Aho-Corasick String Matching on Shared and Distributed-Memory Parallel Architectures

Current trends in high performance and embedded computing include design of increasingly complex hardware architectures with high parallelism, heterogeneous processing elements, and nonuniform communication resources. In order to take hardware and software design decisions, early evaluations of the system nonfunctional properties are needed. These evaluations of system efficiency require electronic system-level information on both algorithms and architecture. Contrary to algorithm models for which a major body of work has been conducted on defining formal models of computation (MoCs), architecture models from the literature are mostly empirical models from which reproducible experimentation requires the accompanying software. In this paper, a precise definition of a model of architecture (MoA) is proposed that focuses on reproducibility and abstraction and removes the overlap previously existing between the notions of MoA and MoC. A first MoA, called the linear system-level architecture model (LSLA), is presented. To demonstrate the generic nature of the proposed new architecture modeling concepts, we show that the LSLA model can be integrated flexibly with different MoCs. LSLA is then used to model the energy consumption of a state-of-the-art multiprocessor system-on-chip (MPSoC) when running an application described using the synchronous dataflow MoC. A method to automatically learn LSLA model parameters from platform measurements is introduced. Despite the high complexity of the underlying hardware and software, a simple LSLA model is demonstrated to estimate the energy consumption of the MPSoC with a fidelity of 86%.

MapReduce was originally proposed as a suitable and efficient approach for analyzing and processing large amounts of data. Since then, many researches contributed with MapReduce implementations for distributed and shared memory architectures. Nevertheless, different architectural levels require different optimization strategies in order to achieve high-performance computing. Such strategies in turn have caused very different MapReduce programming interfaces among these researches. This paper presents some research notes on coding productivity when developing MapReduce applications for distributed and shared memory architectures. As a case study, we introduce our current research on a unified MapReduce domain-specific language with code generation for Hadoop and Phoenix++, which has achieved a coding productivity increase from 41.84% and up to 94.71% without significant performance losses (below 3%) compared to those frameworks.

BackgroundIn the United States, a rare disease is characterized as the one affecting no more than 200,000 patients at a certain period. Patients suffering from rare diseases are often either misdiagnosed or left undiagnosed, possibly due to insufficient knowledge or experience with the rare disease on the part of clinical practitioners. With an exponentially growing volume of electronically accessible medical data, a large volume of information on thousands of rare diseases and their potentially associated diagnostic information is buried in electronic medical records (EMRs) and medical literature.ObjectiveThis study aimed to leverage information contained in heterogeneous datasets to assist rare disease diagnosis. Phenotypic information of patients existed in EMRs and biomedical literature could be fully leveraged to speed up diagnosis of diseases.MethodsIn our previous work, we advanced the use of a collaborative filtering recommendation system to support rare disease diagnostic decision making based on phenotypes derived solely from EMR data. However, the influence of using heterogeneous data with collaborative filtering was not discussed, which is an essential problem while facing large volumes of data from various resources. In this study, to further investigate the performance of collaborative filtering on heterogeneous datasets, we studied EMR data generated at Mayo Clinic as well as published article abstracts retrieved from the Semantic MEDLINE Database. Specifically, in this study, we designed different data fusion strategies from heterogeneous resources and integrated them with the collaborative filtering model.ResultsWe evaluated performance of the proposed system using characterizations derived from various combinations of EMR data and literature, as well as with sole EMR data. We extracted nearly 13 million EMRs from the patient cohort generated between 2010 and 2015 at Mayo Clinic and retrieved all article abstracts from the semistructured Semantic MEDLINE Database that were published till the end of 2016. We applied a collaborative filtering model and compared the performance generated by different metrics. Log likelihood ratio similarity combined with k-nearest neighbor on heterogeneous datasets showed the optimal performance in patient recommendation with area under the precision-recall curve (PRAUC) 0.475 (string match), 0.511 (systematized nomenclature of medicine [SNOMED] match), and 0.752 (Genetic and Rare Diseases Information Center [GARD] match). Log likelihood ratio similarity also performed the best with mean average precision 0.465 (string match), 0.5 (SNOMED match), and 0.749 (GARD match). Performance of rare disease prediction was also demonstrated by using the optimal algorithm. Macro-average F-measure for string, SNOMED, and GARD match were 0.32, 0.42, and 0.63, respectively.ConclusionsThis study demonstrated potential utilization of heterogeneous datasets in a collaborative filtering model to support rare disease diagnosis. In addition to phenotypic-based analysis, in the future, we plan to further resolve the heterogeneity issue and reduce miscommunication between EMR and literature by mining genotypic information to establish a comprehensive disease-phenotype-gene network for rare disease diagnosis.

We propose a parallel synchronous and asynchronous implementation of the coupled simulated annealing (CSA) algorithm in a shared-memory architecture. The original CSA was implemented synchronously in a distributed-memory architecture. It synchronizes at each temperature update, which leads to idling and loss of efficiency when increasing the number of processors. The proposed synchronous CSA (SCSA) is implemented as the original, but in a shared-memory architecture. The proposed asynchronous CSA (ACSA) does not synchronize, allowing a larger parallel efficiency for larger numbers of processors. Results from extensive experiments show that the proposed ACSA presents much better quality of solution when compared to the serial and to the SCSA. The experiments also show that the performance of the proposed ACSA is better than the SCSA for less computationally intensive problems or when a larger number of processing cores are available. Moreover, the parallel efficiency of the ACSA improves by increasing the size of the problem. With the advent of the Multi-core Era, the use of the proposed algorithm becomes more attractive than the original synchronous CSA.

In this paper, the low speed wind over the Great Sphinx is simulated in order to investigate the effects of the wind flow structure on the surface of the statue. The three-dimensional incompressible Navier-Stokes equations are solved based on Chorin-type projection method and using finite difference approximations. The computational mesh of such large scale simulation problems contains several millions of points, therefore the MPI-based parallel solution of Navier-Stokes equations is addressed on distributed-memory parallel architecture. The distributed solution methodology of Navier-Stokes equations is optimized to get the best accuracy and performance. Then, this optimized methodology is used to simulate the low speed northwest wind over the Great Sphinx at Reynolds number of 1000. The wind flow structure over the statue is visualized and intensively studied.

This paper describes extensions to OpenMP that implement data placement features needed for NUMA architectures. OpenMP is a collection of compiler directives and library routines used to write portable parallel programs for shared‐memory architectures. Writing efficient parallel programs for NUMA architectures, which have characteristics of both shared‐memory and distributed‐memory architectures, requires that a programmer control the placement of data in memory and the placement of computations that operate on that data. Optimal performance is obtained when computations occur on processors that have fast access to the data needed by those computations. OpenMP ‐‐ designed for shared‐memory architectures ‐‐ does not by itself address these issues. The extensions to OpenMP Fortran presented here have been mainly taken from High Performance Fortran. The paper describes some of the techniques that the Compaq Fortran compiler uses to generate efficient code based on these extensions. It also describes some additional compiler optimizations, and concludes with some preliminary results.

When a molecule experiences an electric field, its charge distribution is relaxed in response to that field. The energy associated with this relaxation is known as the polarization energy . Computing the polarization energy between a ligand (i.e., a small molecule such as a drug molecule) and a receptor (e.g., a virus molecule) is of utmost importance in drug design, protein-protein docking, virus/bacterium cell analysis, molecular dynamics simulations for determining the molecular conformation with minimal total free energy. We have implemented distributed-memory and distributed shared-memory parallel algorithms for approximating polarization energy of a molecule by extending a prior work for shared-memory (multicore) architectures. This is an octree-based hierarchical algorithm, built on GreengardRokhlin type near and far decomposition of data points (i.e., atoms and points sampled from the molecular surface) which calculates the polarization energy of a molecule using the r6 approximation of Generalized Born (GB) Radii of atoms. Both Poisson-Boltzmann (PB) GeneralizedBorn (GB) models can be used for approximating polarization energy. However, due to high computational costs PB method is rarely used for large molecules such as proteins.

The high performance of the elbow joint angle prediction is essential on the development of the devices based on electromyography (EMG) control. The performance of the prediction depends on the feature of extraction parameters such as window length. In this paper, we evaluated the effect of the window length on the performance of the elbow-joint angle prediction. The prediction algorithm consists of zero-crossing feature extraction and second order of Butterworth low pass filter. The feature was used to extract the EMG signal by varying window length. The EMG signal was collected from the biceps muscle while the elbow was moved in the flexion and extension motion. The subject performed the elbow motion by holding a 1-kg load and moved the elbow in different periods (12 seconds, 8 seconds and 6 seconds). The results indicated that the window length affected the performance of the prediction. The 250 window lengths yielded the best performance of the prediction algorithm of (mean±SD) root mean square error = 5.68%±1.53% and Person’s correlation = 0.99±0.0059.

This paper describes the architecture and implementation of the Data Compression Chip (DCC) for high-performance lossless data compression for multi-channel highspeed data networking applications. The DCC implements an improved adaptive string matching algorithm using a new parallel architecture-the Variable Length String Matcher (VLSM). The VLSM quickly performs an otherwise time-intensive string-matching procedure. Furthermore, by avoiding the typical data structures that require very large and fast memory to store a compression vocabulary, the DCC's VLSM minimizes system costs. By maintaining separate vocabularies for multiple channels, the DCC eliminates the loss of compression efficiency otherwise experienced in typical frame-multiplexed or reset-at-frame applications. The DCC uses a high-performance algorithm with a custom parallel architecture to minimize vocabulary memory (VRAM) requirements and provide very high throughput and compression efficiency for multiple channels. The device has been realized in 0.8 micron CMOS technology with a die area of 9.27 mm/spl times/10.0 mm and power dissipation about 1 W at 20 MHz. It employs a 160 pin PQFP package. >

We provide the first theoretical analysis on the convergence rate of asynchronous mini-batch gradient descent with variance reduction (AsySVRG) for non-convex optimization. Asynchronous stochastic gradient descent (AsySGD) has been broadly used for deep learning optimization, and it is proved to converge with rate of O(1/\sqrt{T}) for non-convex optimization. Recently, variance reduction technique is proposed and it is proved to be able to accelerate the convergence of SGD greatly. It is shown that asynchronous SGD method with variance reduction technique has linear convergence rate when problem is strongly convex. However, there is still no work to analyze the convergence rate of this method for non-convex problem. In this paper, we consider two asynchronous parallel implementations of mini-batch gradient descent method with variance reduction: one is on distributed-memory architecture and the other is on shared-memory architecture. We prove that both methods can converge with a rate of O(1/T) for non-convex optimization, and linear speedup is accessible when we increase the number of workers. We evaluate our methods by optimizing multi-layer neural networks on two real datasets (MNIST and CIFAR-10), and experimental results demonstrate our theoretical analysis.

Since 1964, we had the notion that the instruction set architecture (ISA) is a useful and fairly opaque abstraction layer between hardware and software. Software rode hardware's performance wave while remaining gloriously oblivious to hardware's growing complexity. Unfortunately, the jig is up. We still have ISAs, but the abstraction no longer offers seamless portability---parallel software needs to be tuned for different core counts, and heterogeneous processing elements (CPUs, GPUs, accelerators) further complicate programmability. We are better at building large-scale heterogeneous processors than we are at programming them. Maintaining software across multiple current platforms is difficult and porting to future platforms is also difficult. There have been many technical responses: virtual ISAs (e.g., NVIDIA's PTX), higher-level programming interfaces (e.g., CUDA or OpenCL), and late-stage compilation and platform-specific tailoring (e.g., Android ART), etc. A team of opinionated experts, drawn from the three ASPLOS communities will examine the problem of programmer productivity in the post-ISA world, first from the perspective of their area of expertise and then noting the contributions from the other two communities. What research will save us and how? This wide-ranging debate will frame important research areas for future work while being grounded in frank discussion about what has succeeded in the past. Attendees can expect actionable insight into important research issues as well an entertaining discussion.

This paper addresses the problem of image processing algorithms implementation onto dynamically and reconfigurable architectures. Today, these Systems-on-Chip (SoC), offer the possibility to implement several heterogeneous processing elements in a single chip. It means several processors, few hardware accelerators as well as communication mediums between all these components. Applications for this kind of platform are described with software threads, running on processors, and specific hardware accelerators, running on hardware partitions. This paper focuses on the complex problem of communication management between software and hardware actors for dataflow oriented processing, and proposes solutions to leverage this issue.

A scheme for maintaining a balanced search tree on a distributed memory parallel architecture is described. A general 2/sup P-2/-2/sup P/ (for integer P>or=3) search tree is introduced with a linear array of up to (log/sub 2/ N/(P-2))+1 processors being used to implement such a search tree. As many as ((log/sub 2/ N/(P-2))+1)/2 operations can execute concurrently. Insertion and deletion transformations are described and several search trees are demonstrated on an array of transputer processors. Variations in both the number of processors allocated to the array and the value of P allow the optimal search structure for a given architecture to be determined. >

Just as liquids adapt their shape to the one of their container, liquid architectures feature a high degree of adaptability so that they can provide scalability to applications as they are executed on a wide variety of heterogeneous deployment environments. In this paper we enumerate the properties to be guaranteed by so-called liquid service-oriented architectures and define a set of design constraints that make up a novel architectural style for liquid architectures. These constraints drive the careful construction of a pattern, the Restful Actor (Reactor), which enables to deliver the required scalability by means of replication of its constituent parts. Reactors feature a Restful Web service interface and a composable architecture which is capable of delivering scalability and high performance in a way that is independent from the chosen deployment infrastructure. We discuss how the Reactor can be deployed to run on distributed (shared-nothing) execution environments typical of virtualized Cloud computing environments as well as on modern multicore processors with shared memory architectures.

We present the spectral element method to simulate elastic-wave propagation in realistic geological structures involving complieated free-surface topography and material interfaces for two- and three-dimensional geometries. The spectral element method introduced here is a high-order variational method for the spatial approximation of elastic-wave equations. The mass matrix is diagonal by construction in this method, which drastically reduces the computational cost and allows an efficient parallel implementation. Absorbing boundary conditions are introduced in variational form to simulate unbounded physical domains. The time discretization is based on an energy-momentum conserving scheme that can be put into a classical explicit-implicit predictor/multi-corrector format. Long-term energy conservation and stability properties are illustrated as well as the efficiency of the absorbing conditions. The associated Courant condition behaves as ΔtC < O (nel−1/ndN−2), with nel the number of elements, nd the spatial dimension, and N the polynomial order. In practice, a spatial sampling of approximately 5 points per wavelength is found to be very accurate when working with a polynomial degree of N = 8. The accuracy of the method is shown by comparing the spectral element solution to analytical solutions of the classical two-dimensional (2D) problems of Lamb and Garvin. The flexibility of the method is then illustrated by studying more realistic 2D models involving realistic geometries and complex free-boundary conditions. Very accurate modeling of Rayleigh-wave propagation, surface diffraction, and Rayleigh-to-body-wave mode conversion associated with the free-surface curvature are obtained at low computational cost. The method is shown to provide an efficient tool to study the diffraction of elastic waves by three-dimensional (3D) surface topographies and the associated local effects on strong ground motion. Complex amplification patterns, both in space and time, are shown to occur even for a gentle hill topography. Extension to a heterogeneous hill structure is considered. The efficient implementation on parallel distributed memory architectures will allow to perform real-time visualization and interactive physical investigations of 3D amplification phenomena for seismic risk assessment.