Automatic Data Allocation and Buffer Management for Multi-GPU Machines

Multi-GPU machines are being increasingly used in high-performance computing. Each GPU in such a machine has its own memory and does not share the address space either with the host CPU or other GPUs. Hence, applications utilizing multiple GPUs have to manually allocate and manage data on each GPU. Existing works that propose to automate data allocations for GPUs have limitations and inefficiencies in terms of allocation sizes, exploiting reuse, transfer costs, and scalability. We propose a scalable and fully automatic data allocation and buffer management scheme for affine loop nests on multi-GPU machines. We call it the Bounding-Box-based Memory Manager (BBMM). BBMM can perform at runtime, during standard set operations like union, intersection, and difference, finding subset and superset relations on hyperrectangular regions of array data (bounding boxes). It uses these operations along with some compiler assistance to identify, allocate, and manage data required by applications in terms of disjoint bounding boxes. This allows it to (1) allocate exactly or nearly as much data as is required by computations running on each GPU, (2) efficiently track buffer allocations and hence maximize data reuse across tiles and minimize data transfer overhead, and (3) and as a result, maximize utilization of the combined memory on multi-GPU machines. BBMM can work with any choice of parallelizing transformations, computation placement, and scheduling schemes, whether static or dynamic. Experiments run on a four-GPU machine with various scientific programs showed that BBMM reduces data allocations on each GPU by up to 75% compared to current allocation schemes, yields performance of at least 88% of manually written code, and allows excellent weak scaling.

Data acquisition and processing at ocean bottom for a Tsunami warning system

The design and development of a Bottom Pressure Recorder for a Tsunami Early Warning System is described here. The special requirements that it should satisfy for the specific application of deployment at ocean bed and pressure monitoring of the water column above are dealt with. A high-resolution data digitization and low circuit power consumption are typical ones. The implementation details of the data sensing and acquisition part to meet these are also brought out. The data processing part typically encompasses a Tsunami detection algorithm that should detect an event of significance in the background of a variety of periodic and aperiodic noise signals. Such an algorithm and its simulation are presented. Further, the results of sea trials carried out on the system off the Chennai coast are presented. The high quality and fidelity of the data prove that the system design is robust despite its low cost and with suitable augmentations, is ready for a full-fledged deployment at ocean bed. (C) 2013 Elsevier Ltd. All rights reserved.

Effect of zinc content on the processing map for hot working of α brass

The hot deformation behavior of α brass with varying zinc contents in the range 3%–30% was characterized using hot compression testing in the temperature range 600–900 °C and strain rate range 0.001–100 s−1. On the basis of the flow stress data, processing maps showing the variation of the efficiency of power dissipation (given by Image where m is the strain rate sensitivity) with temperature and strain rate were obtained. α brass exhibits a domain of dynamic recrystallization (DRX) at temperatures greater than 0.85Tm and at strain rates lower than 1 s−1. The maximum efficiency of power dissipation increases with increasing zinc content and is in the range 33%–53%. The DRX domain shifts to lower strain rates for higher zinc contents and the strain rate for peak efficiency is in the range 0.0001–0.05 s−1. The results indicate that the DRX in α brass is controlled by the rate of interface formation (nucleation) which depends on the diffusion-controlled process of thermal recovery by climb.

Effect of Zirconium on the Processing MAPS For Hot-Working of Alpha-Beta Brass

The effect of zirconium on the hot working characteristics of alpha and alpha-beta brass was studied in the temperature range of 500 to 850-degrees-C and the strain rate range of 0.001 to 100 s-1. On the basis of the flow stress data, processing maps showing the variation of the efficiency of power dissipation (given by [2m/(m+1)] where m is the strain rate sensitivity) with temperature and strain rate were obtained. The addition of zirconium to alpha brass decreased the maximum efficiency of power dissipation from 53 to 39%, increased the strain rate for dynamic recrystallization (DRX) from 0.001 to 0.1 s-1 and improved the hot workability. Alpha-beta brasses with and without zirconium exhibit a domain in the temperature range from 550 to 750-degrees-C and at strain rates lower than 1 s-1 with a maximum efficiency of power dissipation of nearly 50 % occurring in the temperature range of 700 to 750-degrees-C and a strain rate of 0.001 s-1. In the domain, the alpha phase undergoes DRX and controls the hot deformation of the alloy whereas the beta phase deforms superplastically. The addition of zirconium to alpha-beta brass has not affected the processing maps as it gets partitioned to the beta phase and does not alter the constitutive behavior of the alpha phase

Processing maps for hot working of Cu-Ni-Zn alloys. Part I - Alpha-nickel silver

The constitutive behaviour of agr — nickel silver in the temperature range 700–950 °C and strain rate range 0.001–100 s–1 was characterized with the help of a processing map generated on the basis of the principles of the ldquodynamic materials modelrdquo of Prasadet al Using the flow stress data, processing maps showing the variation of the efficiency of power dissipation (given by 2m/(m+1) wherem is the strain-rate sensitivity) with temperature and strain rate were obtained, agr-nickel silver exhibits a single domain at temperatures greater than 750 °C and at strain rates lower than 1s–1, with a maximum efficiency of 38% occurring at about 950 °C and at a strain rate of 0.1 s–1. In the domain the material undergoes dynamic recrystallization (DRX). On the basis of a model, it is shown that the DRX is controlled by the rate of interface formation (nucleation) which depends on the diffusion-controlled process of thermal recovery by climb. At high strain rates (10 and 100s–1) the material undergoes microstructural instabilities, the manifestations of which are in the form of adiabatic shear bands and strain markings.

Processing maps for hot working of Cu-Ni-Zn alloys Part II Alpha-Beta nickel silver

The constitutive behaviour of agr-beta nickel silver in the temperature range 600�850 °C and strainrate range 0.001�100s�1 was characterized with the help of a processing map generated on the principles of the dynamic materials model. On the basis of the flow-stress data, processing maps showing the variation of the efficiency of power dissipation (given by [2m/(m+1)], wherem is the strain-rate sensitivity) with temperature and strain rate were obtained, agr-beta nickel silver exhibits a single domain at temperatures greater than 700 °C and at strain rates lower than 1 s�1 with a maximum efficiency of power dissipation of about 42% occurring at about 850 °C and at 0.1 s�1. In the domain, the agr phase undergoes dynamic recrystallization and controls the deformation of the alloy, while the beta phase deforms superplastically. Optimum conditions for the processing of agr-beta nickel silver are 850 °C and 0.1 s�1. The material undergoes unstable flow at strain rates of 10 and 100 s�1 and in the temperature range 600�750 °C, manifestated in the form of adiabatic shear bands.

An automatic parallelization framework for multicomputers

Several researchers have looked into various issues related to automatic parallelization of sequential programs for multicomputers. But there is a need for a coherent framework which encompasses all these issues. In this paper we present a such a framework which takes best advantage of the multicomputer architecture. We resort to tiling transformation for iteration space partitioning and propose a scheme of automatic data partitioning and dynamic data distribution. We have tried a simple implementation of our scheme on a transputer based multicomputer [1] and the results are encouraging.

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.

An Algorithm to Decide Feasibility of Linear Integer Constraints Occurrng in Decision Tables

Abstract-To detect errors in decision tables one needs to decide whether a given set of constraints is feasible or not. This paper describes an algorithm to do so when the constraints are linear in variables that take only integer values. Decision tables with such constraints occur frequently in business data processing and in nonnumeric applications. The aim of the algorithm is to exploit. the abundance of very simple constraints that occur in typical decision table contexts. Essentially, the algorithm is a backtrack procedure where the the solution space is pruned by using the set of simple constrains. After some simplications, the simple constraints are captured in an acyclic directed graph with weighted edges. Further, only those partial vectors are considered from extension which can be extended to assignments that will at least satisfy the simple constraints. This is how pruning of the solution space is achieved. For every partial assignment considered, the graph representation of the simple constraints provides a lower bound for each variable which is not yet assigned a value. These lower bounds play a vital role in the algorithm and they are obtained in an efficient manner by updating older lower bounds. Our present algorithm also incorporates an idea by which it can be checked whether or not an (m - 2)-ary vector can be extended to a solution vector of m components, thereby backtracking is reduced by one component.

Heat-Capacity Measurements - Progress In Experimental-Techniques

The heat capacity of a substance is related to the structure and constitution of the material and its measurement is a standard technique of physical investigation. In this review, the classical methods are first analyzed briefly and their recent extensions are summarized. The merits and demerits of these methods are pointed out. The newer techniques such as the a.c. method, the relaxation method, the pulse methods, the laser flash calorimetry and other methods developed to extend the heat capacity measurements to newer classes of materials and to extreme conditions of sample geometry, pressure and temperature are comprehensively reviewed. Examples of recent work and details of the experimental systems are provided for each method. The introduction of automation in control systems for the monitoring of the experiments and for data processing is also discussed. Two hundred and eight references and 18 figures are used to illustrate the various techniques.

A Real-time clustering system for spatio-temporal signals from network of neurons

Over past few years, the studies of cultured neuronal networks have opened up avenues for understanding the ion channels, receptor molecules, and synaptic plasticity that may form the basis of learning and memory. The hippocampal neurons from rats are dissociated and cultured on a surface containing a grid of 64 electrodes. The signals from these 64 electrodes are acquired using a fast data acquisition system MED64 (Alpha MED Sciences, Japan) at a sampling rate of 20 K samples with a precision of 16-bits per sample. A few minutes of acquired data runs in to a few hundreds of Mega Bytes. The data processing for the neural analysis is highly compute-intensive because the volume of data is huge. The major processing requirements are noise removal, pattern recovery, pattern matching, clustering and so on. In order to interface a neuronal colony to a physical world, these computations need to be performed in real-time. A single processor such as a desk top computer may not be adequate to meet this computational requirements. Parallel computing is a method used to satisfy the real-time computational requirements of a neuronal system that interacts with an external world while increasing the flexibility and scalability of the application. In this work, we developed a parallel neuronal system using a multi-node Digital Signal processing system. With 8 processors, the system is able to compute and map incoming signals segmented over a period of 200 ms in to an action in a trained cluster system in real time.

Health Monitoring Of Composite Structures Based On Acoustic Wave Sensing Using Fiber Optic Sensors

Structural Health Monitoring has gained wide acceptance in the recent past as a means to monitor a structure and provide an early warning of an unsafe condition using real-time data. Utilization of structurally integrated, distributed sensors to monitor the health of a structure through accurate interpretation of sensor signals and real-time data processing can greatly reduce the inspection burden. The rapid improvement of the Fiber Optic Sensor technology for strain, vibration, ultrasonic and acoustic emission measurements in recent times makes it feasible alternative to the traditional strain gauges, PVDF and conventional Piezoelectric sensors used for Non Destructive Evaluation (NDE) and Structural Health Monitoring (SHM). Optical fiber-based sensors offer advantages over conventional strain gauges, and PZT devices in terms of size, ease of embedment, immunity from electromagnetic interference (EMI) and potential for multiplexing a number of sensors. The objective of this paper is to demonstrate the acoustic wave sensing using Extrinsic Fabry-Perot Interferometric (EFPI) sensor on a GFRP composite laminates. For this purpose experiments have been carried out initially for strain measurement with Fiber Optic Sensors on GFRP laminates with intentionally introduced holes of different sizes as defects. The results obtained from these experiments are presented in this paper. Numerical modeling has been carried out to obtain the relationship between the defect size and strain.

Structural Health Monitoring (SHM) Using Strain Gauges, PVDF Film and Fiber Bragg Grating (FBG) Sensors: A Comparative Study

The term Structural Health Monitoring has gained wide acceptance in the recent pastas a means to monitor a structure and provide an early warning of an unsafe conditionusing real-time data. Utilization of structurally integrated, distributed sensors tomonitor the health of a structure through accurate interpretation of sensor signals andreal-time data processing can greatly reduce the inspection burden. The rapidimprovement of the Fiber Bragg Grating sensor technology for strain, vibration andacoustic emission measurements in recent times make them a feasible alternatives tothe traditional strain gauges transducers and conventional Piezoelectric sensors usedfor Non Destructive Evaluation (NDE) and Structural Health Monitoring (SHM).Optical fiber-based sensors offers advantages over conventional strain gauges, PVDFfilm and PZT devices in terms of size, ease of embedment, immunity fromelectromagnetic interference(EMI) and potential for multiplexing a number ofsensors. The objective of this paper is to demonstrate the feasibility of Fiber BraggGrating sensor and compare its utility with the conventional strain gauges and PVDFfilm sensors. For this purpose experiments are being carried out in the laboratory on acomposite wing of a mini air vehicle (MAV). In this paper, the results obtained fromthese preliminary experiments are discussed.

Ultrasonic evaluation of delamination in quasi-isotropic CFRP laminates subjected to low-velocity impact

Ultrasonic C-Scan is used very often to detect flaws and defects in the composite components resulted during fabrication and damages resulting from service conditions. Evaluation and characterization of defects and damages of composites require experience and good understanding of the material as they are distinctly different in composition and behavior as compared to conventional metallic materials. The failure mechanisms in composite materials are quite complex. They involve the interaction of matrix cracking, fiber matrix interface debonding, fiber pullout, fiber fracture and delamination. Generally all of them occur making the stress and failure analysis very complex. Under low-velocity impact loading delamination is observed to be a major failure mode. In composite materials the ultrasonic waves suffer high acoustic attenuation and scattering effect, thus making data interpretation difficult. However these difficulties can be overcome to a greater extent by proper selection of probe, probe parameter settings like pulse width, pulse amplitude, pulse repetition rate, delay, blanking, gain etc., and data processing which includes image processing done on the image obtained by the C-Scan.

Calibration and linearity verification of capacitance type cryo level indicators using cryogenically multiplexed diode array

In space application the precision level measurement of cryogenic liquids in the storage tanks is done using triple redundant capacitance level sensor, for control and safety point of view. The linearity of each sensor element depends upon the cylindricity and concentricity of the internal and external electrodes. The complexity of calibrating all sensors together has been addressed by two step calibration methodology which has been developed and used for the calibration of six capacitance sensors. All calibrations are done using Liquid Nitrogen (LN2) as a cryogenic fluid. In the first step of calibration, one of the elements of Liquid Hydrogen (LH2) level sensor is calibrated using 700mm eleven point discrete diode array. Four wire method has been used for the diode array. Thus a linearity curve for a single element of LH2 is obtained. In second step of calibration, using the equation thus obtained for the above sensor, it is considered as a reference for calibrating remaining elements of the same LH2 sensor and other level sensor (either Liquid Oxygen (LOX) or LH2). The elimination of stray capacitance for the capacitance level probes has been attempted. The automatic data logging of capacitance values through GPIB is done using LabVIEW 8.5.