Amount Of Data Transfer Research Articles

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 GPU implementation for LBG and SOM training

Vector quantization (VQ) is an effective technique applicable in a wide range of areas, such as image compression and pattern recognition. The most time-consuming procedure of VQ is codebook training, and two of the frequently used training algorithms are LBG and self-organizing map (SOM). Nowadays, desktop computers are usually equipped with programmable graphics processing units (GPUs), whose parallel data-processing ability is ideal for codebook training acceleration. Although there are some GPU algorithms for LBG training, their implementations suffer from a large amount of data transfer between CPU and GPU and a large number of rendering passes within a training iteration. This paper presents a novel GPU-based training implementation for LBG and SOM training. More specifically, we utilize the random write ability of vertex shader to reduce the overheads mentioned above. Our experimental results show that our approach can run four times faster than the previous approach.

Exploring parallel capabilities of an innovative numerical method for recovering image velocity vectors field

In this paper an efficient method devoted to estimate the velocity vectors field is investigated. The method is based on a quasi-interpolant operator and involves a large amount of computation. The operations characterizing the computational scheme are ideal for parallel processing because they are local, regular and repetitive. Therefore, the spatial parallelism of the process is studied to rapidly proceed in the computation on distributed multiprocessor systems. The process has shown to be synchronous, with good task balancing and requiring a small amount of data transfer.

A hardware acceleration of the time domain boundary integral equation method for the wave equation in two dimensions

We aimed at reducing the runtime of the time domain boundary integral equation method (TDBIEM) for the two-dimensional scalar wave equation by the help of a special-purpose computer MDGRAPE-2, which is a high-performance vector-parallel computer dedicated to computing long-range forces in classical molecular dynamics simulations. To this end, we presented a formulation to make MDGRAPE-2 compute the layer potential that is the most time-consuming in a run of the TDBIEM, considering to minimise the amount of data-transfer between MDGRAPE-2 and the host computer. The newly implemented TDBIEM code using three MDGRAPE-2s boards was about 109 times faster than the conventional code in an analysis of 5400 boundary elements and 640 time steps. We also demonstrated that the present hardware acceleration helps to perform large-scale simulations such as light propagation in photonic crystals.

Mobile IPv6‐based efficient vertical handoff approach for heterogeneous wireless networks

AbstractThe demand and amount of data transfer in heterogeneous wireless network increases as the techniques of the third/fourth generation mobile communications (3G/4G) are progressed significantly. Vertical handoff provides continuous and seamless transfer between various wireless mobile networks, which have different data transmission techniques and standards such as GPRS, UMTS, cdma2000 and IEEE 802.11 family WLANs. In contrast to the horizontal handoff, the vertical handoff should consider not only the strength of signal power but also QoS classes, network bandwidth and service types, etc. Most of the previous works proposed the received signal strength (RSS) based methods to set thresholds for making the decision of performing vertical handoff or not. The RSS‐based approach yields serious ping‐pong effect when the mobile node moves around the overlay area of two heterogeneous wireless networks, which causes unnecessary handoff and increases the handoff overhead. Although the RSS‐based approach with the hysteresis technique decreases the number of unnecessary handoffs, it suffers from large overhead of handoff and causes several drawbacks, including low network throughput, long handoff delay, and high dropping probability. Therefore, an efficient cost‐based vertical handoff algorithm is proposed herein to reduce unnecessary handoffs while increasing network throughput, decreasing handoff delay, and avoiding connection dropping. Two primary motivations of this paper include the prediction of node mobility from RSS and the adaptive cost‐based competitive on‐line (COL) method to determine the optimal network to handoff to from all available candidates. Furthermore, message flows of vertical handoffs between 3G/UMTS and WLAN as well as horizontal handoff between WLANs with MIPv6 are expressed in detail. Finally, the handoff delays are analyzed for the proposed MIPv6 handoff. Numerical results demonstrate that the proposed cost‐based approach outperforms other approaches in the number of vertical handoffs and connection dropping while yielding competitive throughput. Copyright © 2006 John Wiley & Sons, Ltd.

Client-server speech processing system, apparatus, method, and storage medium

The system implements high-accuracy speech recognition while suppressing the amount of data transfer between the client and server. For this purpose, the client compression-encodes speech parameters by a speech processing unit, and sends the compression-encoded speech parameters to the server. The server receives the compression-encoded speech parameters, a speech processing unit makes speech recognition of the compression-encoded speech parameters, and sends information corresponding to the speech recognition result to the client.

A Parallel Block Scaled Gradient Method With Decentralized Step‐Size For Block Additive Unconstrained Optimization Problems Of Large Distributed Systems

ABSTRACTIn this paper, we propose a modified parallel block scaled gradient method for solving block additive unconstrained optimization problems of large distributed systems. Our method makes two major modifications to the typical parallel block scaled gradient method: First, we include a pre‐processing step which reduces the computational time; second, we propose a decentralized Armijo‐type step‐size rule. This rule circumvents the difficulty of determining a step‐size in a distributed computing environment and enables the proposed parallel algorithm to execute in a distributed computer network with a limited amount of data transfer.We have applied our method to the weighted‐least‐square problems of power system state estimation and demonstrated the convergence of our method by testing numerous examples on a PC network. The speedup ratio of the distributed version of our method tends to increase proportionally with the number of subsystems (or computers).

A Custom Image Convolution DSP with a Sustained Calculation Capacity of >1 GMAC/s and Low I/O Bandwidth

A customized processor for real time image convolution has been designed to increase the performance of an instrument for automated cereal grain quality assessment. Image convolution requires an extensive amount of calculation capacity and a corresponding amount of data transfers, hard to achieve with standard processors in real time. Therefore, a tailored architecture with a streamlined dataflow has been developed with emphasis on a system design perspective. The designed processor has a sustained calculation capacity of >1 GMAC/s and on-chip line buffers reduce the amount of external data transfers. Hence, the complexity of the designed processor has been increased to gain a lower complexity of the complete system. To achieve powerful and versatile filtering the size of the programmable kernel functions have been maximized to 15 × 15.

Performance optimization of VXI-based measurement stations

The paper deals with a procedure for estimating the communication times in VXI measurement stations, mainly used in industrial environment with the aim of optimizing their global performances. At first, the proposed procedure analyzes the communication process in order to share the global communication time into elementary time intervals. Then, it evaluates each of them by a two channel oscilloscope used to probe the handshaking lines of the communication protocols. The basic intervals are then used to find general expressions of the communication time for any amount of data transfers with the same measurement station. On this basis, the factors responsible for the throughput loss deriving from communication bottlenecks can be put in evidence. Thus, the proposed procedure provides an useful investigation tool for increasing the station performance. In order to best put in evidence the benefits of the proposed procedure, its experimental application to a general purpose VXI measurement station is discussed as well. >

Schur complement preconditioned conjugate gradient methods for spline collocation equations

We are interested in the efficient solution of linear second order Partial Differential Equation (PDE) problems on rectangular domains. The PDE discretisation scheme used is of Finite Element type and is based on quadratic splines and the collocation methodology. We integrate the Quadratic Spline Collocation (QSC) discretisation scheme with a Domain Decomposition (DD) technique. We develop DD motivated orderings of the QSC equations and unknowns and apply the Preconditioned Conjugate Gradient (PCG) method for the solution of the Schur Complement (SC) system. Our experiments show that the SC-PCG-QSC method in its sequential mode is very efficient compared to standard direct band solvers for the QSC equations. We have implemented the SC-PCG-QSC method on the iPSC/2 hypercube and present performance evaluation results for up to 32 processors configurations. We discuss a type of nearest neighbour communication scheme, in which the amount of data transfer per processor does not grow with the number of processors. The estimated efficiencies are at the order of 90%.

Parallel ODE solvers

We are interested in the efficient solution of linear second order Partial Differential Equation (PDE) problems on rectangular domains. The PDE discretisation scheme used is of Finite Element type and is based on quadratic splines and the collocation methodology. We integrate the Quadratic Spline Collocation (QSC) discretisation scheme with a Domain Decomposition (DD) technique. We develop DD motivated orderings of the QSC equations and unknowns and apply the Preconditioned Conjugate Gradient (PCG) method for the solution of the Schur Complement (SC) system. Our experiments show that the SC-PCG-QSC method in its sequential mode is very efficient compared to standard direct band solvers for the QSC equations. We have implemented the SC-PCG-QSC method on the iPSC/2 hypercube and present performance evaluation results for up to 32 processors configurations. We discuss a type of nearest neighbour communication scheme, in which the amount of data transfer per processor does not grow with the number of processors. The estimated efficiencies are at the order of 90%.