmit these packets) that can be transmitted via D2D links. We show that finding the optimal IDNC packets to minimize the packet completion time problem is equivalent to finding the maximum independent set of the two-layer IDNC conflict graph. We also design a sub-optimal IDNC mechanism, aIDNC, that efficiently finds IDNC packets to be transmitted over cellular and D2D links. Finally, the efficiency of our proposed approach is verified through numerical experiments.

Robust and computationally efficient anomaly detection in videos is a problem in video surveillance systems. We propose a technique to increase robustness and reduce computational complexity in a Convolutional Neural Network (CNN) based anomaly detector that utilizes the optical flow information of video data. We reduce the complexity of the network by denoising the intermediate layer outputs of the CNN and by using powers-of-two weights, which replaces the computationally expensive multiplication operations with bit-shift operations. Denoising operation during inference forces small valued intermediate layer outputs to zero. The number of zeros in the network significantly increases as a result of denoising, we can implement the CNN about 10% faster than a comparable network while detecting all the anomalies in the testing set. It turns out that denoising operation also provides robustness because the contribution of small intermediate values to the final result is negligible. During training we also generate motion vector images by a Generative Adversarial Network (GAN) to improve the robustness of the overall system. We experimentally observe that the resulting system is robust to background motion.

Large data transfers can result in significant congestion and performance degradation for interactive end-user applications such as web browsing and streaming. While there are existing TCP congestion control algorithms for delivery of large volume data (e.g., LEDBAT, TCP-LP), our results show that these protocols are not effective in cellular networks due to variability in radio channel conditions and the use of cellular schedulers in base stations. We propose Legilimens, an agile TCP variant for cellular downlink transfers, which not only retains desirable properties of existing approaches, but also exploits the properties of the cellular schedulers to estimate load and capacity and addresses the challenges in cellular networks. As a result, Legilimens is able to deliver traffic using only the spare capacity on the downlink. We conduct extensive evaluations of Legilimens in multiple settings-in a large cellular network for real-world performance, on the PhantomNet emulator for controlled experiments, and ns-3 simulator for scaled experiments-all of which demonstrate that Legilimens is superior to existing protocols in transferring large volumes of data without interfering with regular user traffic. Compared to existing low-priority protocols, Legilimens improves the throughput of background flows by 2x on average (up to 5x) without degrading the performance of foreground flows across all the three testbeds.

Edge computing is emerging as a new paradigm to allow processing data at the edge of the network, where data is typically generated and collected by exploiting multiple devices at the edge collectively. However, offloading tasks to other devices leaves the edge computing applications at the complete mercy of an attacker. One of the attacks, which is the focus of this work, is Byzantine attacks, where one or more devices can corrupt the offloaded tasks. Furthermore, exploiting the potential of edge computing is challenging mainly due to the heterogeneous and time-varying nature of the devices at the edge. In this paper, we develop a secure coded cooperative computation mechanism (SC3) that provides both security and computation efficiency guarantees by gracefully combining homomorphic hash functions and coded cooperative computation. Homomorphic hash functions are used against Byzantine attacks and coded cooperative computation is used to improve computation efficiency when edge resources are heterogeneous and time-varying. Simulation results show that SC3 improves task completion delay significantly.

Effective communication among first responders during and in the aftermath of a disaster can affect outcomes dramatically. We seek to build a resilient architecture that allows first responders to communicate even with: 1) damage to infrastructure - civilian and / or specialized communication facilities may be damaged by the disaster, and 2) dynamically formed groups - first responder teams may be formed dynamically in response to a disaster and team member addresses (e.g., phone numbers, network addresses) may not be known to one another. We propose a resilient network architecture that allows efficient communication among first responders during and after a disaster [1]. We seek to support dynamically formed groups for incident response, allowing first responders to securely and conveniently communicate based on roles (names). The architecture supports communication in disasters by 1) building resilience into the framework across all the layers, 2) creating a framework that allows communication by role and identity, rather than addresses, 3) supporting multiple modalities (data, voice) for communication among dynamically formed first responder teams, and 4) providing robust and resilient communication and computing even when facilities are error- and disruption-prone.

This paper studies the Layered Packet Erasure Broadcast Channel (LPE-BC) with Channel Output Feedback (COF), which is a high-SNR approximation of the fading Gaussian BC, proposed by Tse and Yates in 2012 for the case without COF. This model is also a multi-layer generalization of the Binary Erasure Channel (BEC). In a past work, the Authors derived inner and outer bounds to the rate region (set of achievable rates with backlogged arrivals) of the LPE-BC with COF; here, the arrival region (set of exogenous arrival rates for which packet arrival queues are stable) for the same model is analyzed. For the case of K=2 users and Q ≥ 1 layers, the known achievable rate region and the derived arrival region coincide; both strategically employ a. For the case of Q = 2 layers, sufficient conditions are given for the achievable arrival region to coincide with the known converse rate region, thus showing that in those cases the optimal rate and arrival regions coincide.

A large variety of traffic - time-sensitive “foreground” traffic (e.g., web browsing) and time-insensitive “background” traffic (e.g., software updates) - compete for the scarce cellular bandwidth, especially on the downlink. While there is limited in-network support for traffic prioritization, existing endto-end, “low priority transport protocols” exhibit sub-optimal performance in cellular networks. We propose Sneaker, which yields to time-sensitive foreground traffic during periods of congestion and enables time-insensitive background traffic to efficiently utilize any spare capacity. Sneaker achieves the desired goal by randomly dropping packets coming into the base station, based on traffic type and network conditions. Our key contribution is the derivation of the optimal dropping rate and a practical dropping rate, which performs close to optimal. Further, Sneaker co-exists and performs well with existing cellular schedulers and transport protocols.

This paper, motivated by mission-critical and latency-constrained traffic, focuses on delivering different messages to various end-users within hard deadlines on the downlink of a wireless system. A novel deadline outage performance criterion is introduced, which accounts for both the violation of the hard deadline by any of the messages as well as for channel decoding errors due to finite block-length. This formulation allows for the study of scheduling policies under a more refined model of the physical layer channel than is usually assumed in the networking literature. Different scheduling polices under hard deadline constraints are proposed, and the corresponding deadline outage probabilities evaluated. The main result is that, to reduce the deadline outage probability, a scheduling policy should cleverly combine time-sharing and concatenate-and- code.

Cooperative computation is a promising approach for localized data processing at the edge, e.g., for Internet of Things (IoT). Cooperative computation advocates that computationally intensive tasks in a device could be divided into sub-tasks, and offloaded to other devices or servers in close proximity. However, exploiting the potential of cooperative computation is challenging mainly due to the heterogeneous and time-varying nature of edge devices. Coded computation, which advocates mixing data in sub-tasks by employing erasure codes and offloading these sub-tasks to other devices for computation, is recently gaining interest, thanks to its higher reliability, smaller delay, and lower communication costs. In this paper, we develop a coded cooperative computation framework, which we name Coded Cooperative Computation Protocol (C3P), by taking into account the heterogeneous resources of edge devices. C3P dynamically offloads coded sub-tasks to helpers and is adaptive to time-varying resources. We show that (i) task completion delay of C3P is very close to optimal coded cooperative computation solutions, (ii) the efficiency of C3P in terms of resource utilization is higher than 99%, (iii) C3P improves task completion delay significantly as compared to baselines via both simulations and in a test-bed consisting of real Android-based smartphones.

Traditional means for contacting emergency responders depend critically on the availability of the 911 service to request help. Large-scale natural disasters such as hurricanes and earthquakes often result in overloading and sometimes failure of communication facilities. Affected citizens are increasingly using social media to obtain and disseminate information. Social media is not only being used to communicate with first responders but also for people to organically volunteer and seek help from each other, complementing the role of first responders. In this paper, we examine the use of Twitter during two major hurricanes in the U.S. in 2017. We find that there exists a sizable number of people with access to the Internet even in areas where 911 services were down, and they tweet disaster-related information including requests for help. Our analysis indicates that social media can potentially help in disaster management and improve outcomes.

Edge computing is a promising approach for localized data processing for many edge applications and systems including Internet of Things (IoT), where computationally intensive tasks in IoT devices could be divided into sub-tasks and offloaded to other IoT devices, mobile devices, and/or servers at the edge. However, existing solutions on edge computing do not address the full range of challenges, specifically heterogeneity; edge devices are highly heterogeneous and dynamic in nature. In this paper, we develop a predictive edge computing framework with hard deadlines. Our algorithm; PrComp (i) predicts the uncertain dynamics of resources of devices at the edge including energy, computing power, and mobility, and (ii) makes sub-task offloading decisions by taking into account the predicted available resources, as well as the hard deadline constraints of tasks. We evaluate Prcomp on a testbed consisting of real Android-based smartphones, and show that it significantly improves energy consumption of edge devices and task completion delay as compared to baselines.

Optimal deployment and movement of multiple unmanned aerial vehicles (UAVs) is studied. The scenarios of variable data rate with fixed transmission power, and variable transmission power with fixed data rate are considered. First, the optimal deployment of UAVs is studied for the static case of a fixed ground terminal (GT) density. Using high resolution quantization theory, the corresponding best achievable performance (average data rate or transmission power) is determined in the asymptotic regime of a large number of UAVs. Next, the dynamic case where the GT density is allowed to vary periodically through time is considered. For one-dimensional networks, an accurate formula for the total amount of UAV movement that guarantees the best time-averaged performance is determined. In general, the tradeoff between the total UAV movement and the achievable performance is obtained through a Lagrangian approach. A corresponding trajectory optimization algorithm is introduced and shown to guarantee a convergent Lagrangian. Numerical simulations are also carried out to confirm the analytical findings.

As the number of Internet of Things (IoT) devices keeps increasing, data is required to be communicated and processed by these devices at unprecedented rates. Cooperation among wireless devices by exploiting Device- to-Device (D2D) connections is promising, where aggregated resources in a cooperative setup can be utilized by all devices, which would increase the total utility of the setup. In this paper, we focus on the resource allocation problem for cooperating IoT devices with multiple heterogeneous applications. In particular, we develop Application-Aware Cooperative Time allocation (AACT) framework, which optimizes the time that each application utilizes the aggregated system resources by taking into account heterogeneous device constraints and application requirements. AACT is grounded on the concept of Rolling Horizon Control (RHC) where decisions are made by iteratively solving a convex optimization problem over a moving control window of estimated system parameters. The simulation results demonstrate significant performance gains.

Consider a scenario in which a source broadcasts a common content to a group of cooperating mobile devices that are within proximity of each other. Devices in this group may receive only partial content from the source due to packet losses over wireless broadcast links and these packet losses may differ for different devices. The remaining content missing at each device can then be recovered, thanks to cooperation among the devices by exploiting device-to-device (D2D) connections. In this context, the minimum amount of time that guarantees a complete acquisition of the common content at every device is referred to as the “completion time”. It has been shown that instantly decodable network coding (IDNC) reduces the completion time as compared with no network coding in this scenario. However, for applications such as video streaming, not all packets have the same importance and not all devices are interested in the same quality of content. This problem becomes more interesting and challenging when additional, but realistic constraints, such as strict deadline, bandwidth, or limited energy are added in the problem formulation. We assert that direct application of IDNC in such a scenario yields poor performance in terms of content quality and completion time. In this paper, we propose a novel Content- and Loss-Aware IDNC scheme that improves content quality and network coding opportunities jointly by taking into account the contribution of each packet to the desired quality of service (QoS) as well as the channel losses over D2D links. Our proposed Content- and Loss-Aware IDNC (i) maximizes the quality under the completion time constraint, and (ii) minimizes the completion time under the quality constraint. We demonstrate the benefits of Content- and Loss-Aware IDNC through simulations.

The ACM CoNEXT 2016 Student Workshop is held in Irvine, California, USA on December 12, 2016 and co-located with the ACM 12th International Conference on emerging Networking Experiments and Technologies (CoNEXT 2016). The main objective of the workshop is to provide a platform for graduate students in the area of computer networks and communications to present their ongoing research efforts. The workshop is also a unique opportunity for students to network with other junior researchers as well as more experienced ones, receive constructive feedback, guidance, tips, and learn about cutting-edge research problems being tackled by the community. We have constructed an exciting program of 24 refereed presentations and posters, and an invited keynote talk (by Prof. Lixia Zhang, UCLA) that will give participating students an opportunity to hear from a senior researcher.

In this work, we explore the performance of backpressure routing and scheduling for TCP flows over wireless networks. TCP and backpressure are not compatible due to a mismatch between the congestion control mechanism of TCP and the queue size based routing and scheduling of the backpressure framework. We propose a TCP-aware backpressure routing and scheduling mechanism that takes into account the behavior of TCP flows. TCP-aware backpressure provides throughput optimality guarantees in the Lyapunov optimization framework, and gracefully combines TCP and backpressure without making any changes to the TCP protocol. The simulation results show that TCP-aware backpressure (i) improves the throughput of TCP flows significantly, (ii) provides fairness across competing TCP flows, and (iii) accommodates both TCP and non-TCP flows in a wireless network, and improves throughput of these flows without hurting fairness.

We consider a group of mobile users, within proximity of each other, who are interested in watching the same online video. The common practice today is that each user downloads the video independently on her mobile device using her own cellular connection, which wastes access bandwidth and may also lead to poor video quality. We propose a novel cooperative system where each mobile device uses simultaneously two network interfaces: (i) cellular to connect to the video server and download parts of the video and (ii) WiFi to connect locally to all other devices in the group to exchange those parts. Devices cooperate to efficiently utilize all network resources and to adapt to varying wireless network conditions. In the local WiFi network, we exploit overhearing, which we further combine with network coding. The end result is savings in cellular bandwidth and improved user experience. We follow a complete approach, from theory to practice. First, we formulate the problem using a network utility maximization (NUM) framework, decompose the problem, and provide a distributed solution. Then, based on the structure of the NUM solution, we design a system called MicroCast, and we implement a prototype as an Android application. We provide both simulation results of the NUM solution and experimental evaluation. We demonstrate that the proposed approach brings significant performance benefits (namely, faster download on the order of the group size) without battery penalty.

Backpressure routing and scheduling, with its throughput-optimal operation guarantee, is a promising technique to improve throughput in wireless multihop networks. Although backpressure is conceptually viewed as layered, the decisions of routing and scheduling are made jointly, which imposes several challenges in practice. In this work, we present Diff-Max, an approach that separates routing and scheduling and has three strengths: 1) Diff-Max improves throughput significantly; 2) the separation of routing and scheduling makes practical implementation easier by minimizing cross-layer operations; i.e., routing is implemented in the network layer and scheduling is implemented in the link layer; and 3) the separation of routing and scheduling leads to modularity; i.e., routing and scheduling are independent modules in Diff-Max, and one can continue to operate even if the other does not. Our approach is grounded in a network utility maximization (NUM) formulation and its solution. Based on the structure of Diff-Max, we propose two practical schemes: Diff-subMax and wDiff-subMax. We demonstrate the benefits of our schemes through simulation in ns-2.

In this paper, we are interested in improving the performance of TCP flows over wireless networks with a given constructive intersession network coding scheme. We are motivated by the observation that TCP does not fully exploit the potential of the underlying network coding opportunities. In order to improve the performance of TCP flows over coded wireless networks, without introducing changes to TCP itself, we propose a network-coding aware queue management scheme (NCAQM) that is implemented at intermediate network coding nodes and bridges the gap between network coding and TCP rate control. The design of NCAQM is grounded on the network utility maximization (NUM) framework and includes the following mechanisms. NCAQM: 1) stores coded packets at intermediate nodes in order to use the buffer space more efficiently; 2) determines what fraction of the flows should be coded together; and 3) drops packets at intermediate nodes so that it matches the rates of parts of different TCP flows that are coded together. We demonstrate, via simulation, that NCAQM significantly improves TCP throughput compared to TCP over baseline queue management schemes.

Backpressure routing and scheduling, with throughput-optimal operation guarantee, is a promising technique to improve throughput in wireless multi-hop networks. Although backpressure is conceptually viewed as layered, the decisions of routing and scheduling are made jointly, which imposes several challenges in practice. In this work, we present Diff-Max, an approach that separates routing and scheduling and has three strengths: (i) Diff-Max improves throughput significantly, (ii) the separation of routing and scheduling makes practical implementation easier by minimizing cross-layer operations; i.e., routing is implemented in the network layer and scheduling is implemented in the link layer, and (iii) the separation of routing and scheduling leads to modularity; i.e., routing and scheduling are independent modules in Diff-Max, and one can continue to operate even if the other does not. Our approach is grounded in a network utility maximization (NUM) formulation and its solution. Based on the structure of Diff-Max, we propose two practical schemes: Diff-subMax and wDiff-subMax. We demonstrate the benefits of our schemes through simulation in ns-2.

Video streaming is one of the increasingly popular, as well as demanding, applications on smartphones today. In this paper, we consider a group of smartphone users, within proximity of each other, who are interested in watching the same video from the Internet at the same time. The common practice today is that each user downloads the video independently using her own cellular connection, which often leads to poor quality.

In this work, we are interested in a scenario where a group of smartphone users, within proximity of each other, are interested in watching the same video at the same time. The default operation today is that each user with a cellular connection downloads the video independently from the server. However, each phone's individual cellular connection may not be sufficient for providing high video quality.

In this work, we are interested in improving the performance of constructive network coding schemes in lossy wireless environments. We propose I2NC - an approach that combines inter-session and intra-session network coding and has two strengths. First, the error-correcting capabilities of intra-session network coding make our scheme resilient to loss. Second, redundancy allows intermediate nodes to operate without knowledge of the decoding buffers of their neighbors. Based only on the knowledge of the loss rates on the direct and overhearing links, intermediate nodes can make decisions for both intra-session (i.e., how much redundancy to add in each flow) and inter-session (i.e., what percentage of flows to code together) coding. Our approach is grounded on a network utility maximization (NUM) formulation of the problem. We propose two practical schemes, I2NC-state and I2NC-stateless, which mimic the structure of the NUM optimal solution. We also address the interaction of our approach with the transport layer. We demonstrate the benefits of our schemes through simulation in GloMoSim.

In this paper, we study delay-optimized network coding for video streaming over wireless networks with network coding capabilities. It has been demonstrated that network coding can increase throughput in wireless networks, by mixing packets from different flows into a single packet, thus increasing the information content per transmission. However, network coding potential is not always fully exploited in this setting, because there may not be enough packets at intermediate nodes to do network coding due to bursty nature of transport protocols, packet losses, and difference in path delays. One way to deal with this problem is to delay packets at intermediate nodes in order to create more network coding opportunities. However, introducing large or varying delays eventually hurts video traffic, which requires low delay and delay jitter. In this paper, we study this tradeoff in coded wireless networks, and we propose a packet delaying scheme at intermediate nodes to maximize video quality.

Media flows coexist with TCP-based data traffic on the Internet and are required to be TCP-friendly. The TCP protocol slowly increases its sending rate until episodes of congestion occur, and then it quickly reduces its rate to remove congestion. However, media flows can be sensitive to even brief episodes of congestion. In this paper, we are interested in protecting media flows from TCP-induced congestion while maintaining their TCP friendliness. In particular, we consider media flows carried over the TCP-Friendly Rate Control (TFRC) protocol and we design algorithms that dynamically adapt the level of forward error correction (FEC) based on the congestion state of the network. To this end, first, we investigate the loss and delay characteristics of TFRC flows in several TCP-induced congestion scenarios, and we develop novel predictors of loss events based on packet delay information. Second, we use these predictors to dynamically adapt the level of FEC protection based on the predicted level of congestion. We show that this technique can significantly improve the overhead versus reliability trade-off compared to fixed FEC. Third, we select the FEC and original media packets within each FEC block, in a rate-distortion optimized way, and we show that this technique significantly improves media quality.

In this paper, we study rate control and scheduling over wireless networks with intersession network coding, as a utility maximization problem. We demonstrate that making rate control and scheduling aware of the underlying network coding increases throughput. The key intuition is that network coding introduces new network coded flows and eventually new conflicts between nodes, which should be taken into account both in rate control and in scheduling. We compare the network coding-aware to the network coding-unaware schemes in two cases: (i) optimal control and (ii) practical, suboptimal control. Our main goal is to make the case for network coding-aware rate control and scheduling, via simulation of representative examples. Along the way, we also propose a practical scheme that approximates the optimal control.

In this paper, we study video streaming over wireless networks with network coding capabilities. We build upon recent work, which demonstrated that network coding can increase throughput over a broadcast medium, by mixing packets from different flows into a single packet, thus increasing the information content per transmission. Our key insight is that, when the transmitted flows are video streams, network codes should be selected so as to maximize not only the network throughput but also the video quality. We propose video-aware opportunistic network coding schemes that take into account both the decodability of network codes by several receivers and the importance and deadlines of video packets. Simulation results show that our schemes significantly improve both video quality and throughput. This work is a first step towards content-aware network coding.

This paper considers the problem of multiple video streaming over wireless local area networks. In particular, we propose video streaming techniques to improve the quality of video streams over the link from an access point to multiple wireless clients. The proposed video streaming techniques decide on clients to be served, packets to be (re)transmitted, and forward error correction (FEC) rates by considering video, channel, and network properties in a rate-distortion optimized manner. Multi-client rate distortion optimized video streaming is the first technique operated by an access point to jointly optimize client, packet, and FEC rate selection for video streaming. Second, we propose multi-client sub-optimal video streaming technique that optimizes packet and FEC rate, but not client. The client is selected by using the results of optimization to reduce the computational complexity. The simulation results show that the proposed video streaming techniques achieve signal-to-noise ratio (SNR) improvement up-to 5 dB as compared with time-division multiple access (TDMA) and packet adaptive TDMA.

In this paper, we investigate rate distortion optimized streaming of H.264 coded video sequences over time-varying wireless channels. Our main objective is to minimize average end-to-end distortion to satisfy a certain quality of service (QoS) by considering the media and channel characteristics such as packet importance, packet dependencies, decoding deadlines, channel state information and channel capacity. In particular, we propose a sender-driven optimized joint ARQ-FEC scheme that decides on packet transmissions, re-transmissions as well as the FEC rate used in each transmission. The optimized joint ARQ-FEC scheme along with packet scheduling aims to minimize a weighted sum of distortion and total transmission cost under capacity constraints. The other objective in this paper is to satisfy the QoS of the optimized joint ARQ-FEC scheme without minimizing end-to-end distortion. We improved two approximate algorithms that use the media and channel characteristics to transmit the best selected and FEC coded packet. The real-time simulation results with H.264 sequences demonstrate the efficacy of the all proposed algorithms over the classical error recovery scheme uses ARQ and FEC.