Synergistic Execution of Stream Programs on Multicores with Accelerators

The StreamIt programming model has been proposed to exploit parallelism in streaming applications oil general purpose multicore architectures. The StreamIt graphs describe task, data and pipeline parallelism which can be exploited on accelerators such as Graphics Processing Units (GPUs) or CellBE which support abundant parallelism in hardware. In this paper, we describe a novel method to orchestrate the execution of if StreamIt program oil a multicore platform equipped with an accelerator. The proposed approach identifies, using profiling, the relative benefits of executing a task oil the superscalar CPU cores and the accelerator. We formulate the problem of partitioning the work between the CPU cores and the GPU, taking into account the latencies for data transfers and the required buffer layout transformations associated with the partitioning, as all integrated Integer Linear Program (ILP) which can then be solved by an ILP solver. We also propose an efficient heuristic algorithm for the work-partitioning between the CPU and the GPU, which provides solutions which are within 9.05% of the optimal solution on an average across the benchmark Suite. The partitioned tasks are then software pipelined to execute oil the multiple CPU cores and the Streaming Multiprocessors (SMs) of the GPU. The software pipelining algorithm orchestrates the execution between CPU cores and the GPU by emitting the code for the CPU and the GPU, and the code for the required data transfers. Our experiments on a platform with 8 CPU cores and a GeForce 8800 GTS 512 GPU show a geometric mean speedup of 6.94X with it maximum of 51.96X over it single threaded CPU execution across the StreamIt benchmarks. This is a 18.9% improvement over it partitioning strategy that maps only the filters that cannot be executed oil the GPU - the filters with state that is persistent across firings - onto the CPU.

An investigation of well-characterized small gold clusters by photoelectron spectroscopy, tunneling spectroscopy, and cognate techniques

After microscopic characterization of the size distributions of gold clusters, deposited on carbon substrates by vacuum evaporation or by soft landing, Au(4f') binding energy of the clusters has been measured as a function of the mean cluster size. Similar measurements have been carried out on Au clusters prepared from sols by chemical means and high-nuclearity cluster compounds. In general, small clusters with a mean diameter of $2 nm show significantly larger binding energies than the bulk metal value, due to the onset of nonmetallicity. Nonmetallicity manifests itself in terms of a tunneling conductance gap only in clusters of diameter ;5 1 nm containing 40 atoms or fewer.

A study of chemically prepared small metal clusters in the nonmetallic regime

Small gold clusters [mean diameter (d)[less, similar] 1.4 nm], unlike larger clusters, show a higher Au(4f) binding energy relative to the bulk value and the presence of a conductance gap in tunnelling measurements, just as the molecular cluster compound, Au55(PPh3)12Cl6; small platinum clusters show similar nonmetallic features.

Bubble size measurement in a gas-liquid ejector for a sodium chloride-air system

A high speed photographic technique has been employed to measure the Sauter mean diameter of bubbles experimentally in a gas liquid ejector using a sodium chloride-air system. The measured values are compared with the theoretically predicted maximum bubble size diameter using Sprow's correlation. Bubble size as a function of the liquid flow rate and also of its distance from the throat of the ejector has been reported in this paper. The results obtained for this non-reactive system are also compared with those obtained earlier for the air-water system.

Electrochemical co-deposition of bimetallic Pt-Ru nanoclusters dispersed on poly(3,4-ethylenedioxythiophene) and electrocatalytic behavior for methanol oxidation

Nanoclusters of bimetallic Pt-Ru are electrochemically deposited on conductive polymer, poly(3,4-ethylenedioxythiophene)(PEDOT), which is also electrochemically deposited on a carbon paper substrate. The bimetallic deposition is carried out in an acidic electrolyte consisting of chloroplatinic acid and ruthenium chloride at 0.0 V versus saturated calomel electrode (SCE) on PEDOT coated carbon paper. A thin layer PEDOT on a carbon paper substrate facilitates the formation of uniform, well-dispersed, nano clusters of Pt-Ru of mean diameter of 123 nm, which consist of nanosize particles. In the absence of PEDOT, the size of the clusters is about 251 nm, which are unevenly distributed on carbon paper substrate. Cyclic voltammetry studies suggest that peak currents of methanol oxidation are several times greater on PtRu-PEDOT electrode than on Pt-Ru electrode in the absence of PEDOT. (C) 2011 Elsevier B.V. All rights reserved.

Automatic Compilation of MATLAB Programs for Synergistic Execution on Heterogeneous Processors

MATLAB is an array language, initially popular for rapid prototyping, but is now being increasingly used to develop production code for numerical and scientific applications. Typical MATLAB programs have abundant data parallelism. These programs also have control flow dominated scalar regions that have an impact on the program's execution time. Today's computer systems have tremendous computing power in the form of traditional CPU cores and throughput oriented accelerators such as graphics processing units(GPUs). Thus, an approach that maps the control flow dominated regions to the CPU and the data parallel regions to the GPU can significantly improve program performance. In this paper, we present the design and implementation of MEGHA, a compiler that automatically compiles MATLAB programs to enable synergistic execution on heterogeneous processors. Our solution is fully automated and does not require programmer input for identifying data parallel regions. We propose a set of compiler optimizations tailored for MATLAB. Our compiler identifies data parallel regions of the program and composes them into kernels. The problem of combining statements into kernels is formulated as a constrained graph clustering problem. Heuristics are presented to map identified kernels to either the CPU or GPU so that kernel execution on the CPU and the GPU happens synergistically and the amount of data transfer needed is minimized. In order to ensure required data movement for dependencies across basic blocks, we propose a data flow analysis and edge splitting strategy. Thus our compiler automatically handles composition of kernels, mapping of kernels to CPU and GPU, scheduling and insertion of required data transfer. The proposed compiler was implemented and experimental evaluation using a set of MATLAB benchmarks shows that our approach achieves a geometric mean speedup of 19.8X for data parallel benchmarks over native execution of MATLAB.

Dispersion and vaporization of biofuels and conventional fuels in a crossflow pre-mixer

In lean premixed pre-vaporized (LPP) combustion, controlled atomization, dispersion and vaporization of different types of liquid fuel in the premixer are the key factors required to stabilize the combustion process and improve the efficiency. The dispersion and vaporization process for biofuels and conventional fuels sprayed into a crossflow pre-mixer have been simulated and analyzed with respect to vaporization rate, degree of mixedness and homogeneity. Two major biofuels under investigation are Ethanol and Rapeseed Methyl Esters (RME), while conventional fuels are gasoline and jet-A. First, the numerical code is validated by comparing with the experimental data of single n-heptane and decane droplet evaporating under both moderate and high temperature convective air now. Next, the spray simulations were conducted with monodispersed droplets with an initial diameter of 80 mu m injected into a turbulent crossflow of air with a typical velocity of 10 m/s and temperature of around 800K. Vaporization time scales of different fuels are found to be very different. The droplet diameter reduction and surface temperature rise were found to be strongly dependent on the fuel properties. Gasoline droplet exhibited a much faster vaporization due a combination of higher vapor pressure and smaller latent heat of vaporization compared to other fuels. Mono-dispersed spray was adopted with the expectation of achieving more homogeneous fuel droplet size than poly-dispersed spray. However, the diameter histogram in the zone near the pre-mixer exit shows a large range of droplet diameter distributions for all the fuels. In order to improve the vaporization performance, fuels were pre-heated before injection. Results show that the Sauter mean diameter of ethanol improved from 52.8% of the initial injection size to 48.2%, while jet-A improved from 48.4% to 18.6% and RME improved from 63.5% to 31.3%. The diameter histogram showed improved vaporization performance of jet-A. (C) 2011 Elsevier Ltd. All rights reserved.

Experimental Studies on High-Pressure Spray Structure of Biofuels

The high-pressure spray characteristics of biofuels, specifically, Pongamia oil and its blends with diesel are studied for various gas pressures. Two single-hole solenoid injectors with nozzle diameters of 200 and 260 mu m are used along with a high-pressure common-rail direct-injection system to inject fuel into a high-pressure spray visualization chamber. The spray structure is characterized using a high-speed laser-based shadowgraphy technique. The spray structure of Pongamia oil revealed the presence of an intact liquid core at low gas pressure. At high gas pressures, the spray atomization of the Pongamia oil showed marked improvement. The spray tip penetration of Pongamia oil and its blends with diesel is higher compared to that of diesel for all test conditions. The spray cone angle of Pongamia oil and 50% Pongamia oil blend with diesel is lower as compared to that of diesel. Both these observations are attributed to the presence of large droplets carrying higher momentum in oil and blend. The droplet size is measured at an injection pressure of 1000 bar and gas pressure of 30 bar at 25 mm below the nozzle tip using the particle/droplet image.analysis (PDIA) method. The droplet size measurements have shown that the Sauter mean diameter (SMD) in the spray core of Pongamia oil is more than twice that of diesel. The spray tip penetration of the 20% blend of Pongamia with diesel (P20) is similar to that of diesel but the SMD is 50% higher. Based on experimental data, appropriate spray tip penetration correlation is proposed for the vegetable oil fuels such as Pongamia.

Spray characterization of gasoline-ethanol blends from a multi-hole port fuel injector

This work reports the measured spray structure and droplet size distributions of ethanol-gasoline blends for a low-pressure, multi-hole, port fuel injector (PFI). This study presents previously unavailable data for this class of injectors which are widely used in automotive applications. Specifically, gasoline, ethanol, and gasoline-ethanol blends containing 10%, 20% and 50% ethanol were studied using laser backlight imaging, and particle/droplet image analysis (PDIA) techniques. The fuel mass injected, spray structure and tip penetrations, droplet size distributions, and Sauter mean diameter were determined for the blends, at two different injection pressures. Results indicate that the gasoline and ethanol sprays have similar characteristics in terms of spray progression and droplet sizes in spite of the large difference in viscosity. It appears that the complex mode of atomization utilized in these injectors involving interaction of multiple fuel jets is fairly insensitive to the fuel viscosity over a range of values. This result has interesting ramifications for existing gasoline fuel systems which need to handle blends and even pure ethanol, which is one of the renewable fuels of the future. (C) 2012 Elsevier Ltd. All rights reserved.

Bubble size measurement and error analysis in a gas liquid ejector

Bubble size in a gas liquid ejector has been measured using the image technique and analysed for estimation of Sauter mean diameter. The individual bubble diameter is estimated by considering the two dimensional contour of the ellipse, for the actual three dimensional ellipsoid in the system by equating the volume of the ellipsoid to that of the sphere. It is observed that the bubbles are of oblate and prolate shaped ellipsoid in this air water system. The bubble diameter is calculated based on this concept and the Sauter mean diameter is estimated. The error between these considerations is reported. The bubble size at different locations from the nozzle of the ejector is presented along with their percentage error which is around 18%.

AN EXPERIMENTAL STUDY ON EFFERVESCENT ATOMIZATION OF BIO-OIL FUELS

This paper presents the work on detailed characterization of effervescent spray of Jatropha and Pongamia pure plant oils. The spray characteristics of these biofuels are compared with those of diesel. Both macroscopic and microscopic spray characteristics at different injection pressures and gas-to-liquid ratio (GLR) have been studied. The particle/droplet imaging analysis (PDIA) technique along with direct imaging methods are used for the purpose of spray characterization. Due to their higher viscosity, pure plant oils showed poor atomization compared to diesel and a blend of diesel and pure plant oil at a given GLR. Pure plant oil sprays showed a lower spray cone angle when compared to diesel and blends at lower GLRs. However, the difference is not significant at higher GLRs. Droplet size measurements at 100 mm downstream of the exit orifice showed reduction in Sauter mean diameter (SMD) diameter with increase in GLR. A radial variation in the SMD is observed for the blend and pure plant oils. Pure oils showed a larger variation when compared to the blend. Spray unsteadiness has been characterized based on the image-to-image variation in the mean droplet diameter and fluctuations in the spray cone angle. Results showed that pure plant oil sprays are more unsteady at lower GLRs when compared to diesel and blend. A critical GLR is identified at which the spray becomes steady. The three regimes of spray operation, namely ``steady spray,'' ``pulsating spray,'' and ``spray and unbroken liquid jet'' are identified in the injection pressure-GLR parameter space for these pure plant oils. Two-phase flow imaging inside the exit orifice shows that for the pure plant oils, the flow is highly transient at low GLRs and the bubbly, slug, and annular two-phase flow regimes are all observed. However, at higher GLRs where the spray is steady, only the annular flow regime is observed.

Automatic compilation of MATLAB programs for synergistic execution on heterogeneous processors

MATLAB is an array language, initially popular for rapid prototyping, but is now being increasingly used to develop production code for numerical and scientific applications. Typical MATLAB programs have abundant data parallelism. These programs also have control flow dominated scalar regions that have an impact on the program's execution time. Today's computer systems have tremendous computing power in the form of traditional CPU cores and throughput oriented accelerators such as graphics processing units(GPUs). Thus, an approach that maps the control flow dominated regions to the CPU and the data parallel regions to the GPU can significantly improve program performance. In this paper, we present the design and implementation of MEGHA, a compiler that automatically compiles MATLAB programs to enable synergistic execution on heterogeneous processors. Our solution is fully automated and does not require programmer input for identifying data parallel regions. We propose a set of compiler optimizations tailored for MATLAB. Our compiler identifies data parallel regions of the program and composes them into kernels. The problem of combining statements into kernels is formulated as a constrained graph clustering problem. Heuristics are presented to map identified kernels to either the CPU or GPU so that kernel execution on the CPU and the GPU happens synergistically and the amount of data transfer needed is minimized. In order to ensure required data movement for dependencies across basic blocks, we propose a data flow analysis and edge splitting strategy. Thus our compiler automatically handles composition of kernels, mapping of kernels to CPU and GPU, scheduling and insertion of required data transfer. The proposed compiler was implemented and experimental evaluation using a set of MATLAB benchmarks shows that our approach achieves a geometric mean speedup of 19.8X for data parallel benchmarks over native execution of MATLAB.

A method for measurement of planar liquid volume fraction in dense sprays

A methodology for measurement of planar liquid volume fraction in dense sprays using a combination of Planar Laser-Induced Fluorescence (PLIF) and Particle/Droplet Imaging Analysis (PDIA) is presented in this work. The PLIF images are corrected for loss of signal intensity due to laser sheet scattering, absorption and auto-absorption. The key aspect of this work pertains to simultaneously solving the equations involving the corrected PLIF signal and liquid volume fraction. From this, a quantitative estimate of the planar liquid volume fraction is obtained. The corrected PLIF signal and the corrected planar Mie scattering can be also used together to obtain the Sauter Mean Diameter (SMD) distribution by using data from the PDIA technique at a particular location for calibration. This methodology is applied to non-evaporating sprays of diesel and a more viscous pure plant oil at an injection pressure of 1000 bar and a gas pressure of 30 bar in a high pressure chamber. These two fuels are selected since their viscosity values are very different with a consequently very different spray structure. The spatial distribution of liquid volume fraction and SMD is obtained for two fuels. The proposed method is validated by comparing liquid volume fraction obtained by the current method with data from PDIA technique. (C) 2012 Elsevier Inc. All rights reserved.

Experimental studies on air-assisted impinging jet atomization

In the present study, a novel air-assisted impinging jet atomization is demonstrated. A configuration in which a gas jet is directed on to the impinging point of two liquid jets is used to improve the atomization. The effect of liquid properties such as viscosity and surface tension, angle between liquid jets and gas injection orifice diameter on spray characteristics has been experimentally studied. Backlit imaging and particle/droplet imaging and analysis techniques are utilized to characterize the sprays. The experimental results indicate that the effect of liquid viscosity is significant on the liquid sheet break up formed by the impinging jets. However, surface tension does not affect the spray structure significantly in this mode of atomization. At low liquid jet velocity, the prompt mode of atomization is observed where as atomization occurs in classical mode at higher liquid jet velocity. Results showed that variation in the angle between liquid jets do not affect the breakup phenomenon significantly. The spray angle is computed by finding the angle between the lines joining the impinging point and spray edge at an axial distance of 15 mm downstream of the impinging point from the ensemble-averaged data over 100 spray images. It was observed that effect of liquid jets impinging angle on the spray angle is higher at higher liquid velocity. Higher viscosity liquids exhibit lower spray angles. Droplet size measurements indicate a radial variation in the spray. An overall Sauter Mean Diameter (SMD) value is obtained by combining the droplet statistics at all radial locations at a fixed axial location. A very interesting trend is that the SMD is constant beyond a critical Gas to Liquid Ratio (GLR) and momentum ratio for a large variation in liquid viscosity and surface tension. This observation has important ramifications for fuel flexible systems. (C) 2013 Elsevier Ltd. All rights reserved.

Toward a Scalable Working Set Size Estimation Method and Its Application for Chip Multiprocessors

It is essential to accurately estimate the working set size (WSS) of an application for various optimizations such as to partition cache among virtual machines or reduce leakage power dissipated in an over-allocated cache by switching it OFF. However, the state-of-the-art heuristics such as average memory access latency (AMAL) or cache miss ratio (CMR) are poorly correlated to the WSS of an application due to 1) over-sized caches and 2) their dispersed nature. Past studies focus on estimating WSS of an application executing on a uniprocessor platform. Estimating the same for a chip multiprocessor (CMP) with a large dispersed cache is challenging due to the presence of concurrently executing threads/processes. Hence, we propose a scalable, highly accurate method to estimate WSS of an application. We call this method ``tagged WSS (TWSS)'' estimation method. We demonstrate the use of TWSS to switch-OFF the over-allocated cache ways in Static and Dynamic NonUniform Cache Architectures (SNUCA, DNUCA) on a tiled CMP. In our implementation of adaptable way SNUCA and DNUCA caches, decision of altering associativity is taken by each L2 controller. Hence, this approach scales better with the number of cores present on a CMP. It gives overall (geometric mean) 26% and 19% higher energy-delay product savings compared to AMAL and CMR heuristics on SNUCA, respectively.