CUDA-for-clusters: a system for efficient execution of CUDA kernels on multi-core clusters

Rapid advancements in multi-core processor architectures coupled with low-cost, low-latency, high-bandwidth interconnects have made clusters of multi-core machines a common computing resource. Unfortunately, writing good parallel programs that efficiently utilize all the resources in such a cluster is still a major challenge. Various programming languages have been proposed as a solution to this problem, but are yet to be adopted widely to run performance-critical code mainly due to the relatively immature software framework and the effort involved in re-writing existing code in the new language. In this paper, we motivate and describe our initial study in exploring CUDA as a programming language for a cluster of multi-cores. We develop CUDA-For-Clusters (CFC), a framework that transparently orchestrates execution of CUDA kernels on a cluster of multi-core machines. The well-structured nature of a CUDA kernel, the growing popularity, support and stability of the CUDA software stack collectively make CUDA a good candidate to be considered as a programming language for a cluster. CFC uses a mixture of source-to-source compiler transformations, a work distribution runtime and a light-weight software distributed shared memory to manage parallel executions. Initial results on running several standard CUDA benchmark programs achieve impressive speedups of up to 7.5X on a cluster with 8 nodes, thereby opening up an interesting direction of research for further investigation.

Measurement Based Impromptu Deployment of a Multi-Hop Wireless Relay Network

We study the problem of optimal sequential (''as-you-go'') deployment of wireless relay nodes, as a person walks along a line of random length (with a known distribution). The objective is to create an impromptu multihop wireless network for connecting a packet source to be placed at the end of the line with a sink node located at the starting point, to operate in the light traffic regime. In walking from the sink towards the source, at every step, measurements yield the transmit powers required to establish links to one or more previously placed nodes. Based on these measurements, at every step, a decision is made to place a relay node, the overall system objective being to minimize a linear combination of the expected sum power (or the expected maximum power) required to deliver a packet from the source to the sink node and the expected number of relay nodes deployed. For each of these two objectives, two different relay selection strategies are considered: (i) each relay communicates with the sink via its immediate previous relay, (ii) the communication path can skip some of the deployed relays. With appropriate modeling assumptions, we formulate each of these problems as a Markov decision process (MDP). We provide the optimal policy structures for all these cases, and provide illustrations of the policies and their performance, via numerical results, for some typical parameters.

Exploring Transfer Learning Approaches for Head Pose Classification from Multi-view Surveillance Images

Head pose classification from surveillance images acquired with distant, large field-of-view cameras is difficult as faces are captured at low-resolution and have a blurred appearance. Domain adaptation approaches are useful for transferring knowledge from the training (source) to the test (target) data when they have different attributes, minimizing target data labeling efforts in the process. This paper examines the use of transfer learning for efficient multi-view head pose classification with minimal target training data under three challenging situations: (i) where the range of head poses in the source and target images is different, (ii) where source images capture a stationary person while target images capture a moving person whose facial appearance varies under motion due to changing perspective, scale and (iii) a combination of (i) and (ii). On the whole, the presented methods represent novel transfer learning solutions employed in the context of multi-view head pose classification. We demonstrate that the proposed solutions considerably outperform the state-of-the-art through extensive experimental validation. Finally, the DPOSE dataset compiled for benchmarking head pose classification performance with moving persons, and to aid behavioral understanding applications is presented in this work.

Generating Efficient Data Movement Code for Heterogeneous Architectures with Distributed-Memory

Programming for parallel architectures that do not have a shared address space is extremely difficult due to the need for explicit communication between memories of different compute devices. A heterogeneous system with CPUs and multiple GPUs, or a distributed-memory cluster are examples of such systems. Past works that try to automate data movement for distributed-memory architectures can lead to excessive redundant communication. In this paper, we propose an automatic data movement scheme that minimizes the volume of communication between compute devices in heterogeneous and distributed-memory systems. We show that by partitioning data dependences in a particular non-trivial way, one can generate data movement code that results in the minimum volume for a vast majority of cases. The techniques are applicable to any sequence of affine loop nests and works on top of any choice of loop transformations, parallelization, and computation placement. The data movement code generated minimizes the volume of communication for a particular configuration of these. We use a combination of powerful static analyses relying on the polyhedral compiler framework and lightweight runtime routines they generate, to build a source-to-source transformation tool that automatically generates communication code. We demonstrate that the tool is scalable and leads to substantial gains in efficiency. On a heterogeneous system, the communication volume is reduced by a factor of 11X to 83X over state-of-the-art, translating into a mean execution time speedup of 1.53X. On a distributed-memory cluster, our scheme reduces the communication volume by a factor of 1.4X to 63.5X over state-of-the-art, resulting in a mean speedup of 1.55X. In addition, our scheme yields a mean speedup of 2.19X over hand-optimized UPC codes.

Wireless Aircraft Fuel Quantity Indication System

A wireless fuel quantity indication system (FQIS) has been developed using an RFID-enabled sensing platform. The system comprises a fully passive tag, modified reader protocol, capacitive fuel probe, and auxiliary antenna for additional energy harvesting. Results of fluid testing show sensitivity to changes in fluid height of less than 0.25in. An RF-DC harvesting circuit was developed, which delivers up to 5dBm of input power through a remote radio frequency (RF) source. Testing was conducted in a loaded reverberation chamber to emulate the fuel tank environment. Results demonstrate feasibility of the remote source to power the sensor with less than 1W of maximum transmit power and under 100ms dwell time (100mW average power) into the tank. This indicates adequate coverage for large transport aircraft at safe operating levels with a sample rate of up to 1 sample/s.

Direct Link-Aware Optimal Relay Selection and a Low Feedback Variant for Underlay CR

Cooperative relaying combined with selection has been extensively studied in the literature to improve the performance of interference-constrained secondary users in underlay cognitive radio (CR). We present a novel symbol error probability (SEP)-optimal amplify-and-forward relay selection rule for an average interference-constrained underlay CR system. A fundamental principle, which is unique to average interference-constrained underlay CR, that the proposed rule brings out is that the choice of the optimal relay is affected not just by the source-to-relay, relay-to-destination, and relay-to-primary receiver links, which are local to the relay, but also by the direct source-to-destination (SD) link, even though it is not local to any relay. We also propose a simpler, practically amenable variant of the optimal rule called the 1-bit rule, which requires just one bit of feedback about the SD link gain to the relays, and incurs a marginal performance loss relative to the optimal rule. We analyze its SEP and develop an insightful asymptotic SEP analysis. The proposed rules markedly outperform several ad hoc SD link-unaware rules proposed in the literature. They also generalize the interference-unconstrained and SD link-unaware optimal rules considered in the literature.

QoS Aware Trust Metric based Framework for Wireless Sensor Networks

Wireless Sensor Networks have gained popularity due to their real time applications and low-cost nature. These networks provide solutions to scenarios that are critical, complicated and sensitive like military fields, habitat monitoring, and disaster management. The nodes in wireless sensor networks are highly resource constrained. Routing protocols are designed to make efficient utilization of the available resources in communicating a message from source to destination. In addition to the resource management, the trustworthiness of neighboring nodes or forwarding nodes and the energy level of the nodes to keep the network alive for longer duration is to be considered. This paper proposes a QoS Aware Trust Metric based Framework for Wireless Sensor Networks. The proposed framework safeguards a wireless sensor network from intruders by considering the trustworthiness of the forwarder node at every stage of multi-hop routing. Increases network lifetime by considering the energy level of the node, prevents the adversary from tracing the route from source to destination by providing path variation. The framework is built on NS2 Simulator. Experimental results show that the framework provides energy balance through establishment of trustworthy paths from the source to the destination. (C) 2015 The Authors. Published by Elsevier B.V.

Emergent Intelligence Based QoS Routing in MANET

Mobile Ad hoc Networks (MANETs) are self-organized, infrastructureless, decentralized wireless networks consist of a group of heterogeneous mobile devices. Due to the inherent characteristics of MANE -Ts, such as frequent change of topology, nodes mobility, resource scarcity, lack of central control, etc., makes QoS routing is the hardest task. QoS routing is the task of routing data packets from source to destination depending upon the QoS resource constraints, such as bandwidth, delay, packet loss rate, cost, etc. In this paper, we proposed a novel scheme of providing QoS routing in MANETs by using Emergent Intelligence (El). The El is a group intelligence, which is derived from the periodical interaction among a group of agents and nodes. We logically divide MANET into clusters by centrally located static agent, and in each cluster a mobile agent is deployed. The mobile agent interacts with the nodes, neighboring mobile agents and static agent for collection of QoS resource information, negotiations, finding secure and reliable nodes and finding an optimal QoS path from source to destination. Simulation and analytical results show that the effectiveness of the scheme. (C) 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.ore/licenscs/by-nc-nd/4.0/). Peer-review under responsibility of the Conference Program Chairs

PLUTO plus : Near-Complete Modeling of Affine Transformations for Parallelism and Locality

Affine transformations have proven to be very powerful for loop restructuring due to their ability to model a very wide range of transformations. A single multi-dimensional affine function can represent a long and complex sequence of simpler transformations. Existing affine transformation frameworks like the Pluto algorithm, that include a cost function for modern multicore architectures where coarse-grained parallelism and locality are crucial, consider only a sub-space of transformations to avoid a combinatorial explosion in finding the transformations. The ensuing practical tradeoffs lead to the exclusion of certain useful transformations, in particular, transformation compositions involving loop reversals and loop skewing by negative factors. In this paper, we propose an approach to address this limitation by modeling a much larger space of affine transformations in conjunction with the Pluto algorithm's cost function. We perform an experimental evaluation of both, the effect on compilation time, and performance of generated codes. The evaluation shows that our new framework, Pluto+, provides no degradation in performance in any of the Polybench benchmarks. For Lattice Boltzmann Method (LBM) codes with periodic boundary conditions, it provides a mean speedup of 1.33x over Pluto. We also show that Pluto+ does not increase compile times significantly. Experimental results on Polybench show that Pluto+ increases overall polyhedral source-to-source optimization time only by 15%. In cases where it improves execution time significantly, it increased polyhedral optimization time only by 2.04x.

Direct Link-Aware Relay Selection for Average Interference-Constrained Underlay Cognitive Radio

Cooperative relaying combined with selection exploits spatial diversity to significantly improve the performance of interference-constrained secondary users in an underlay cognitive radio network. We present a novel and optimal relay selection (RS) rule that minimizes the symbol error probability (SEP) of an average interference-constrained underlay secondary system that uses amplify-and-forward relays. A key point that the rule highlights for the first time is that, for the average interference constraint, the signal-to-interference-plus-noise-ratio (SINR) of the direct source-to-destination (SI)) link affects the choice of the optimal relay. Furthermore, as the SINR increases, the odds that no relay transmits increase. We also propose a simpler, more practical, and near-optimal variant of the optimal rule that requires just one bit of feedback about the state of the SD link to the relays. Compared to the SD-unaware ad hoc RS rules proposed in the literature, the proposed rules markedly reduce the SEP by up to two orders of magnitude.

Ink-jet printing of carbon nanotube thin film transistors

Ink-jet printing is an important process for placing active electronics on plastic substrates. We demonstrate ink-jet printing as a viable method for large area fabrication of carbon nanotube (CNT) thin film transistors (TFTs). We investigate different routes for producing stable CNT solutions ("inks"). These consist of dispersion methods for CNT debundling and the use of different solvents, such as N -methyl-2-pyrrolidone. The resulting printable inks are dispensed by ink-jet onto electrode bearing silicon substrates. The source to drain electrode gap is bridged by percolating networks of CNTs. Despite the presence of metallic CNTs, our devices exhibit field effect behavior, with effective mobility of ∼0.07 cm2 /V s and ON/OFF current ratio of up to 100. This result demonstrates the feasibility of ink-jet printing of nanostructured materials for TFT manufacture. © 2007 American Institute of Physics.

Experiment on the thermocapillary convection of a mercury liquid bridge in a floating half zone

A liquid bridge of a floating half zone consisting of liquid mercury sealed in a glass tube with nitrogen atmosphere was used for the experiment of thermocapillary convection with a low Prandtl number liquid. A non-contacted diagnostic method was developed to monitor the surface flow and the surface oscillation. A growing surface film (or skin) is a crucial source to suppress thermocapillary convection, and is discussed in this paper. For the case of a mercury Liquid bridge, the critical Marangoni number was obtained as 900, and the oscillatory frequency was around 5 Hz.

A Breakdown Criterion of Free Molecular Flows and an Optimum Analysis of EBPVD

Two important issues in electron beam physical vapor deposition (EBPVD) are addressed. The first issue is a validity condition of the classical cosine law widely used in the engineering context. This requires a breakdown criterion of the free molecular assumption on which the cosine law is established. Using the analytical solution of free molecular effusion flow, the number of collisions (N-c) for a particle moving from an evaporative source to a substrate is estimated that is proven inversely proportional to the local Knudsen number at the evaporation surface. N-c = 1 is adopted as a breakdown criterion of the free molecular assumption, and it is verified by experimental data and DSMC results. The second issue is how to realize the uniform distributions of thickness and component over a large-area thin film. Our analysis shows that at relatively low evaporation rates the goal is easy achieved through arranging the evaporative source positions properly and rotating the substrate.

Flow similariy applied in convection-based tilt sensor

The environment temperature has inevitable effects on property of the convect ion-based tilt sensors. It not only redefines the application, but also prevents the improvement of the sensor performance. Numerical simulation of the fluid flow in the chamber of a sensor was performed and the influence of the environment temperature was studied in this paper. At zero tilt angle, the temperature distribution along the perpendicular line cross the heat source at various environment temperatures was presented. It was found that the flow varied dramatically at different environment temperatures, which would cause the output signal vary accordingly, even when the tilt angle was kept at a constant, because this device works by sensing the change of flow. At the same condition, we present the numerical results when the temperature difference across the heat source and the environment was kept at the same, in those results, it was found that the temperature difference at every point along the perpendicular line cross the heat source keep the same, this result confirms the similarity principle of nature convection. Second, A method of eliminating environment temperature infect on property of convect ion-based tilt sensor, which is based on the theory of flow similarity, is proposed. It was found that a thermal transistance can be piped on the circuit of heat source to compensate the temperature of the heat source. A compensative circuit was specially designed which can keep flow similarity by changing heat source temperature in order to eliminate the influence of environment temperature. The experiment results show that above 70% temperature drift can be eliminated by this compensative circuit.

Wave propagation in an inhomogeneous medium

This paper is in two parts. In the first part we give a qualitative study of wave propagation in an inhomogeneous medium principally by geometrical optics and ray theory. The inhomogeneity is represented by a sound-speed profile which is dependent upon one coordinate, namely the depth; and we discuss the general characteristics of wave propagation which result from a source placed on the sound channel axis. We show that our mathematical model of the sound- speed in the ocean actually predicts some of the behavior of the observed physical phenomena in the underwater sound channel. Using ray theoretic techniques we investigate the implications of our profile on the following characteristics of SOFAR propagation: (i) the sound energy traveling further away from the axis takes less time to travel from source to receiver than sound energy traveling closer to the axis, (ii) the focusing of sound energy in the sound channel at certain ranges, (iii) the overall ray picture in the sound channel.

In the second part a more penetrating quantitative study is done by means of analytical techniques on the governing equations. We study the transient problem for the Epstein profile by employing a double transform to formally derive an integral representation for the acoustic pressure amplitude, and from this representation we obtain several alternative representations. We study the case where both source and receiver are on the channel axis and greatly separated. In particular we verify some of the earlier results derived by ray theory and obtain asymptotic results for the acoustic pressure in the far-field.