Automatic data decomposition for message-passing machines

This paper considers the problem of multi-agent distributed optimization. In this problem, there are multiple agents in the system, and each agent only knows its local cost function. The objective for the agents is to collectively compute a common minimum of the aggregate of all their local cost functions. In principle, this problem is solvable using a distributed variant of the traditional gradient-descent method, which is an iterative method. However, the speed of convergence of the traditional gradient-descent method is highly influenced by the conditioning of the optimization problem being solved. Specifically, the method requires a large number of iterations to converge to a solution if the optimization problem is ill-conditioned. In this paper, we propose an iterative pre-conditioning approach that can significantly attenuate the influence of the problem's conditioning on the convergence-speed of the gradient-descent method. The proposed pre-conditioning approach can be easily implemented in distributed systems and has minimal computation and communication overhead. For now, we only consider a specific distributed optimization problem wherein the individual local cost functions of the agents are quadratic. Besides the theoretical guarantees, the improved convergence speed of our approach is demonstrated through experiments on a real data-set.

This work considers the problem of decentralized coordination between multiple non-holonomic vehicles, each navigating to a specified goal. By augmenting the Generalized Roundabout Policy (GRP), which guarantees collision avoidance, this paper improves the performance and liveness characteristics for such problems. These gains are achieved by integrating a second hybrid policy with GRP that updates the desired direction for each vehicle based on a dynamic priority scheme. In this scheme, minimalistic communication between vehicles is employed, such that information is periodically exchanged when changes in the high-level operating mode or prioritization occur. This information exchange is taking place only locally and data are exchanged only between neighboring vehicles. Additionally, each agent selects a control using only this local information and rules established by the two underlying hybrid automata. The proposed technique scales well due to its decentralized nature and as the computational complexity depends on the maximum number of vehicles in communication range for a vehicle. This paper presents simulations which show that the proposed approach can solve problems faster than using GRP alone, as well as solve instances in which GRP fails to find a solution, with minimal communication and computational overhead.

Efficient access to spectrum is critical to maintain reliable and robust communication in military networks, especially for ground vehicles on the move. High mobility and frequent topology variations make spectrum management in battlefields challenging. In this paper, we present a decentralized, hierarchical approach to assign spectrum channels among vehicles in a fair and efficient manner. Making use of group mobility, we propose a low-complexity allocation scheme with minimum communication and computation overhead. Experimental results demonstrate that the proposed algorithm can reduce the overhead by 10+ fold with maintaining reasonable spectrum utilization.

Tripartite (Diffie-Hellman) Key Exchange (3KE), introduced by Joux (ANTS-IV 2000), represents today the only known class of group key exchange protocols, in which computation of unauthenticated session keys requires one round and proceeds with minimal computation and communication overhead. The first one-round authenticated 3KE version that preserved the unique efficiency properties of the original protocol and strengthened its security towards resilience against leakage of ephemeral (session-dependent) secrets was proposed recently by Manulis, Suzuki, and Ustaoglu (ICISC 2009).In this work we explore sufficient conditions for building such protocols. We define a set of admissible polynomials and show how their construction generically implies 3KE protocols with the desired security and efficiency properties. Our result generalizes the previous 3KE protocol and gives rise to many new authenticated constructions, all of which enjoy forward secrecy and resilience to ephemeral key-leakage under the gap Bilinear Diffie-Hellman assumption in the random oracle model.

Rasterization hardware provides interactive frame rates for rendering dynamic scenes, but lacks the ability of ray tracing required for efficient global illumination simulation. Existing ray tracing based methods yield high quality renderings but are far too slow for interactive use. We present a new parallel global illumination algorithm that perfectly scales, has minimal preprocessing and communication overhead, applies highly efficient sampling techniques based on randomized quasi-Monte Carlo integration, and benefits from a fast parallel ray tracing implementation by shooting coherent groups of rays. Thus a performance is achieved that allows for applying arbitrary changes to the scene, while simulating global illumination including shadows from area light sources, indirect illumination, specular effects, and caustics at interactive frame rates. Ceasing interaction rapidly provides high quality renderings.

The smart water monitoring system (SWMS) integrated with IoT provides improved services to the society with minimal effort. At the same time, IoT connectivity brings various active, passive, and physical security threats to the system. The conventional cryptosystems are infeasible for resource-constrained IoT devices and susceptible to physical attacks. Physically unclonable function (PUF) emerged as a low-cost, convenient hardware security primitive. In this work, we have proposed a secure (against active, passive, and physical attacks) authentication and key sharing protocol for the SWMS with low overhead. The extensive security analysis illustrates the durability of the protocol. The performance analysis shows the proposed protocol's minimal computation, communication overhead, and resource utilization. Using a mathematical PUF model (i.e., soft model) reduces the server-side burden of storing PUF challenge-response pairs. Moreover, the protocol deals with the noisy PUF responses to maintain the correctness and accuracy of the authentication process. Finally, a comparison with the existing works establishes the proposed protocol's superiority in the SWMS.

In this paper, we investigate dense short packet forwarding for clustering-based massive Internet-of-Things (mIoT). The objective is to support the data forwarding with minimal communication overhead while satisfying the differentiated latency constraints from the transport layer perspective. To this end, we propose a dense packet assembling scheme, named ShuttleBus, for forwarding devices in mIoT to achieve effective data merging. The assembling scheme is designed based on the stream multiplexing mechanism of the Quick UDP Internet Connection (QUIC) protocol. With ShuttleBus, the payload data sent from IoT devices are extracted as independent frames belonging to different data streams. The ShuttleBus can bundle data frames from multiple streams into a single packet while ensuring data integrity of these streams. Furthermore, we develop a resilient packing mechanism in packet assembling to merge data received from IoT devices within a cluster. In addition, a latency-oriented scheduling mechanism for backlogged QUIC data is established to guarantee satisfactory delivery of diverse transmission tasks. To accommodate the dynamic network environment, we tailor a learning-based algorithm to determine the optimal packet assembling time adaptively. We evaluate the performance of ShuttleBus under various network load conditions. Both analytical and experimental results demonstrate that the proposed scheme significantly reduces communication overhead and enhances data delivery performance under stringent latency constraints.

Tree-structured computations are relatively easy to process in parallel. As leaf processes are recursively spawned they can be assigned to independent processors in a multicomputer network. However, to achieve good performance the on-line mapping algorithm must maintain load balance, i.e., distribute processes equitably among processors. Additionally, the algorithm itself must be distributed in nature, and process allocation must be completed via message-passing with minimal communication overhead. This paper investigates bounds on the performance of deterministic and randomized algorithms for on-line tree embeddings. In particular, we study trade-offs between computation overhead (load imbalance) and communication overhead (message congestion). We give a simple technique to derive lower bounds on the congestion that any on-line allocation algorithm must incur in order to guarantee load balance. This technique works for both randomized and deterministic algorithms. We prove that the advantage of randomization is limited. Optimal bounds are achieved for several networks, including multidimensional grids and butterflies.

The key concerns to enhance the lifetime of IoT-enabled Underwater Wireless Sensor Networks (IoT-UWSNs) are energy-efficiency and reliable data delivery under constrained resource. Traditional transmission approaches increase the communication overhead, which results in congestion and affect the reliable data delivery. Currently, many routing protocols have been proposed for UWSNs to ensure reliable data delivery and to conserve the node’s battery with minimum communication overhead (by avoiding void holes in the network). In this paper, adaptive energy-efficient routing protocols are proposed to tackle the aforementioned problems using the Shortest Path First (SPF) with least number of active nodes strategy. These novel protocols have been developed by integrating the prominent features of Forward Layered Multi-path Power Control One (FLMPC-One) routing protocol, which uses 2-hop neighbor information, Forward Layered Multi-path Power Control Two (FLMPC-Two) routing protocol, which uses 3-hop neighbor information and ’Dijkstra’ algorithm (for shortest path selection). Different Packet Sizes (PSs) with different Data Rates (DRs) are also taken into consideration to check the dynamicity of the proposed protocols. The achieved outcomes clearly validate the proposed protocols, namely: Shortest Path First using 3-hop neighbors information (SPF-Three) and Breadth First Search with Shortest Path First using 3-hop neighbors information (BFS-SPF-Three). Simulation results show the effectiveness of the proposed protocols in terms of minimum Energy Consumption (EC) and Required Packet Error Rate (RPER) with a minimum number of active nodes at the cost of affordable delay.

Monitoring is an issue of primary concern in current and next generation networked systems. For ex, the objective of sensor networks is to monitor their surroundings for a variety of different applications like atmospheric conditions, wildlife behavior, and troop movements among others. Similarly, monitoring in data networks is critical not only for accounting and management, but also for detecting anomalies and attacks. Such monitoring applications are inherently continuous and distributed, and must be designed to minimize the communication overhead that they introduce. In this context we introduce and study a fundamental class of problems called where we must return the aggregate frequency count of an event that is continuously monitored by distributed nodes with a user-specified accuracy whenever the actual count exceeds a given threshold value.In this paper we propose to address the problem of thresholded counts by setting local thresholds at each monitoring node and initiating communication only when the locally observed data exceeds these local thresholds. We explore algorithms in two categories: static and adaptive thresholds. In the static case, we consider thresholds based on a linear combination of two alternate strategies, and show that there exists an optimal blend of the two strategies that results in minimum communication overhead. We further show that this optimal blend can be found using a steepest descent search. In the adaptive case, we propose algorithms that adjust the local thresholds based on the observed distributions of updated information. We use extensive simulations not only to verify the accuracy of our algorithms and validate our theoretical results, but also to evaluate the performance of our algorithms. We find that both approaches yield significant savings over the naive approach of centralized processing.

This paper describes an automatic data distribution method which deal with both the alignment and the distribution problems in a single optimization phase, as opposed to sequentially solving these two inter-dependent approaches as done by previous work. The core of this work is called the Communication-Parallelism Graph, which describes the relationships among array dimensions of the same and different array references regarding communication and parallelism. The overall data distribution problem is then formulated as a linear 0–1 integer programming problem, where the objective function to be minimized is the total execution time. The solution is static in the sense that the layout of the arrays does not change during the execution of the program. We also show the feasibility of using this approach to solve the problem in terms of compilation time and quality of the solutions generated.

In this paper, we propose an efficient technique for run-time application mapping onto Network-on-Chip (NoC) platforms with multiple voltage levels. Our technique consists of a region selection algorithm and a heuristic for run-time application mapping which minimizes the communication energy consumption, while still providing the required performance guarantees. The proposed technique allows for new applications to be easily added to the system platform with minimal inter-processor communication overhead. Moreover, our approach scales very well for large designs. Finally, the experimental results show as much as 50% communication energy savings compared to arbitrary mapping solutions.

Personalized Federated Learning (pFL) can improve the accuracy of local models and provide enhanced edge intelligence without exposing the raw data in Mobile Edge Computing (MEC). However, in the MEC environment with constrained communication resources, transmitting the entire model between the server and the clients in traditional pFL methods imposes substantial communication overhead, which can lead to inaccurate personalization and degraded performance of mobile clients. In response, we propose a Communication-Efficient pFL architecture to enhance the performance of personalized models while minimizing communication overhead in MEC. First, a Knowledge-Aware Parameter Coaching method (KAPC) is presented to produce a more accurate personalized model by utilizing the layer-wise parameters of other clients with adaptive aggregation weights. Then, convergence analysis of the proposed KAPC is developed in both the convex and non-convex settings. Second, a Bidirectional Layer Selection algorithm (BLS) based on self-relationship and generalization error is proposed to select the most informative layers for transmission, which reduces communication costs. Extensive experiments are conducted, and the results demonstrate that the proposed KAPC achieves superior accuracy compared to the state-of-the-art baselines, while the proposed BLS substantially improves resource utilization without sacrificing performance.

Spurred by advances in communication technologies, mobile computing and databases that are distributed have become widespread. Such a computing environment involves data that is stored at geographically dispersed locations, and the so-called “slim” computing devices such as palmtops and wearable computers. The decentralized nature of data storage and this new paradigm in computing give rise to several issues, such as security, communication overhead, computational load demands and scalability, that are not adequately addressed by traditional centralized data mining techniques. It is essential that algorithms designed for distributed data mining scenarios mitigate some of these issues. This paper attempts to adapt one centralized data mining technique, decision tree learning, to such an environment. It presents a scalable algorithm that can be used to build decision trees from a distributed, heterogeneous database while minimizing communication overheads. This paper also shows how a decision tree may be represented in terms of its Fourier spectrum. It uses this Fourier spectrum based technique to aggregate decision trees built at the various distributed sites, simplifying the model built during the data mining stage, and notes some additional advantages of the Fourier spectrum approach.

This chapter reports on the merging of geospatial data transformation, high-performance computing (HPC), and cyberinfrastructure (CI) domains for map projection transformation through performance profiling and tuning of pRasterBlaster, a parallel map projection transformation program. pRasterBlaster is built on the desktop version of mapIMG. Profiling was employed in an effort to identify and resolve computational bottlenecks that could prevent the program from scaling to thousands of processors for map projection on large raster datasets. Performance evaluation of a parallel program is critical to achieving projection transformation as factors such as the number of processors, overhead of communications, and input/output (I/O) all contribute to efficiency in an HPC environment. Flaws in the workload distribution algorithm, in this reported work, could hardly be observed when the number of processors was small. Without being exposed to large-scale supercomputers through software integration efforts, such flaws might remain unidentified. Overall, the two computational bottlenecks highlighted in this chapter, workload distribution and data-dependent load balancing, showed that in order to produce scalable code, profiling is an important process and scaling tests are necessary to identify bottlenecks that are otherwise difficult to discover.