Deepfakes have emerged as a significant threat to online communications, enabling nearly indistinguishable impersonation of executives, public figures, and trusted contacts during video calls. While state-of-the-art deepfake detection models can achieve high accuracy offline, deploying them in real-time video conferencing systems remains challenging: the added computation quickly violates interactive latency budgets and greatly limits scalability. Our empirical analysis reveals that video decoding and frame movement dominate the detection pipeline, together accounting for approximately 86.6% of per-frame processing time.

Mobile devices run deep learning models for various purposes, such as image classification and speech recognition. Due to the resource constraints of mobile devices, researchers have focused on either making a lightweight deep neural network (DNN) model using model pruning or generating an efficient code using compiler optimization. It was observed that the straightforward integration between model compression and compiler auto-tuning often fails to produce the most efficient model for a target device. We propose CPrune, a compiler-informed model pruning for efficient target-aware DNN execution to support an application with a required target accuracy. To address real-world deployment scenarios with resource or latency constraints, we further introduce RB-CPrune, a predictive variant that eliminates iterative tuning by using a learned latency estimator. CPrune makes a lightweight DNN model through informed pruning based on the structural information of subgraphs built during the compiler tuning process. Our experimental results show that CPrune increases the DNN execution speed up to 2.73× compared to the state-of-the-art TVM auto-tune while meeting the accuracy requirement.

Reference signals from cellular networks present an untapped and abundant signal of opportunity for high quality radio frequency (RF) propagation measurements. Commercial-off-the-shelf mobile phones continuously capture and report Reference Signal Received Power (RSRP) measurements, making them an easily crowdsource-able data source for RF propagation modeling in path geometries and frequencies relevant to cellular communications, broadcast, and short-range outdoor communications systems. However, it remains unclear whether crowdsourced mobile phone RSRP measurements can meet the stringent accuracy requirements of measured propagation data in support of RF propagation model validation and improvement.

Live volumetric video streaming enables immersive user experiences but poses significant challenges due to the high bandwidth requirement that 3D representations entail. Recent research has focused on reducing volumetric video bandwidth, but it struggles to effectively address temporal redundancy under real-time constraints, limiting its applicability for live streaming scenarios. To address these challenges, we present DeltaStream, a novel live volumetric video streaming system that efficiently encodes 3D point clouds by leveraging 2D information. By utilizing 2D RGB and depth frames, DeltaStream efficiently infers inter-frame changes to reduce the streaming bandwidth of volumetric video. Furthermore, DeltaStream introduces an adaptive block-based approach that can reduce the client-side decoding load. Through extensive evaluations, our results demonstrate that DeltaStream reduces bandwidth by up to 71% with 1.63× faster decoding speed while maintaining visual quality compared to state-of-the-art systems.

On-device deep neural network (DNN) inference is often desirable for user experience and privacy. Existing solutions have fully utilized resources to minimize inference latency. However, they result in severe energy inefficiency by completing DNN inference much earlier than the required service interval. It poses a new challenge of how to make DNN inferences in a punctual and energy-efficient manner. To tackle this challenge, we propose a new resource allocation strategy for DNN processing, namelypunctual lazinessthat disperses its workload as efficiently as possible over time within its strict delay constraint. This strategy is particularly beneficial for neural workloads since a DNN comprises a set of popular operators whose latency and energy consumption are predictable. Through this understanding, we propose NeuroBalancer, an operator-aware core and memory frequency scaling framework that balances those frequencies as efficiently as possible while making timely inferences. We implement and evaluate NeuroBalancer on off-the-shelf Android devices with various state-of-the-art DNN models. Our results show that NeuroBalancer successfully meets a given inference latency requirements while saving energy consumption up to 43.9% and 21.1% compared to the Android's default governor and up to 42.1% and 18.6% compared to SysScale, the state-of-the-art mobile governor on CPU and GPU, respectively.

Offloading neural network inferences from resource-constrained mobile devices to an edge server over wireless networks is becoming more crucial as the neural networks get heavier. To this end, recent studies have tried to make this offloading process more efficient. However, the most fundamental question on extracting and offloading the minimal amount of necessary information that does not degrade the inference accuracy has remained unanswered. We call such an ideal offloading semantic offloading and propose N-epitomizer, a new offloading framework that enables semantic offloading, thus achieving more reliable and timely inferences in highly-fluctuated or even low-bandwidth wireless networks. To realize N-epitomizer, we design an autoencoder-based scalable encoder trained to extract the most informative data and scale its output size to meet the latency and accuracy requirements of inferences over a network. We also accelerate N-epitomizer by exploiting light-weight knowledge distillation for the encoder design and decoder slimming for the decoder design, reducing its overall computation time significantly. Moreover, we extend our N-epitomizer to support multiple DNNs by extracting and offloading the union of the essential information required for each DNN. Our evaluation shows that N-epitomizer achieves exceptionally high compression for images without compromising inference accuracy, which is 21$\times$, 77$\times$, and 192$\times$higher than JPEG compression, and 20$\times$, 55$\times$, and 86$\times$higher than the state-of-the-art DNN-aware image compression GRACE for semantic segmentation, depth estimation, and classification, respectively. Our results show N-epitomizer’s strong potential as the first semantic offloading system to guarantee end-to-end latency even under highly varying cellular networks.

Network fluctuations can cause unpredictable degradation of the user’s quality of experience (QoE) on real-time video streaming. The intrinsic property of real-time video streaming, which generates delay-sensitive and chunk-based video frames, makes the situation even more complicated. Although previous approaches have tried to alleviate this problem by controlling the video bitrate based on the current network capacity estimate, they do not take into account the explicit queueing delay experienced by the video frame in determining the bitrate of upcoming video frames. To tackle this problem, we propose a new real-time video streaming system, Exstream, that can adapt to dynamic network conditions with the help of video bitrate control method and bandwidth estimation method designed to support real-time video streaming environments. Exstream explicitly estimates the queueing delay experienced by the video frame based on the transmission time budget that each frame can maximally utilize, which depends on the frame generation interval, and adjusts the bitrate of newly generated video frames to suppress the queueing delay level close to zero. Our comprehensive experiments demonstrate that Exstream achieves lower frame delay than four existing systems, Salsify, WebRTC, Skype, and Hangouts without frequent video frame skip.

In this paper, we conduct a measurement study on operational 5G networks deployed across different frequency bands (mmWave and sub-6GHz) and server locations (mobile edge and Internet cloud). Specifically, we assess 5G performance in both uplink and downlink across multiple operators’ networks. We then carry out extensive comparisons of transport-layer protocols using ten different algorithms in full-fledged 5G networks, including an edge computing environment. Finally, we evaluate representative mobile applications over the 5G network with and without edge servers. Our comprehensive measurements provide several insights that affect the experience of 5G users: (i) With a 5G edge server, existing TCP congestion control algorithms can achieve throughput up to 1.8Gbps with only a single flow. (ii) The maximum TCP receive buffer size, which is set by off-the-shelf 5G phones, can limit the throughput performance of 5G networks, which is not observed in 4G LTE-A networks. (iii) Despite significant latency gains in download-centric applications, the 5G edge service provides limited benefits to CPU-intensive tasks or those that use significant uplink bandwidth. To our knowledge, this is the first measurement-driven understanding of 5G edge computing “in the wild,” which can provide an answer to how edge computing would perform in real 5G networks.

A user's movement path can be precisely and concisely described as a concatenation of straight lines having the user's turns as their end points. Learning such a path description or representation from inertial measurement unit (IMU) sensors enables various mobile and IoT applications, as it allows efficient processing of the movement path data. It is, however, non-trivial to learn a succinct yet accurate path description from IMU sensor readings in the mobile device of a moving useron the flydue to the dynamically changing behaviors and the technical difficulty in detecting the user's turns. We propose PATHLIT, a novel online path description learning system based on IMU signals. PATHLIT learns position vectors of a user from IMU sensor readings by our custom-made self-attention network model. Once each position vector is learned, PATHLIT also decides whether or not to take it as a part of the resulting path description by our efficient online algorithm developed under the minimum description length principle, which essentially detects the user's turns along the path. We conduct extensive experiments on two large datasets. The experiment results show that PATHLIT achieves superior performance over state-of-the-art algorithms by up to 50% in absolute trajectory error using only 15% of trajectory data points.

As datacenters become the new computing platform, integrating server-centric distributed systems into modern network hardware is gaining interest under the diminishing Moore's law. Researchers want to build reusable primitives that can take advantage of modern network hardware and offload common system components of a broad range of applications. In this paper, we present Mostly Reliable Totally Ordered Multicast, a reusable network primitive that can embed reliable group communication into the network. MRTOM is a network-centric approach that handles message replication, ordering, and reliable delivery using a network fast path, freeing server CPUs for application logic. MRTOM can be implemented in the programmable switches and edge interfaces (e.g., Smart-NICs), significantly reducing network traffic compared to the existing approaches and improving job finish time amid packet loss. With MRTOM, we were able to accelerate multiple high-performance applications whose fast path can be totally offloaded into the network. For example, a Paxos application, MRTOM-Paxos, achieves > 1,100,000 transactions/secs and$23\mu \mathrm{s}$minimum latency. A replicated key-value store, MRTOM-KV, also shows significant latency reduction with eBPF/XDP in the Linux Kernel, which is further improved by offloading into SmartNICs.

Floor labels of crowdsourced RF signals are crucial for many smart-city applications, such as multi-floor indoor localization, geofencing, and robot surveillance. To build a prediction model to identify the floor number of a new RF signal upon its measurement, conventional approaches using the crowdsourced RF signals assume that at least few labeled signal samples are available on each floor. In this work, we push the envelope further and demonstrate that it is technically feasible to enable such floor identification with only one floor-labeled signal sample on the bottom floor while having the rest of signal samples unlabeled. We propose FIS-ONE, a novel floor identification system with only one labeled sample. FIS-ONE consists of two steps, namely signal clustering and cluster indexing. We first build a bipartite graph to model the RF signal samples and obtain a latent representation of each node (each signal sample) using our attention-based graph neural network model so that the RF signal samples can be clustered more accurately. Then, we tackle the problem of indexing the clusters with proper floor labels, by leveraging the observation that signals from an access point can be detected on different floors, i.e., signal spillover. Specifically, we formulate a cluster indexing problem as a combinatorial optimization problem and show that it is equivalent to solving a traveling salesman problem, whose (near-)optimal solution can be found efficiently. We have implemented FIS-ONE and validated its effectiveness on the Microsoft dataset and in three large shopping malls. Our results show that FIS- ONE outperforms other baseline algorithms significantly, with up to 23 % improvement in adjusted rand index and 25% improvement in normalized mutual information using only one floor-labeled signal sample.

In applications such as elderly care, dementia anti-wandering and pandemic control, it is important to ensure that people are within a predefined area for their safety and well-being. We propose GEM, a practical, semi-supervised Geofencing system with network EMbedding, which is based only on ambient radio frequency (RF) signals. GEM models measured RF signal records as a weighted bipartite graph. With access points on one side and signal records on the other, it is able to precisely capture the relationships between signal records. GEM then learns node embeddings from the graph via a novel bipartite network embedding algorithm called BiSAGE, based on a Bipartite graph neural network with a novel bi-level SAmple and aggreGatE mechanism and non-uniform neighborhood sampling. Using the learned embeddings, GEM finally builds a one-class classification model via an enhanced histogram-based algorithm for in-out detection, i.e., to detect whether the user is inside the area or not. This model also keeps on improving with newly collected signal records. We demonstrate through extensive experiments in diverse environments that GEM shows state-of-the-art performance with up to 34% improvement in F-score. BiSAGE in GEM leads to a 54% improvement in F-score, as compared to the one without BiSAGE.

Age of Information (AoI) is a new metric that indicates the freshness of information; especially, it is used to design schedulers in 5G networks. Recently, various AoI schedulers have been proposed to guarantee AoI deadlines for emerging applications in which the timeliness of information is critical. However, these schedulers face challenges in meeting the deadlines for all source nodes in situations wherein system resources are limited. To overcome this limitation, a violation-tolerant scheduler has been proposed; however, it does not consider the variance of violations over consecutive data, thereby resulting in significant information deficiencies over long periods of time. In this study, we consider the problem of scheduling a stream with ($m, k$)-firm deadlines, wherein at least$m$of any$k$consecutive data points must satisfy their deadlines. To address this problem, we propose a dynamic priority assignment technique and a priority-based physical resource block allocation algorithm. In addition, we introduce an open radio access network-based system model to enable practical deployment of the proposed solution. We built a 5G simulation environment using ns-3 and captured performance variations under realistic deployment scenarios.

With the rapid advances of deep learning and the commercialization of high-definition cameras in mobile and embedded devices, the demands from latency-critical applications such as AR and XR for high-quality video analytics (HVA) are soaring. By the nature of HVA aiming at enabling detailed analytics even for small objects, its on-device implementation is suffering from thermal and battery issues, which makes offloaded HVA an attractive solution. This work provides unique observations on the tradeoff pertaining to offloaded HVA: the frame encoding time, the frame transmission time, and the HVA accuracy. Our observations pose a fundamental question: given a latency budget, how to choose the encoding option that properly combines between the encoding time and the transmission time to maximize the HVA accuracy. To answer this question, we propose an offloaded HVA system, ENTRO, which exploits this tradeoff in real-time to maximize the HVA accuracy under the latency budget. Our extensive evaluations with ENTRO implemented on Nvidia AGX Xavier and Samsung Galaxy S20 Ultra over WiFi networks show 8.8× improvement in latency without accuracy loss compared to DDS, the state-of-the-art offloaded video analytics. Our evaluation over commercial 5G and LTE networks also indicates that ENTRO flexibly adapts its encoding option under the tradeoff and enables the latency-bounded HVA with 4K frames.

Modern Deep Neural Network (DNN) frameworks use tensor operators as the main building blocks of DNNs. However, we observe that operator-based construction of DNNs incurs significant drawbacks in parallelism in the form of synchronization barriers. Synchronization barriers of operators confine the scope of parallel computation to each operator and obscure the rich parallel computation opportunities that exist across operators. To this end, we present ASPEN, a novel parallel computation solution for DNNs that achieves fine-grained dynamic execution of DNNs, which (1) removes the operator barriers and expresses DNNs in dataflow graphs of fine-grained tiles to expose the parallel computation opportunities across operators, and (2) exploits these opportunities by dynamically locating and scheduling them in runtime. This novel approach of ASPEN enables opportunistic parallelism, a new class of parallelism for DNNs that is unavailable in the existing operator-based approaches. ASPEN also achieves high resource utilization and memory reuse by letting each resource asynchronously traverse depthwise in the DNN graph to its full computing potential. We provide challenges and solutions to our approach and show that our proof-of-concept implementation of ASPEN on CPU shows exceptional performance, outperforming state-of-the-art inference systems of TorchScript and TVM by up to 3.2$\times$ and 4.3$\times$, respectively.

Many studies have applied machine learning to bitrate control to increase Quality of Experience (QoE) of video streaming services in highly dynamic networks. However, their solutions mainly focused on HTTP adaptive streaming with one-to-one connections. This paper studies video conferencing applications where multi-party, full-duplex communication happens among participants. In particular, we propose Muno, a Deep Reinforcement Learning (DRL)-based bandwidth prediction framework for multi-party video conferencing systems. Muno learns and predicts an appropriate bitrate for each connection in a multi-party conferencing call. We trained Muno to maximize the QoE of individual connections by constructing a feedback loop between a media server and DRL servers. Our experimental results show that Muno achieves a higher video streaming rate and lower delay compared to state-of-the-art rulebased algorithms.

Video conferencing has become a daily necessity, but protocols to support video conferencing have yet to keep pace despite the innovation in next-generation networks. As video resolutions increase and mobile applications using multiple cameras for photos and videos become popular, the need to meet the Quality of Experience (QoE) requirements is growing. Multipath protocols could be a possible solution.

Age of information (AoI) is used to evaluate the performance of 5G health monitoring systems because stale data can be fatal for patients with serious illness. Recently, grant-free ultrareliable and low latency communications (URLLC) have shown greater potential of minimizing AoI than conventional grant-based approaches; however, existing grant-free schedulers cannot provide guaranteed performance in 5G health monitoring systems because they involve two fundamental problems in time and frequency domains, namely the joint scheduling problem and physical resource block (PRB) allocation. In this study, we investigate two resource allocation problems for the first time, aiming to enable grant-free URLLC to minimize AoI in 5G health monitoring systems. Specifically, we propose two adaptive solutions based on an open radio access network-coordinated wireless system: 1) a joint scheduling algorithm and 2) an adaptive PRB allocation algorithm. To verify the effectiveness of the proposed solutions, we built a simulation environment similar to a real health monitoring system and captured the performance variations under realistic deployment scenarios.

Edge computing capabilities in 5G wireless networks promise to benefit mobile users: computing tasks can be offloaded from user devices to nearby edge servers, reducing users’ experienced latencies. Few works have addressed how this offloading should handle long-term user mobility: as devices move, they will need to offload to different edge servers, which may require migrating data or state information from one edge server to another. In this paper, we introduce MoDEMS, a system model and architecture that provides a rigorous theoretical framework and studies the challenges of such migrations to minimize the service provider cost and user latency. We show that this cost minimization problem can be expressed as an integer linear programming problem, which is hard to solve due to resource constraints at the servers and unknown user mobility patterns. We show that finding the optimal migration plan is in general NP-hard, and we propose alternative heuristic solution algorithms that perform well in both theory and practice. We finally validate our results with real user mobility traces, ns-3 simulations, and an LTE testbed experiment. Migrations reduce the latency experienced by users of edge applications by 33% compared to previously proposed migration approaches.

Today’s Internet must support applications with increasingly dynamic and heterogeneous connectivity requirements, such as video streaming and the Internet of Things. Yet current network management practices generally rely on pre-specified network configurations, which may not be able to cope with dynamic application needs. Moreover, even the best-specified policies will find it difficult to cover all possible scenarios, given applications’ increasing heterogeneity and dynamic network conditions, e.g., on volatile wireless links. In this work, we instead propose a model-free learning approach to find the optimal network policies for current network flow requirements. This approach is attractive as comprehensive models do not exist for how different policy choices affect flow performance under changing network conditions. However, it can raise new challenges for online learning algorithms: policy configurations can affect the performance of multiple flows sharing the same network resources, and this performance coupling limits the scalability and optimality of existing online learning algorithms. In this work, we extend multi-armed bandit frameworks to propose new online learning algorithms for protocol selection with provably sublinear regret under certain conditions. We validate the optimality and scalability of our algorithms through data-driven simulations and testbed experiments. (An extended abstract of this work was accepted by IEEE ICNP as a short paper Zhanget al. (2019)).

As convolution has empowered many smart applications, dynamic convolution further equips it with the ability to adapt to diverse inputs. However, the static and dynamic convolutions are either layout-agnostic or computation-heavy, making it inappropriate for layout-specific applications, e.g., face recognition and medical image segmentation. We observe that these applications naturally exhibit the characteristics of large intra-image (spatial) variance and small cross-image variance. This observation motivates our efficient translation variant convolution (TVConv) for layout-aware visual processing. Technically, TVConv is composed of affinity maps and a weight-generating block. While affinity maps depict pixel-paired relationships gracefully, the weight-generating block can be explicitly over-parameterized for better training while maintaining efficient inference. Although conceptually simple, TVConv significantly improves the efficiency of the convolution and can be readily plugged into various network architectures. Extensive experiments on face recognition show that TVConv reduces the computational cost by up to 3.1 × and improves the corresponding throughput by 2.3× while maintaining a high accuracy compared to the depthwise convolution. Moreover, for the same computation cost, we boost the mean accuracy by up to 4.21%. We also conduct experiments on the optic disc/cup segmentation task and obtain better generalization performance, which helps mitigate the critical data scarcity issue. Code is available at https://github.com/JierunChen/TVConv.

We study the problem of floor identification for radiofrequency (RF) signal samples obtained in a crowdsourced manner, where the signal samples are highly heterogeneous and most samples lack their floor labels. We propose GRAFICS, a graph embedding-based floor identification system. GRAFICS first builds a highly versatile bipartite graph model, having APs on one side and signal samples on the other. GRAFICS then learns the low-dimensional embeddings of signal samples via a novel graph embedding algorithm named E-LINE. GRAFICS finally clusters the node embeddings along with the embeddings of a few labeled samples through a proximity-based hierarchical clustering, which eases the floor identification of every new sample. We validate the effectiveness of GRAFICS based on two large-scale datasets that contain RF signal records from 204 buildings in Hangzhou, China, and five buildings in Hong Kong. Our experiment results show that GRAFICS achieves highly accurate prediction performance with only a few labeled samples (96% in both micro- and macro-F scores) and significantly outperforms several state-of-the-art algorithms (by about 45% improvement in micro-F score and 53% in macro-F score).

Edge computing capabilities in 5G wireless networks promise to benefit mobile users: computing tasks can be offloaded from user devices to nearby edge servers, reducing users’ experienced latencies. Few works have addressed how this offloading should handle long-term user mobility: as devices move, they will need to offload to different edge servers, which may require migrating data or state information from one edge server to another. In this paper, we introduce MoDEMS, a system model and architecture that provides a rigorous theoretical framework and studies the challenges of such migrations to minimize the service provider cost and user latency. We show that this cost minimization problem can be expressed as an integer linear programming problem, which is hard to solve due to resource constraints at the servers and unknown user mobility patterns. We show that finding the optimal migration plan is in general NP-hard, and we propose alternative heuristic solution algorithms that perform well in both theory and practice. We finally validate our results with real user mobility traces, ns-3 simulations, and an LTE testbed experiment. Migrations reduce the latency experienced by users of edge applications by 33% compared to previously proposed migration approaches.

The demand for video conferencing applications has seen explosive growth while users still often face unsatisfactory quality of experience (QoE). Video conferencing applications adopt Forward Error Correction (FEC) as a recovery mechanism to meet tight latency requirements and overcome packet losses prevalent in the network. However, many studies mainly focused on video rate control by neglecting the complex interactions of this video recovery mechanism on the rate control and its impact on the user QoE. Deciding the right amount of FEC for the current video rate under a dynamically changing network environment is not straightforward. For instance, the higher FEC may enhance the tolerance to packet losses, but it may increase latency due to FEC processing overhead and hurt the video quality due to the additional bandwidth used for FEC. To address this issue, we propose R-FEC which is a reinforcement learning (RL) based framework for video and FEC bitrate decisions in video conferencing. R-FEC aims to improve overall QoE by automatically learning through the results of past decisions and adjusting video and FEC bitrates to maximize the user QoE while minimizing the congestion in the network. Our experiments show that R-FEC outperforms the state-of-the-art solutions in video conferencing, with up to 27% improvement in its video rate and 6dB PSNR improvement in video quality over the default WebRTC.

Edge computing systems benefit from knowledge of short-term mobility from 5G technologies, as tasks offloaded from user devices can be placed at the edge to reduce their latencies. However, as devices move, they will need to offload their tasks to different edge servers, which may require migrating data from one edge server to another. In this paper, we introduce MoDEMS, a system architecture through which we provide a rigorous theoretical framework to study the challenges of such migrations to minimize the service provider cost and user latency. We show that this cost minimization problem can be expressed as an integer linear programming problem, which is challenging to solve due to resource constraints at the servers and unknown user mobility patterns. We show that finding the optimal migration plan is in general NP-hard, and we propose alternative heuristic solution algorithms. We finally validate our results with realistic user mobility traces.

Video conferencing applications have seen explosive growth both in the number of available applications and their use. However, there have been few studies on the detailed analysis of video conferencing applications with respect to network dynamics, yet understanding these dynamics is essential for network design and improving these applications. In this paper, we carry out an in-depth measurement and modeling study on the rate control algorithms used in six popular commercial video conferencing applications. Based on macroscopic behaviors commonly observed across these applications in our extensive measurements, we construct a unified architecture to model the rate control mechanisms of individual applications. We then reconstruct each application's rate control by inferring key parameters that closely follow its rate control and quality adaptation behaviors. To our knowledge, this is the first work that reverse-engineers rate control algorithms of popular video conferencing applications, which are often unknown or hidden as they are proprietary software. We confirm our analysis and models using an end-to-end testbed that can capture the dynamics of each application under a variety of network conditions. We also show how we can use these models to gain insights into the particular behaviors of an application in two practical scenarios.

DVFS (dynamic voltage and frequency scaling) is a system-level technique that adjusts voltage and frequency levels of CPU/GPU at runtime to balance energy efficiency and high performance. DVFS has been studied for many years, but it is considered still challenging to realize a DVFS that performs ideally for mobile devices for two main reasons: i) an optimal power budget distribution between CPU and GPU in a power-constrained platform can only be defined by the application performance, but conventional DVFS implementations are mostly application-agnostic; ii) mobile platforms experience dynamic thermal environments for many reasons such as mobility and holding methods, but conventional implementations are not adaptive enough to such environmental changes. In this work, we propose a deep reinforcement learning-based frequency scaling technique, zTT. zTT learns thermal environmental characteristics and jointly scales CPU and GPU frequencies to maximize the application performance in an energy-efficient manner while achieving zero thermal throttling. Our evaluations for zTT implemented on Google Pixel 3a and NVIDIA JETSON TX2 platform with various applications show that zTT can adapt quickly to changing thermal environments, consistently resulting in high application performance with energy efficiency. In a high-temperature environment where a rendering application with the default mobile DVFS fails to keep producing more than a target frame rate, zTT successfully manages to do so even with 23.9% less average power consumption.

The sixth generation of DOCSIS standard is currently under development for the provisioning of multi-Gbps services over cable networks. Building upon DOCSIS 3.1 (D3.1), DOCSIS 4.0 (D4) introduces several features including full-duplex transmission and extended-spectrum, which benefit from subcarrier-level OFDM modulation configurations to adapt to varying channel conditions. To exploit the full potential of D4, we propose a softwarized adaptive subcarrier modulation management framework. The optimization system consists of (i) a clustering mechanism that classifies CMs (Cable Modems) with a similar channel condition into the same group using a sparsified K-means algorithm and (ii) an efficient profile generation mechanism to balance achieved channel throughput and packet error rate within the same group. Then, we implement key elements of the softwarized system using a virtualized network function in our DOCSIS experimental testbed that enables the programmatic control of OFDM channels using D4 performance parameters. Our experimental results show that the proposed optimization function offers significant improvements in OFDM channel throughput over current industry management practices. Furthermore, we confirm via simulations that using a novel clustering algorithm for the classification of CM populations and a new bit-loading method measurably enhances channel performance in large-scale distributed deployment scenarios.

This paper presents a new approach to memory disaggregation called FluidMem that leverages the userfault mechanism in Linux to achieve full memory disaggregation in software. FluidMem enables dynamic and transparent resizing of an unmodified Virtual Machine’s (VM’s) memory footprint in the cloud. As a result, a VM’s memory footprint can seamlessly scale over multiple machines or even be downsized to a near-zero footprint on a given server. FluidMem’s architecture provides flexibility to cloud operators to manage remote memory without requiring guest intervention, while also supporting paging out the entirety of a VM’s pages within its address space. FluidMem integrates with a remote memory backend in a modular way, easily supporting systems such as RAMCloud to harness remote memory. We demonstrate FluidMem outperforms an existing memory disaggregation approach based on network swap. Microbenchmarks are evaluated to characterize the latency of different components of the FluidMem architecture, and two memory-intensive applications are demonstrated using FluidMem, the Graph500 benchmark, and MongoDB. Additionally, we show FluidMem can flexibly and efficiently grow and shrink the memory footprint of a VM as defined by a cloud provider.

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.

As video calls and personal broadcasting become popular, the demand for mobile live streaming over cellular uplink channels is growing fast. However, current live streaming solutions are known to suffer from frequent uplink throughput fluctuations causing unnecessary video stalls and quality drops. As a remedy to this problem, we propose PERCEIVE, a deep learning-based uplink throughput prediction framework. PERCEIVE exploits a 2-stage LSTM (Long Short Term Memory) design and makes throughput predictions for the next 100ms. Our extensive evaluations show that PERCEIVE, trained with LTE network traces from three major operators in the U.S., achieves high accuracy in the uplink throughput prediction with only 7.67% mean absolute error and outperforms existing prediction techniques. We integrate PERCEIVE with WebRTC, a popular video streaming platform from Google, as a rate adaptation module. Our implementation on the Android phone demonstrates that it can improve PSNR by up to 6dB (4x) over the default WebRTC while providing less streaming latency.

Emerging applications like virtual reality (VR), augmented reality (AR), and 360-degree video aim to exploit the unprecedentedly low latencies promised by technologies like the tactile Internet and mobile 5G networks. Yet these promises are still unrealized. In order to fulfill them, it is crucial to understand where packet delays happen, which impacts protocol performance such as throughput and latency. In this work, we empirically find that sender-side protocol stack delays can cause high end-to-end latencies, though existing solutions primarily address network delays. Unfortunately, however, current latency diagnosis tools cannot even distinguish between delays on network links and delays in the end hosts. To close this gap, we present ELEMENT, a latency diagnosis framework that decomposes end-to-end TCP latency into endhost and network delays, without requiring admin privileges at the sender or receiver.

Ensuring high availability for applications despite unpredictable cloud component failure events is a well-known problem in managing cloud infrastructure. One proposed solution uses a VM redundancy approach, reserving cloud resources for backup VMs that can substitute for primary ones in case of a failure event. However, this solution decreases the cloud resource utilization, since the backup resources usually remain idle. In this paper, we propose ECHO, a cloud resource management system that overbooks these backup VMs by optimizing the overbooking rate tradeoff between maximizing the cloud resource utilization, and thus maximizing the cloud provider's revenue; and improving application availability, thus satisfying users. Specifically, ECHO first obtains the optimal overbooking rate required to achieve a cloud provider's desired resource utilization level. It then computes the optimal (required) number of backup VMs that are required to maintain a given application availability level. Our extensive experimental and simulation results show that using ECHO can increase the number of accepted applications with satisfied availability by about 30%, while increasing the defined resource utilization at the same time.

Today’s Internet must support applications with increasingly dynamic and heterogeneous connectivity requirements, such as video streaming and the Internet of Things. Yet current network management practices generally rely on pre-specified flow configurations, which cannot cover all possible scenarios. In this work, we instead propose a model-free learning approach to automatically optimize the policies for heterogeneous network flows. This approach is attractive as no existing comprehensive models quantify how different policy choices affect flow performance under dynamically changing network conditions. We extend multi-armed bandit frameworks to propose new online learning algorithms for protocol selection, addressing the challenge of policy configurations affecting the performance of multiple flows sharing the same network resources. This performance coupling limits the scalability and optimality of existing online learning algorithms. We theoretically prove that our algorithm achieves a sublinear regret and demonstrate its optimality and scalability through data-driven simulations.

4G LTE networks across the world (e.g., United States, Europe, and South Korea) use the same mechanism to broadcast emergency alerts. These alerts include AMBER, severe weather alerts, and the (unblockable) Presidential Alert in the US. We demonstrate the ability to spoof these alerts by forcing any 4G phone in the area of our malicious cell tower to receive and display a fabricated message. This demonstration uses a commercially-available software-defined radio, an LTE base station, and our modifications to the open-source NextEPC and srsLTE libraries to send the Presidential Alert to phones volunteered from the audience.

This paper presents a fully distributed scheduling framework called CASTLE (Client-side Adaptive Scheduler That minimizes Load and Energy), which jointly optimizes the spectral efficiency of cellular networks and battery consumption of smart devices. To do so, we focus on scenarios when many smart devices compete for cellular resources in the same base station: spreading out transmissions over time so that only a few devices transmit at once improves both spectral efficiency and battery consumption. To this end, we devise two novel features in CASTLE. First, we explicitly consider inter-cell interference for accurate cellular load estimation. Based on our observations, we exploit the RSRQ (Reference Signal Received Quality) and SINR as features in a machine learning algorithm to accurately estimate the cellular load. Second, we propose a fully distributed scheduling algorithm that coordinates transmissions between clients based on the locally estimated load level at each client. Our formulation for minimizing battery consumption at each device leads to an optimized backoff-based algorithm that fits practical environments. To evaluate these features, we prototype a complete LTE system testbed consisting of mobile devices, eNodeBs, EPC (Evolved Packet Core) and application servers. Our comprehensive experimental results show that CASTLE's load estimation is up to 91% accurate, and that CASTLE achieves higher spectral efficiency with less battery consumption, compared to existing centralized scheduling algorithms as well as a distributed CSMA-like protocol. Furthermore, we develop a light-weight SDK that can expedite the deployment of CASTLE into smart devices and evaluate it in a commercial LTE network.

Modern cell phones are required to receive and display alerts via the Wireless Emergency Alert (WEA) program, under the mandate of the Warning, Alert, and Response Act of 2006. These alerts include AMBER alerts, severe weather alerts, and (unblockable) Presidential Alerts, intended to inform the public of imminent threats. Recently, a test Presidential Alert was sent to all capable phones in the United States, prompting concerns about how the underlying WEA protocol could be misused or attacked. In this paper, we investigate the details of this system, and develop and demonstrate the first practical spoofing attack on Presidential Alerts, using both commercially available hardware as well as modified open source software. Our attack can be performed using a commercially-available software defined radio, and our modifications to the open source NextEPC and srsLTE software libraries. We find that with only four malicious portable base stations of a single Watt of transmit power each, almost all of a 50,000-seat stadium can be attacked with a 90% success rate. The true impact of such an attack would of course depend on the density of cell phones in range; fake alerts in crowded cities or stadiums could potentially result in cascades of panic. Fixing this problem will require a large collaborative effort between carriers, government stakeholders, and cell phone manufacturers. To seed this effort, we also discuss several defenses to address this threat in both the short and long term.

We present the demonstration of a fully distributed scheduling framework called CASTLE (Client-side Adaptive Scheduler That minimizes Load and Energy) that jointly optimizes the spectral efficiency of cellular networks and battery consumption of smart devices. To do so, we focus on scenarios when many smart devices compete for cellular resources in the same base station: spreading out transmissions over time so that only a few devices transmit at once and improves both spectral efficiency and battery consumption. To this end, we devise two novel features in CASTLE. First, we explicitly consider inter-cell interference for accurate cellular load estimation in our machine learning algorithm. Second, we propose a fully distributed scheduling algorithm that coordinates transmissions between clients based on the locally estimated load level at each client. Our formulation for minimizing battery consumption at each device leads to an optimized back off-based algorithm that fits practical environments. Our comprehensive experimental results show that CASTLE's load estimation is up to 91 % accurate, and that CASTLE achieves higher spectral efficiency with less battery consumption, compared to existing centralized scheduling algorithms as well as a distributed CSMA-like protocol. Furthermore,we develop a light-weight SDK that can expedite the deployment of CASTLE into smart devices and evaluate it in a commercial LTE network.

The explosive growth of multimedia data traffic in wired and wireless networks have led Internet service providers (ISPs) to use penalty mechanisms like throttling, capping, overage fees to manage network congestion; however, such measures are harmful to the Internet ecosystem. Therefore, we use ideas from economics to create incentive-based, as opposed to penalty-based, solutions for data plans. In particular, we explore time-dependent pricing (TDP) - a form of dynamic pricing that manages congestion by offering time-varying discounts to incentivize users to shift some data traffic temporally. To realize TDP data plans in practice, we provide (i) an optimization model to compute time-dependent prices, (ii) a system implementation for deployment in operational networks, and (iii) experiments with two cellular networks for demonstrating feasibility. Our results show that the users respond to such pricing plans by using higher volume of traffic in lower-priced (off-peak) periods and benefit from a lower $/GB fee, while the ISPs benefit from a higher revenue due to increase in off-peak usage and lower peak-to-average traffic ratio in their network. This suggests that such a pricing solution can incentivize users to modify their usage behavior and enable better revenue management in multimedia-rich networks.

In serverless computing, developers define a function to handle an event, and the serverless framework horizontally scales the application as needed. The downside of this function-based abstraction is it limits the type of application supported and places a bound on the function to be within the physical resource limitations of the server the function executes on. In this paper we propose a new abstraction for serverless computing: a developer supplies a process and the serverless framework seamlessly scales out the process's resource usage across the datacenter. This abstraction enables processing to not only be more general purpose, but also allows a process to break out of the limitations of a single server – making serverless computing more serverless. To realize this abstraction, we propose ServerlessOS, comprised of three key components: (i) a new disaggregation model, which leverages disaggregation for abstraction, but enables resources to move fluidly between servers for performance; (ii) a cloud orchestration layer which manages fine-grained resource allocation and placement throughout the application's lifetime via local and global decision making; and (iii) an isolation capability that enforces data and resource isolation across disaggregation, effectively extending Linux cgroup functionality to span servers.

Since the diagnosis of severe bufferbloat in mobile cellular networks, a number of low-latency congestion control algorithms have been proposed. However, due to the need for continuous bandwidth probing in dynamic cellular channels, existing mechanisms are designed to cyclically overload the network. As a result, it is inevitable that their latency deviates from the smallest possible level (i.e., minimum RTT). To tackle this problem, we propose a new low-latency congestion control, ExLL, which can adapt to dynamic cellular channels without overloading the network. To do so, we develop two novel techniques that run on the cellular receiver: 1) cellular bandwidth inference from the downlink packet reception pattern and 2) minimum RTT calibration from the inference on the uplink scheduling interval. Furthermore, we incorporate the control framework of FAST into ExLL's cellular specific inference techniques. Hence, ExLL can precisely control its congestion window to not overload the network unnecessarily. Our implementation of ExLL on Android smartphones demonstrates that ExLL reduces latency much closer to the minimum RTT compared to other low-latency congestion control algorithms in both static and dynamic channels of LTE networks.

VM redundancy is the foundation of resilient cloud applications. While active-active approaches combined with load balancing and autoscaling are usually resource efficient, the stateful nature of many cloud applications often necessitates 1+1 (or 1+n) active-standby approaches. Keeping the standbys, however, could result in inefficient utilization of cloud resources. We explore an intriguing cloud-based solution, where standby VMs from active-standby applications are selectively overbooked to reduce resources reserved for failures. The approach requires careful VM placement to avoid a situation where multiple standby VMs activate simultaneously on the same host and thus cannot get the full resource entitlement. Indeed today's clouds do not have this visibility to the applications. We rectify this situation through ShadowBox, a novel redundancy-aware VM scheduler that optimizes the placement and activation of standby VMs, while assuring applications' resource entitlements. Evaluation on a large-scale cloud shows that ShadowBox can significantly improve resource utilization (i.e., more than 2.5 times than traditional approaches) while minimizing the impact on applications' entitlements.

Fog computing is an emerging architecture that aims to run applications on multiple devices that lie on a continuum from cloud servers to personal user smartphones. These architectures allow applications to optimize over the information stored at and functionalities run on each device, based on individual device capabilities. We demonstrate the benefits of this approach for mobile video streaming. Existing HAS (HTTP adaptive streaming) techniques often suffer from problems like unstable video quality and suboptimal resource utilization. We find that a lack of coordination prevents both clientand network-side HAS techniques from solving them. However, our fog approach can exploit existing telecommunication APIs, which expose network capabilities to applications, in order to coordinate between clients and the network. Our coordinated HAS solution, FLARE, optimizes the total utility of all clients in a cell while maintaining stable video quality and supporting user- and device-specific needs. We implement FLARE on a commodity LTE femtocell and use the implementation to conduct the first comparison of HAS players on an LTE femtocell. By conducting extensive experiments using the ns-3 simulator, we also demonstrate that FLARE (i) enhances the average video bitrate, (ii) achieves stable video quality, and (iii) balances the throughput of simultaneous video and data flows, compared to other representative HAS solutions.

Video streaming applications based on Peer-to-Peer (P2P) systems are popular for their scalability, which is hard to achieve with traditional client-server approaches. In particular, layered video streaming has been much-studied due to its ability to differentiate users' streaming qualities in heterogeneous user environments. Previous work, however, has shown that user misbehavior (e.g., free-riding and protocol deviation) poses a serious threat to P2P systems that are not equipped with proper incentive mechanisms. We propose a method to disincentivize such misbehavior. Our SVC-TChain is a layered P2P video streaming method based on scalable video coding (SVC), which uses the recently proposed T-Chain incentive mechanism to discourage free-riding. After introducing T-Chain, we present the first analytical framework to study SVC piece selection with multiple video layers, using it to efficiently choose SVC-TChain's optimal piece selection parameters and thus discourage deviations from the piece selection policy. Extensive experimental results show that SVC-TChain outperforms layered extensions of BiTos and Give-to-Get, two popular P2P video streaming approaches, both in the absence of user misbehavior and when some users misbehave.

Current COTS WiFi based work on wireless motion sensing extracts human movements such as keystroking and hand motion mainly from amplitude training to classify different types of motions, as obtaining meaningful phase values is very challenging due to time-varying phase noises occurred with the movement. However, the methods based only on amplitude training are not very practical since their accuracy is not environment and location independent. This paper proposes an effective phase noise calibration technique which can be broadly applicable to COTS WiFi based motion sensing. We leverage the fact that multi-path for indoor environment contains certain static paths, such as reflections from wall or static furniture, as well as dynamic paths due to human hand and arm movements. When a hand moves, the phase value of the signal from the hand rotates as the path length changes and causes the superposition of signals over static and dynamic paths in antenna and frequency domain. To evaluate the effectiveness of the proposed technique, we experiment with a prototype system that can track hand gestures in a non-intrusive manner, i.e. users are not equipped with any device, using COTS WiFi devices. Our evaluation shows that calibrated phase values provide much rich, yet robust information on motion tracking -- 80th percentile angle estimation error up to 14 degrees, 80th percentile tracking error up to 15 cm, and its robustness to the environment and the speed of movement.

The growing volume of mobile data traffic has led many Internet service providers (ISPs) to cap the monthly data usage of their users and to charge overage fees, when the data caps are exceeded. Yet data caps imperfectly capture the reality of heterogeneous data usage over a month-even the same user may have varied requirements from month to month. In response, some ISPs are providing alternative avenues for users to customize data plans to their needs. In this paper, we examine a secondary data market, as for example created by China Mobile Hong Kong, in which users can buy and sell leftover data caps from one another. While similar to an auction in that users submit bids to buy and sell data, it differs from traditional double auctions in that the ISP serves as the middleman between buyers and sellers. Such a market faces two questions. First, can users learn each others' trading behavior well enough for the market to function, and second, do ISPs have a financial incentive to offer such a market? Different users' abilities to trade data depend on others, thus forcing users to not only optimize the amounts of data they bid, but also to learn and adjust for other users' trading behavior. We derive users' optimal behavior and propose an algorithm for ISPs to match buyers and sellers. We compare the optimal matchings for different ISP objectives and derive conditions under which the secondary market increases ISP revenue: while the ISP loses revenue from overage fees, it can assess administration fees and profit from the differences between the buyer and seller prices. Finally, we use one year of usage data from 100 U.S. mobile users to simulate the market dynamics and to illustrate that sustainable conditions for a revenue increase for the ISP can hold in practice.

In this paper, we propose a simple, distributed, but highly efficient fairness-enforcing incentive mechanism for cooperative computing. The proposed mechanism, called triangle chaining (T-Chain), enforces reciprocity to avoid the exploitable aspects of the schemes that allow free-riding. In T-Chain, symmetric key cryptography provides the basis for a lightweight, almost-fair exchange protocol, which is coupled with a pay-it-forward mechanism. This combination increases the opportunity for multi-lateral exchanges and further maximizes the resource utilization of participants, each of whom is assumed to operate solely for his or her own benefit. T-Chain also provides barrier-free entry to newcomers with flexible resource allocation, allowing them to immediately benefit, and, therefore, is suitable for dynamic environments with high churn (i.e., turnover). T-Chain is distributed and simple to implement, as no trusted third party is required to monitor or enforce the scheme, nor is there any reliance on reputation information or tokens.

As more devices gain Internet connectivity, more information needs to be exchanged between them. For instance, cloud servers might disseminate instructions to clients, or sensors in the Internet of Things might send measurements to each other. In such scenarios, information spreads faster when users have an incentive to contribute data to others. While many works have considered this problem in peer-to-peer scenarios, none have rigorously theorized the performance of different design choices for the incentive mechanisms. In particular, different designs have different ways of "bootstrapping" new users (distributing information to them) and preventing "free-riding" (receiving information without uploading any in return). We classify incentive mechanisms in terms of reciprocity-, altruism-, and reputation-based algorithms, and then analyze the performance of these three basic and three hybrid algorithms. We show that the algorithms lie along a tradeoff between fairness and efficiency, with altruism and reciprocity at the two extremes. The three hybrids all leverage their component algorithms to achieve similar efficiency. The reputation hybrids are the most fair and can nearly match altruism's bootstrapping speed, but only the reciprocity/reputation hybrid can match reciprocity's zero-tolerance for free-riding. It therefore yields better fairness and efficiency when free-riders are present. We validate these comparisons with extensive experimental results.

Cloud service providers (CSPs) often face highly dynamic user demands for their resources, which can make it difficult for them to maintain consistent quality-of-service. Some CSPs try to stabilize user demands by offering sustained-use discounts to jobs that consume more instance-hours per month. These discounts present an opportunity for users to pool their usage together into a single ``job.'' In this paper, we examine the viability of a middleman, the cloud virtual service provider (CVSP), that rents cloud resources from a CSP and then resells them to users. We show that the CVSP's business model is only viable if the average job runtimes and thresholds for sustained-use discounts are sufficiently small; otherwise, the CVSP cannot simultaneously maintain low job waiting times while qualifying for a sustained-use discount. We quantify these viability conditions by modeling the CVSP's job scheduling and then use this model to derive users' utility-maximizing demands and the CVSP's profit-maximizing price, as well as the optimal number of instances that the CVSP should rent from the CSP. We verify our results on a one-month trace from Google's production compute cluster, through which we first validate our assumptions on the job arrival and runtime distributions, and then show that the CVSP is viable under these workload traces. Indeed, the CVSP can earn a positive profit without significantly impacting the CSP's revenue, indicating that the CSP and CVSP can coexist in the cloud market.

To cope with recent exponential increases in demand for mobile data, wireless Internet service providers (ISPs) are increasingly changing their pricing plans and deploying Wi-Fi hotspots to offload their mobile traffic. However, these ISP-centric approaches for traffic management do not always match the interests of mobile users. Users face a complex, multi-dimensional tradeoff between cost, throughput, and delay in making their offloading decisions: while they may save money and receive a higher throughput by waiting for Wi-Fi access, they may not wait for Wi-Fi if they are sensitive to delay. To navigate this tradeoff, we develop Adaptive bandwidth Management through USer-Empowerment (AMUSE), a functional prototype of a practical, cost-aware Wi-Fi offloading system that takes into account a user's throughput-delay tradeoffs and cellular budget constraint. Based on predicted future usage and Wi-Fi availability, AMUSE decides which applications to offload to what times of the day. Since nearly all traffic flows from mobile devices are TCP flows, we introduce a new receiver-side bandwidth allocation mechanism to practically enforce the assigned rate of each TCP application. Thus, AMUSE users can optimize their bandwidth rates according to their own cost-throughput-delay tradeoff without relying on support from different apps’ content servers. Through a measurement study of 20 smartphone users’ traffic usage traces, we observe that though users already offload a large amount of some application types, our framework can offload a significant additional portion of users’ cellular traffic. We implement AMUSE on Windows 7 tablets and evaluate its effectiveness with 3G and Wi-Fi usage data obtained from a trial with 37 mobile users. Our results show that AMUSE improves user utility; when compared with AMUSE, other offloading algorithms yield 14 and 27 percent lower user utilities for light and heavy users, respectively. Intelligently managing users’ competing interests for cost, throughput, and delay can therefore improve their offloading decisions.

Public cloud storage has recently surged in popularity. However, cloud storage providers (CSPs) today offer fairly rigid services, which cannot be customized to meet individual users' needs. We propose a distributed, client-defined architecture that integrates multiple autonomous CSPs into one unified cloud and allows individual clients to specify their desired performance levels and share files. We design, implement, and deploy CYRUS (Client-defined privacY-protected Reliable cloUd Service), a practical system that realizes this architecture. CYRUS ensures user privacy and reliability by scattering files into smaller pieces across multiple CSPs, so that no one CSP can read users' data. We develop an algorithm that sets reliability and privacy parameters according to user needs and selects CSPs from which to download user data so as to minimize latency. To accommodate multiple autonomous clients, we allow clients to upload simultaneous file updates and detect conflicts after the fact from the client. We finally evaluate the performance of a CYRUS prototype that connects to four popular commercial CSPs in both lab testbeds and user trials, and discuss CYRUS's implications for the cloud storage market.

In this paper, we propose a simple, distributed, but highly efficient fairness-enforcing incentive mechanism for cooperative computing. The proposed incentive scheme, called Triangle Chaining (T-Chain), enforces reciprocity to minimize the exploitable aspects of other schemes that allow free-riding. In T-Chain, symmetric key cryptography provides the basis for a lightweight, almost-fair exchange protocol, which is coupled with a pay-it-forward mechanism. This combination increases the opportunity for multi-lateral exchanges and further maximizes the resource utilization of participants, each of whom is assumed to operate solely for his or her own benefit. T-Chain also provides barrier-free entry to newcomers with flexible resource allocation, providing them with immediate benefits, and therefore is suitable for dynamic environments with high churn (i.e., Turnover). TChain is distributed and simple to implement, as no trusted third party is required to monitor or enforce the scheme, nor is there any reliance on reputation information or tokens.

In January 2014, AT&T introduced sponsored data to the U.S. mobile data market, allowing content providers (CPs) to subsidize users' cost of mobile data. As sponsored data gains traction in industry, it is important to understand its implications. This work considers CPs' choice of how much content to sponsor and the implications for users, CPs, and ISPs (Internet service providers). We first formulate a model of user, CP, and ISP interaction for heterogeneous users and CPs and derive their optimal behaviors. We then show that these behaviors can reverse our intuition as to how user demand and utility change with different user and CP characteristics. While all three parties can benefit from sponsored data, we find that sponsorship disproportionately favors less cost-constrained CPs and more cost-constrained users, exacerbating CP inequalities but making user demand more even. We also show that users' utilities increase more than CPs' with sponsored data. We finally illustrate these results in practice through numerical simulations with data from a commercial pricing trial and introduce a framework for CPs to decide which, in addition to how much, content to sponsor.

The growing volume of mobile data traffic has led many Internet service providers (ISPs) to cap their users' monthly data usage, with overage fees for exceeding their caps. In this work, we examine a secondary data market in which users can buy and sell leftover data caps from each other. China Mobile Hong Kong recently introduced such a market. While similar to an auction in that users submit bids to buy and sell data, it differs from traditional double auctions in that the ISP serves as the middleman between buyers and sellers. We derive the optimal prices and amount of data that different buyers and sellers are willing to bid in this market and then propose an algorithm for ISPs to match buyers and sellers. We compare the optimal matching for different ISP objectives and derive conditions under which an ISP can obtain higher revenue with the secondary market: while the ISP loses revenue from overage fees, it can assess administration fees and take the differences between the buyer and seller prices. Finally, we use one year of usage data from 100 U.S. mobile users to illustrate that the conditions for a revenue increase can hold in practice.

Recent increases in network traffic have led to severe congestion in broadband networks. We propose to mitigate this problem with a two-level edge-based solution that incentivizes users to moderate their bandwidth usage based on their actual needs. In the first level, home gateways are given QoE (quality of experience) credits that they can spend to receive more bandwidth at congested times; to ensure fairness, the credits are redistributed to other gateways after they are spent. We show that this scheme guarantees long-term fairness and maximizes users' total satisfaction at the equilibrium. In the second level, each gateway allocates bandwidth among its users and apps according to its own priorities. Gateways can thus customize their bandwidth allocation depending on individual preferences. We develop a prototype of this second-level allocation on commodity wireless routers. We then consider an example scenario and show by simulation and implementation results that our solution outperforms an equal bandwidth allocation, increasing users' overall utility and fairly allocating bandwidth across users.

In this paper, we optimize the use of existing DSL low-power sleeping modes (L2 and L3) in order to improve the energy efficiency of DSL access networks. Given that switching on a DSL line can cause instability by increasing the amount of time-varying crosstalk in the cable bundle, and that it takes energy and time to switch a DSL line on and off, we develop a method to optimally choose the appropriate sleeping state based on the line and traffic characteristics. We further develop and prove the structural properties of the optimal policy for switching to the appropriate sleeping state, allowing transitions between submodes with different power and transmit rate characteristics. We also present techniques that guarantee stable sleep mode operation. Using a realistic DSL simulator, we demonstrate the three-way tradeoff among energy consumption, delay performance, and stability. The increased flexibility of control introduced by our approach improves the energy-delay Pareto optimal tradeoff, and results in a more energy efficient and stable DSL operation compared to existing power saving policies.

To alleviate the congestion caused by rapid growth in demand for mobile data, wireless service providers (WSPs) have begun encouraging users to offload some of their traffic onto supplementary network technologies, e.g., offloading from 3G or 4G to WiFi or femtocells. With the growing popularity of such offerings, a deeper understanding of the underlying economic principles and their impact on technology adoption is necessary. To this end, we develop a model for user adoption of a base technology (e.g., 3G) and a bundle of the base plus a supplementary technology (e.g., 3G + WiFi). Users individually make their adoption decisions based on several factors, including the technologies' intrinsic qualities, negative congestion externalities from other subscribers, and the flat access rates that a WSP charges. We then show how these user-level decisions translate into aggregate adoption dynamics and prove that these converge to a unique equilibrium for a given set of exogenously determined system parameters. We fully characterize these equilibria and study adoption behaviors of interest to a WSP. We then derive analytical expressions for the revenue-maximizing prices and optimal coverage factor for the supplementary technology and examine some resulting nonintuitive user adoption behaviors. Finally, we develop a mobile app to collect empirical 3G/WiFi usage data and numerically investigate the profit-maximizing adoption levels when a WSP accounts for its cost of deploying the supplemental technology and savings from offloading traffic onto this technology.

In an era of 108% annual growth in demand for mobile data and $10/GB overage fees, Internet Service Providers (ISPs) are experiencing severe congestion and in turn are hurting consumers with aggressive pricing measures. But smarter practices, such as time-dependent pricing (TDP), reward users for shifting their non-critical demand to off-peak hours and can potentially benefit both users and ISPs. Although dynamic TDP ideas have existed for many years, dynamic pricing for mobile data is only now gaining interest among ISPs. Yet TDP plans require not only systems engineering but also an understanding of economic incentives, user behavior and interface design. In particular, the HCI aspects of communicating price feedback signals from the network and the response of mobile data users need to be studied in the real world. But investigating these issues by deploying a virtual TDP data plan for real ISP customers is challenging and rarely explored. To this end, we carried out the first TDP trial for mobile data in the US with 10 families. We describe the insights gained from the trial, which can help the HCI community as well as ISPs, app developers and designers create tools that empower users to better control their usage and save on their monthly bills, while also alleviating network congestion.

Mobile users face a tradeoff between cost, throughput, and delay in making their offloading decisions. To navigate this tradeoff, we propose AMUSE (Adaptive bandwidth Management through USer-Empowerment), a practical, costaware WiFi offloading system that takes into account a user's throughput-delay tradeoffs and cellular budget constraint. Based on predicted future usage and WiFi availability, AMUSE decides which applications to offload to what times of the day. To practically enforce the assigned rate of each TCP application, we introduce a receiver-side TCP bandwidth control algorithm that adjusts the rate by controlling the TCP advertisement window from the user side. We implement AMUSE on Windows 7 tablets and evaluate its effectiveness with 3G and WiFi usage data obtained from a trial with 25 mobile users. Our results show that AMUSE improves user utility.

To alleviate the congestion caused by rapid growth in demand for mobile data, ISPs have begun encouraging users to offload some of their traffic onto a supplementary, better quality network technology, e.g., offloading from 3G or 4G to WiFi and femtocells. With the growing popularity of such offerings, a deeper understanding of the underlying economic principles and their impact on technology adoption is necessary. To this end, we develop a model for user adoption of a base wireless technology and a bundle of the base plus a supplementary technology. In our model, individual users make their adoption decisions based on several factors, including the technologies' intrinsic qualities, throughput degradation due to congestion externalities from other subscribers, and the flat access rates that an ISP charges. We study the adoption dynamics and show that they converge to a unique equilibrium for a given set of exogenously determined system parameters. In particular, we characterize the occurrence of interesting adoption behaviors, including a possible decrease in the adoption of the supplementary technology as its coverage increases. Similar behaviors occur at an ISP's profit-maximizing prices and the optimal coverage area for the supplementary technology. To account for the potential benefits from offloading in practice, we collect 3G and WiFi usage and location data from twenty mobile users. We then use this data to numerically investigate the profit-maximizing adoption levels when an ISP accounts for its cost of deploying the supplemental technology and savings from offloading traffic onto this technology.

Rapid increases in the demand for broadband data are increasingly causing a growth in costs for communication service providers (CSPs). Yet under the current pricing plans, CSPs' revenue has not kept pace with these costs. Thus, many CSPs are considering Smart Data Pricing (SDP) as a way to reduce cost or increase revenue. Before offering such novel data plans, however, CSPs must conduct trials of the specific data plans proposed. Due to the complexity of necessary changes in network equipment and a need to carefully design the trial in order to understand customer behavior, planning such trials is not only a critical precursor to SDP deployment, but also a nontrivial undertaking in itself. This paper discusses general principles of trial design and proposes two methods for estimating their effectiveness. We first give an introduction to the goals of SDP research and review three possible SDP approaches. We then discuss the importance of pre-trial participant surveys and some technical considerations of implementing the trial infrastructure for a particular SDP algorithm. Finally, we show how the CSP may extrapolate from the trial results to estimate the SDP trial's benefits, in terms of changes in traffic patterns and a reduction in spectrum requirements. We conclude with some remarks about future work.

Large online service providers (OSPs) often build private backbone networks to interconnect data centers in multiple locations. These data centers house numerous applications that produce multiple classes of traffic with diverse performance objectives. Applications in the same class may also have differences in relative importance to the OSP's core business. By controlling both the hosts and the routers, an OSP can perform both application rate-control and network routing. However, centralized management of both rates and routes does not scale due to excessive message-passing between the hosts, routers, and management systems. Similarly, fully-distributed approaches do not scale and converge slowly. To overcome these issues, we investigate two semi-centralized designs that lie at practical points along the spectrum between fully-distributed and fully-centralized solutions. We achieve scalability by distributing computation across multiple tiers of an optimization machinery. Our first design uses two tiers, representing the backbone and classes, to compute class-level link bandwidths and application sending rates. Our second design has an additional tier representing individual data centers. Using optimization, we show that both designs provably maximize the aggregate utility over all traffic classes. Simulations on realistic backbones show that the 3-tier design is more scalable, but converges slower than the 2-tier design.

Switching off a DSL line to a low-power sleeping state is becoming an important method to enhance energy efficiency of DSL broadband access networks. Although a low-power (L2) state and an off (L3) state are already defined in DSL standards, they have not been fully exploited due to concern about the resulting time-varying crosstalk. In this paper, we develop a method to optimally choose the appropriate sleeping state based on the modem's switching cost characteristics and the power consumption incurred during each sleeping state. We further develop the optimal policy for switching to the appropriate sleeping state, allowing transitions between operating modes with different power and transmit rate. Numerical results on our realistic DSL simulator show that more flexibility in control, introduced by our approach, improves the energy-delay Pareto optimal tradeoff, and results in a more energy efficient and stable DSL system compared to the existing power saving policies.

Today's data centers need efficient traffic management to improve resource utilization in their networks. In this work, we study a joint tenant (e.g., server or virtual machine) placement and routing problem to minimize traffic costs. These two complementary degrees of freedom—placement and routing—are mutually-dependent, however, are often optimized separately in today's data centers. Leveraging and expanding the technique of Markov approximation, we propose an efficient online algorithm in a dynamic environment under changing traffic loads. The algorithm requires a very small number of virtual machine migrations and is easy to implement in practice. Performance evaluation that employs the real data center traffic traces under a spectrum of elephant and mice flows, demonstrates a consistent and significant improvement over the benchmark achieved by common heuristics.

ISPs around the world have begun to offer new pricing plans for wireless data, such as usage-based pricing in the U.S., to mitigate recent growth in bandwidth demand. Time Dependent Pricing (TDP) represents a next step in this direction [1, 2]. With TDP, ISPs can shift traffic to off-peak periods, thus reducing their cost, while consumers save money by choosing the time of usage. TDP uses a feedback control loop between an ISP and its users to account for users' responses to offered prices in optimizing the future prices. We have implemented such a TDP system and are presently conducting a user trial at Princeton while planning larger trials with commercial ISPs. This demo will introduce the audience to our system's ISP- and user-side features. On the ISP side, we show the current network congestion, while on the user side, we show device UIs displaying the offered prices, the device usage history, and automated scheduling of applications to keep users within a specified budget.

The two largest U.S. wireless ISPs have recently moved towards usage-based pricing to better manage the growing demand on their networks. Yet usage-based pricing still requires ISPs to over-provision capacity for demand at peak times of the day. Time-dependent pricing (TDP) addresses this problem by considering when a user consumes data, in addition to how much is used. We present the architecture, implementation, and a user trial of an end-to-end TDP system called TUBE. TUBE creates a price-based feedback control loop between an ISP and its end users. On the ISP side, it computes TDP prices so as to balance the cost of congestion during peak periods with that of offering lower prices in less congested periods. On mobile devices, it provides a graphical user interface that allows users to respond to the offered prices either by themselves or using an "autopilot" mode. We conducted a pilot TUBE trial with 50 iPhone or iPad 3G data users, who were charged according to our TDP algorithms. Our results show that TDP benefits both operators and customers, flattening the temporal fluctuation of demand while allowing users to save money by choosing the time and volume of their usage.

Smart grids are capable of two-way communication between individual user devices and the electricity provider, enabling providers to create a control-feedback loop using time-dependent pricing. By charging users more in peak and less in off-peak hours, the provider can induce users to shift their consumption to off-peak periods, thus relieving stress on the power grid and the cost incurred from large peak loads. We formulate the electricity provider's cost minimization problem in setting these prices by considering consumers' device-specific scheduling flexibility and the provider's cost structure of purchasing electricity from an electricity generator. Consumers' willingness to shift their device usage is modeled probabilistically, with parameters that can be estimated from real data. We develop an algorithm for computing day-ahead prices, and another algorithm for estimating and refining user reaction to the prices. Together, these two algorithms allow the provider to dynamically adjust the offered prices based on user behavior. Numerical simulations with data from an Ontario electricity provider show that our pricing algorithm can significantly reduce the cost incurred by the provider.

Energy efficient and stable operation of the DSL broadband access infrastructure has become an essential part of the emerging trend towards green communications. One promising means to obtain energy savings is the use of low power "sleep modes", putting DSL modems to sleep when they are not used. Executing the optimal sleeping policies is, however, not straightforward since turning the modem ON and OFF consumes both energy and time, and it also impacts the stability of the DSL network. We present an analytic framework providing optimal sleeping policies that achieve a Pareto-optimal tradeoff between energy consumption and delay performance. Furthermore, we present mechanisms achieving stable sleep mode operation that improve overall stability of DSL systems. Using a realistic DSL simulator, we demonstrate the three-way tradeoff between energy consumption, delay performance and stability.

Charging different prices for Internet access at different times induces users to spread out their bandwidth consumption across times of the day. Potential impact on ISP revenue, congestion management, and consumer behavior can be significant, yet some fundamental questions remain: is it feasible to operate time dependent pricing and how much benefit can it bring? We develop an efficient way to compute the cost-minimizing time-dependent prices for an Internet service provider (ISP), using both a static session-level model and a dynamic session model with stochastic arrivals. A key step is choosing the representation of the optimization problem so that the resulting formulations remain computationally tractable for large-scale problems. We next show simulations illustrating the use and limitation of time-dependent pricing. These results demonstrate that optimal prices, which "reward'' users for deferring their sessions, roughly correlate with demand in each period, and that changing prices based on real-time traffic estimates may significantly reduce ISP cost. The degree to which traffic is evened out over times of the day depends on the time-sensitivity of sessions, cost structure of the ISP, and amount of traffic not subject to time-dependent prices. Finally, we present our system integration and implementation, called TUBE, and proof-of-concept experimentation.

Congestion control in wireless multi-hop networks is challenging and complicated because of two reasons. First, interference is ubiquitous and causes loss in the shared medium. Second, wireless multihop networks are characterized by the use of diverse and dynamically changing routing paths. Traditional end point based congestion control protocols are ineffective in such a setting resulting in unfairness and starvation. This paper adapts the optimal theoretical work of Tassiulas and Ephremedes on cross-layer optimization of wireless networks involving congestion control, routing and scheduling, for practical solutions to congestion control in multi-hop wireless networks. This work is the first that implements in real off-shelf radios, a differential backlog based MAC scheduling and router-assisted backpressure congestion control for multi-hop wireless networks. Our adaptation, called DiffQ, is implemented between transport and IP and supports legacy TCP and UDP applications. In a network of 46 IEEE 802.11 wireless nodes, we demonstrate that DiffQ far outperforms many previously proposed "practical" solutions for congestion control.

In this demo, we showcase DiffQ - a congestion control protocol inspired by theoretical cross-layer optimization approaches. DiffQ can support congestion control for network flows that use either single-path or opportunistic multi-path routing. Our demo will focus on the performance in single-path routing environments, where contemporary end-point congestion control algorithms like TCP face severe unfairness or even starvation. This is primarily due to the interaction of such protocols with MAC layer unfairness. We demonstrate micro (5 flows) as well as macro-evaluations (60 flows) of such cases. Our demo is conducted on WiSeNet - a 70-node wireless mesh test-bed hosted in the computer science building at NCSU. Distributed over a 100,000 sq ft building, this is one of the largest test-bed installations both in terms of number of nodes and coverage area, hence an ideal testing ground for such scenarios. Experimental results like throughput, MAC-layer statistics, delay, routing path flaps and network buffer overflows are recorded and displayed in real-time and enable a bird's eye-view of the entire network status and allow us to point out various phenomenon as they happen.

Window growth function for congestion control is a strong determinant of protocol behaviors, especially its second and higher-order behaviors associated with the distribution of transmission rates, its variances, and protocol stability. This paper presents a new stochastic tool, called convex ordering, that provides an ordering of any convex function of transmission rates of two protocols and valuable insights into high order behaviors of protocols. As the ordering determined by this tool is consistent with any convex function of rates, it can be applied to any unknown metric for protocol performance that consists of some high-order moments of transmission rates, as well as those already known such as rate variance. Using the tool, it is analyzed that a protocol with a growth function that starts off with a concave function and then switches to a convex function (e.g., an odd order function such as x3and x5) around the maximum window size in the previous loss epoch, gives the smallest rate variation under a variety of network conditions. Among existing protocols, BIC and CUBIC have this window growth function. Experimental and simulation results confirm the analytical findings.