Search

We study the convergence time of local search for a standard machine scheduling problem in which jobs are assigned to identical or related machines. Local search corresponds to the best response dynamics that arises when jobs selfishly try to minimize their costs. We assume that each machine runs a coordination mechanism that determines the order of execution of jobs assigned to it. We obtain various new polynomial and pseudo-polynomial bounds for the well-studied coordination mechanisms Makespan and Shortest-Job-First, using worst-case and smoothed analysis. We also introduce a natural coordination mechanism FIFO, which takes into account the order in which jobs arrive at a machine, and study both its impact on the convergence time and its price of anarchy.

The main objective of this paper is to demonstrate how simple fluid-flow model of TCP (Transport Control Protocol) behavior allows to exploit robust control methods earlier developed for nonlinear time delayed feedback systems. Test setup consists of “far” and “near”, in terms of transport delay, groups of TCP flows and AQM (Active Queue Management) router. It was proposed to control “near” flows for “far” long-lived flows stabilization in uncertain network parameters environment. The comparison of RED (Random Early Detect) queue controller and robust controller simulation and experimental results are presented.

In this paper expressions of outage probability under Nakagami-\(m\) fading for Coded Cooperative Communication have been developed. Performance of outage has been simulated for various factors such as allocated Rate, cooperation level, mean channel SNR and number of relays. Diversity order has been calculated for Coded Cooperative Communication with multiple relays. Outage probability expression for Space Time Cooperation with multiple partners under Nakagami-m have also been developed and analyzed. A critical comparative analysis is done for No Cooperation, Coded Cooperation and Space Time Cooperation. Simulation results highlight the benefits of Coded Cooperation over the other two schemes under the slow fading circumstances.

We analyze the efficiency of servers equipped with state-of-the-art general-purpose multicore processors as well as platforms based on accelerators such as graphics processing units (GPUs) and the Intel Xeon Phi. Following the proposal recently advocated in the High Performance Conjugate Gradient (HPCG) benchmark, we leverage for this purpose efficient implementations of ILUPACK, a preconditioned solver for sparse linear systems that comprises numerical kernels and data access patterns analogous to those of HPCG. Our study analyzes the (computational) performance and energy efficiency, with two different metrics for each: time/floating-point throughput for the former; and energy/floating-point throughput-per-Watt for the latter. This work was supported by the CICYT project TIN2011-23283 of MINECO and FEDER, and the EU Project FP7 318793 “EXA2GREEN”.

We present a method for fast volume rendering using GPGPU for Cloud 3D Visualization. Each of the threads is processed in parallel on the GPU. Our algorithm is to use the computational power of graphics processors to speed up the rendering process of the Shear-Warp algorithm. Our GPU-based method provides real-time frame rates and outperforms the GPU-based implementation. Our experimental results show that our algorithm is much faster than the CPU-based Shear-Warp volume visualization in terms of rendering speed and image quality. It can be launched on GPU computing clusters provided by cloud infrastructures such as Amazon EC2.

Depth First Search (DFS) tree is a fundamental data structure for graphs used in solving various algorithmic problems. However, very few results are known for maintaining DFS tree in a dynamic environment - insertion or deletion of edges. The only non-trivial result for this problem is by Franciosa et al. [4]. They showed that, for a directed acyclic graph on n vertices, a DFS tree can be maintained in O(n) amortized time per edge insertion. They stated it as an open problem to maintain a DFS tree dynamically in an undirected graph or general directed graph.

A priority queue is a fundamental data structure that maintains a dynamic ordered set of keys and supports the followig basic operations: insertion of a key, deletion of a key, and finding the smallest key. The complexity of the priority queue is closely related to that of sorting: A priority queue can be used to implement a sorting algorithm trivially. Thorup [11] proved that the converse is also true in the RAM model. In particular, he designed a priority queue that uses the sorting algorithm as a black box, such that the per-operation cost of the priority queue is asymptotically the same as the per-key cost of sorting. In this paper, we prove an analogous result in the external memory model, showing that priority queues are computationally equivalent to sorting in external memory, under some mild assumptions. The reduction provides a possibility for proving lower bounds for external sorting via showing a lower bound for priority queues.

The aim of this paper is to show that the multidimensional Monte Carlo integration can be efficiently implemented on clusters with hybrid GPU-accelerated nodes using recently developed parallel versions of LCG and LFG pseudorandom number generators. We explain how to utilize multiple GPUs and all available cores of CPUs within a single node and how to extend computations on all available nodes of a cluster using MPI. The results of experiments performed on a Tesla-based GPU cluster are also presented and discussed.

We are interested in the resolution of the 3D Helmholtz equation for real applications. Solving this problem numerically is a computational challenge due to the large memory requirements of the matrices and vectors involved.For these cases, the massive parallelism of GPU architectures and the high performance at lower energy of the multicores can be exploited. To do a fair comparison between the benefit of accelerating the three-dimensional Helmholtz equation using GPU architectures and multicore platforms, this paper describes three different parallelization schemes on a multi-GPU cluster and also includes an evaluation of their performance. The three parallel schemes consist of:(1) using the multicore processors (CPU version), (2) using the GPU devices (GPU version); and (3) using a hybrid implementation which combines CPU cores and GPU devices simultaneously (hybrid version).Experimental results show that our hybrid implementation outperforms the other approaches in terms of performance.

The recent trend of using Graphics Processing Units (GPUs) for high performance computations is driven by the high ratio of price performance for these units, complemented by their cost effectiveness. Such kinds of units are increasingly being deployed not only as accelerators for supercomputer installations, but also in GPU-enabled nodes in Grid and Cloud installations. At first glance computational fluid dynamics (CFD) solvers match perfectly to GPU resources, because these solvers make intensive calculations and use relatively small memory. Nevertheless, there are scarce results about the practical use of this serious advantage of GPU over CPU, especially for calculations of viscous, compressible, heat conductive gas flows with double precision accuracy. In our work we present calculation of unsteady, viscous, compressible and heat conductive gas with double precision accuracy using GPU-enabled version of the algorithm SIMPLE-TS, written on standard OpenCL. As a test case we model the flow past a square in a microchannel at supersonic speed with Mach number \(\mathrm{{M}}=2.43\) on AMD Radeon HD 7950 GPU and achieve 90 GFlops, which is 46 times faster than the CPU serial code run on Intel Xeon X5560.

In the high performance computing market, CPU-GPU hybrid computing is gradually becoming the tendency. Entropic Lattice Boltzmann method(ELBM) parallelization, like many parallel algorithms in the field of rapid scientific and engineering computing, has given rise to much attention for applications of computational fluid dynamics. This paper presents an efficient implementation of ELBM for a D3Q19 lattice Boltzmann flow simulation in a hybrid CPU-GPU computing environment, which is consisted of AMD multi-core CPU with NVIDIA graphics processors unit (GPU). To overcome the GPU memory size limitation and communication overhead, we propose a set of techniques that can be used to develop efficient ELBM algorithms for the hybrid system. Considering that the respective contributions of CPU versus GPU and the lattice scale of application, the optimal load balancing model is built. These approaches result in an efficient implementation with balanced use of multi-core processor and a graphics processor based on software framework OpenMP and CUDA. Finally, we show that the comparison of performance results using both CPU and GPU with using either a CPU core or a GPU.

As most research focuses on improving the acquisition sensitivity and speed in AWGN (Additive White Gaussion Noise) channel, this paper will present the theoretical analysis of acquisition performance for GNSS signal over frequency non-selective fading channel. The complementary cumulative distribution function for the correlation peak value higher than a threshold R th is derived when the channel is slow and fast fading respectively. For the Rician fading, when the threshold R th is larger than a special value, the acquisition performance over slow fading channel will surpass the performance over fast fading channel. Conversely, the performance for fast fading will be better. But, both of them are inferior to the performance over AWGN channel. For the Rayleigh channel, theoretical analysis shows that the acquisition performance for GNSS signal over slow fading channel exceeds the performance over fast fading channel. Finally, the acquisition for GPS signal transmitted over frequency non-selective fading channel has been simulated and the simulation results confirm the validity of the derivations and analysis.

We consider the vector scheduling problem, a natural generalization of the classical makespan minimization problem to multiple resources. Here, we are given $n$ jobs represented as $d$-dimensional vectors in $[0,1]^d$ and $m$ identical machines, and the goal is to assign the jobs to machines such that the maximum {\em load} of each machine over all the coordinates is at most $1$. For fixed $d$, the problem admits an approximation scheme, and the best known running time is $n^{f(\epsilon,d)}$ where $f(\epsilon,d) = (1/\epsilon)^{\tilde{O}(d)}$ ($\tilde{O}$ supresses polylogarithmic terms in $d$). In particular, the dependence on $d$ is doubly exponential. In this paper we show that a double exponential dependence on $d$ is necessary, and give an improved algorithm with essentially optimum running time. Specifically, we show that: \begin{itemize} \item For any $\epsilon0$. \item We complement these lower bounds with a $(1+\epsilon)$-approximation that runs in time $\exp((1/\epsilon)^{O(d \log \log d)}) + nd$. This gives the first efficient approximation scheme (EPTAS) for the problem. \end{itemize}

Function of smart antennas enhancing the OFDM based WLAN’s immunity to multipath fading environments has been summarized. Time domain implementation of smart antenna algorithm(pre-FFT algorithm) has been explained and a simple modification associated with the weight update of pre-FFT algorithm has been proposed Through computer simulations, the performances of the both algorithms have been shown and it was observed that the proposed algorithm contributed to obtain acceptable performance in multipath fading scenario.

Emerging applications of wireless sensor networks (WSNs) require real-time quality of service (QoS) guarantees to be provided by the network. Due to the non-deterministic impacts of the wireless channel and queuing state, probabilistic analysis of QoS is essential. For most WSNs applications, the end-to-end delay for packet delivery and the energy consumption are the most important QoS metrics. In this chapter, a comprehensive cross-layer probabilistic analysis framework is presented to investigate the probabilistic evaluation of QoS performance provided by WSNs. In particular, the QoS performance is evaluated in two levels. In the node level, using a Discrete-Time Markov queueing model, the distribution of single-hop delay and single-node energy consumption and lifetime are analyzed. In the network level, based on the node level analysis, the distributions of end-to-end delay, the network lifetime, and the event detection delay are then analyzed. Fluid models are utilized in the network level analysis. The framework also considers a realistic channel environments. Compared to the first-order QoS statistics, such as the mean and the variance, the distribution of QoS metrics reveals the relationship between the performance and reliability with QoS-based operations in WSNs. Using the framework, effective network development can be performed.

In this work, we consider a discrete-time \(Geo/G/1\) retrial queue with preemptive resume, Bernoulli feedback and general retrial times. We analyze the Markov chain underlying the considered queueing system and derive the generating functions of the system state, the orbit size and the system size distribution. Using probability generating function technique, some interesting and important performance measures are obtained. We also investigate the stochastic decomposition property and present some numerical examples.

Discrete-time queues are commonly used in the performance analysis of many computer and communications systems. As the independent and identically distributed (i.i.d.) arrival traffic model frequently adopted (due to its analytical tractability) in the literature does not capture some important properties of real-world traffics, such as self-similarity, long-range dependence, and burstiness, it is more realistic to model the arrival traffic as a time-correlated process.

This paper deals with the research of finitesource retrial queues whose service rate depends on queue length. Two- and three-dimensional models that describe threshold and hysteresis control policies are taken into account. Explicit vector-matrix representations of stationary distributions are main results in both cases.

We consider the problem of approximating optimal hub labelings in the context of labeling algorithms for the shortest path problem. A previous result was a O(logn) approximating for minimizing the total label size. We give an O(logn)-approximation algorithm for the maximum label size. We also give O(logn)-approximation algorithms for natural generalizations of the problem: Minimizing an lp norm of the labeling and simultaneously minimizing lp and lq norms.

In this paper, for permutation flowshops with two machines or more than two machines, as the number of jobs tends to infinity, the properties of asymptotic distribution of makespan are proposed. We introduce several conclusions in queuing theory to the scheduling problem, and convert the distribution of makespan to the distribution of waiting time in the queue. In two-machine conditions, the asymptotic distribution of makespan is proved to be the right half of a normal distribution. In m-machine conditions (m > 2), the asymptotic distribution of waiting time is proved under certain assumptions, and the bounds of probability distribution functions of waiting times are given.

In order to reduce network energy consumption and prolong the lifetime of wireless sensor networks, this paper improves the LEACH to an energy efficient multi-hop routing algorithm. LEACH in the cluster creation, data transmission, the update phase of the cluster was modified in proposed algorithm. The algorithm updates the cluster head reasonably and adjusts the structure of the cluster to reduce the energy consumption in cluster establishment phrase. In data transmission, it lowers energy consumption by inter-cluster and intra-cluster multi-hop transmission. The simulation runs the algorithm on NS2. The results show that the new algorithm’s effectiveness in reducing energy consumption by comparing it with LEACH, LEACH-C,DEEUC.

We analyze two parallel finite element implementations of the 2D time-dependent advection diffusion problem, one for multi-core clusters and one for CUDA-enabled GPUs, and compare their performances in terms of time and energy consumption. The parallel CUDA-enabled GPU implementation was derived from the multi-core cluster version. Our experimental results show that a desktop machine with a single CUDA-enabled GPU can achieve performance higher than a 24-machine (96 cores) cluster in this class of finite element problems. Also, the CUDA-enabled GPU implementation consumes less than one twentieth of the energy (Joules) consumed by the multi-core cluster implementation while solving a whole instance of the finite element problem.

In this paper we study a multi-server retrial queueing model in which customers arrive according to a Markovian arrival process (MAP) and the service times are assumed to be of phase type (PH-type). An arriving customer finding all servers busy will enter into a (retrial) orbit of infinite size. The customers in orbit will try to capture a free server after a random amount of time which is assumed to be of PH-type. Thus, every customer in the orbit has his/her own phase type distribution before attempting to get into service. Due to the complexity of the model and lack of attention to such models in the literature, we study this via simulation. After validating our simulated results against known results (both exact and approximation) for some special cases, we illustrate how one can underestimate or overestimate some key system performance measures by incorrectly assuming the retrial times to be exponential.

In this work, we apply a common economic tool, namely money, to coordinate network packets. In particular, we present a network economy, called PacketEconomy, where each flow is modeled as a population of rational network packets, and these packets can self-regulate their access to network resources by mutually trading their positions in router queues. We consider a corresponding Markov model of trade and show that there are Nash equilibria (NE) where queue positions and money are exchanged directly between the network packets. This simple approach, interestingly, delivers significant improvements for packets and routers.

The approach of accelerating application with GPUs already delivers impressive computational performance compared to the traditional CPU. The hardware architecture of GPU is a significant departure from CPUs, hence the redesign and validation of the numerical algorithm are necessary. The spectral-finite-difference schemes usually used in the direction numerical simulation (DNS) for turbulent channel flows are studied here. In order to validate the numerical accuracy, the scalar diffusion equation is first solved with this scheme, and the results from GPU and CPU are validated with the analytical solution. The major computational kernels of the scheme are the fast Fourier transfer (FFT) and the linear equation solver, which are both implemented on GPU. The performance study of the scalar diffusion equation shows at least 20\(\times \) speedup. For 3D Navier-Stokes equation, the performance on a single Nvidia S2050 card shows 25 times speedup.

This chapter addresses \({M}^{X}/G/1/n\) queues, i.e., single server batch Markovian arrival queues with finite customer waiting space of size n. Taking profit of the Markov regenerative structure of these systems, we develop an efficient recursive procedure to compute the probability mass function of the number of losses in busy-periods starting with an arbitrary number of customers in the system. The derived computational procedure is easy to implement and leads to a fast numerical computation of the loss probabilities. To illustrate the effectiveness of the procedure, loss probabilities are computed for a wide variety of queues, with different capacities, batch size distributions, and arrival and service parameters.

We consider a multi-server queueing system with session arrivals that operates in random environment. A session consists of a random number of customers. Sessions arrive to the system in Markovian arrival flow. Inter-arrival times of customers within each session are random. The service time of a customer has a phase type distribution. A numerically stable algorithm for calculation of the stationary distribution of system states is presented. The main performance measures are calculated.

The paper presents a new model of a multiservice queueing system with limited queue and state-dependent dynamic resource sharing of the server between individual classes of calls. The advantage of the proposed model is the possibility to evaluate the average parameters of queues for individual classes of calls. The proposed model can be taken as a basis for further research on multiservice queueing systems.

We discuss the application of an efficient numerical algorithm to sensitivity analysis of the GI/M/1 queue. Specifically, we use a numerical approach based on the Taylor series expansion to examine the robustness of the GI/M/1 queue to some specific perturbations in the arrival process: linear and non-linear perturbations. For each kind of perturbation we approximately compute the sensitivity of the main characteristics of the GI/M/1 queue corresponding to the case where the arrival processes are lightly different from that of the nominal queue. Numerical examples are presented to illustrate the accuracy of the proposed approach.

This paper presents a new hybrid solver based on Schur complement method, in which computations are distributed between multiple CPUs and GPUs. In this solver, the Schur complement is computed either on CPUs for small problem size or on GPUs for large problem sizes. To solve the interface system, we propose a new multi-GPU algorithm that implements conjugate gradient method with explicit preconditioning. Experiments with wrap spring simulation on hybrid multi-core multi-GPU cluster demonstrate efficiency of the proposed method.

Several highly optimized implementations of Finite Difference schemes are discussed. The combination of vectorization and an interleaved data layout, spatial and temporal loop tiling algorithms, loop unrolling, and parameter tuning lead to efficient computational kernels in one to three spatial dimensions, truncation errors of order two to twelve, and isotropic and compact anisotropic stencils. The kernels are implemented on and tuned for several processor architectures like recent Intel Sandy Bridge, Ivy Bridge and AMD Bulldozer CPU cores, all with AVX vector instructions as well as Nvidia Kepler and Fermi and AMD Southern and Northern Islands GPU architectures, as well as some older architectures for comparison. The kernels are either based on a cache aware spatial loop or on time-slicing to compute several time steps at once. Furthermore, vector components can either be independent, grouped in short vectors of SSE, AVX or GPU warp size or in larger virtual vectors with explicit synchronization. The optimal choice of the algorithm and its parameters depend both on the Finite Difference stencil and on the processor architecture.

This paper studies an infinite-server queue in a Markov environment, that is, an infinite-server queue with arrival rates and service times depending on the state of a Markovian background process. Scaling the arrival rates $\lambda_i$ by a factor $N$ and the rate $q_{ij}$ of the background process by a factor $N^\alpha$, with $\alpha \in \mathbb R^+$, we establish a central limit theorem as $N$ tends to $\infty$. We find different scaling regimes, based on the value of $\alpha$. Remarkably, for $\alpha < 1$, we find a central limit theorem with a non-square-root scaling but rather with $N^{\alpha/2}$; in the expression for the variance deviation matrices appear.

An analogue of M t /M t /S/S Erlang loss system for a queue with group services is introduced and considered. Weak ergodicity of the model is studied. We obtain the bounds on the rate of convergence to the limiting characteristics and consider two concrete queueing models with finding of their main limiting characteristics.

In this paper, a novel optical packet switch is proposed, which uses limited wavelength converters and recirculation fiber delay lines to resolve optical packet contentions. A control algorithm is proposed to schedule the optical packets. Performance of the proposed switch is evaluated by means of simulation experiments. Packet loss rate and mean delay time of the proposed switch are analyzed in detailed.

This paper proposes an acceleration strategy for SPH on single-node multi-GPU platform. First the acceleration strategy for SPH on single-GPU is studied in conjunction with the characteristics of architecture. Then the changing pattern of SPH's computation time has been discussed. Based on the fact that the changing pattern is rather slow, using a simple dynamic load balancing algorithm an acceptable load balance is obtained on multi-GPU. Finally, an almost linear speedup is achieved on multi-GPU by further optimizing dynamic load balancing algorithm and communication strategy among multiple GPUs

In this paper we analyze a discrete-time queueing system in which an arriving customer can decide, with a certain probability, to go directly to the server expelling out of the system the customer that is currently in service or to join the queue in the last place. The arrivals are assumed to be geometrical and the service times are arbitrarily distributed. We present some numerical examples in order to illustrate the effect of the parameters on several performance characteristics.

In this paper, a queuing model based on the CAN is established and its simulation and application are considered. The specific steps for, through the hypothesis model related parameters, make the whole model CAN achieve stable state. Through the calculation model of the probability of no data transmission, data in the model in the probability of transmission, data sending average queue goal number, average each data queue time, average each data in the model of the time and other related parameters modeling. By solving the queuing model, the data in the CAN bus to send on quantitative analysis, which CAN be used to CAN bus to transfer data distribution scheme optimization.

We consider a queueing system with Poisson arrivals and exponentially distributed service time and FCFS service discipline. The service of a customer is started at the moment of arrival (in case of free system) or at moments differing from it by the multiples of a given cycle time T (in case of occupied server or waiting queue). The waiting time is always the multiple of cycle time T, one finds its generating function and mean value. The characteristics of service are illustrated by numerical examples. If we measure the waiting time by means of number of cycles, we can optimize the cycle time T.

This paper describes an efficient implementation of Bayesian hierarchical model for space time variability of latent heat flux. Model parameters are estimated using Gibbs sampler. Many computations in the Gibbs sampling scheme are time intensive. Some time intensive computations in the model are matrix operations such as matrix multiplication, matrix inverse, matrix LU decomposition. We have used graphical processing unit (GPU) to run time intensive matrix operations. Some time intensive operations are transformed to matrix operations to take advantage of GPU. To run the model in GPU model is implemented in Compute Unified Device Architecture (CUDA). Two GPUs are used to run the model one is NVIDIA Graphics card Tesla C2075 with 448 cores and other is NVIDIA Graphics card GT 520 with 48 cores. To study the gain in speedup on the GPU we have implemented the model using single threaded C to run on CPU. A comparative study of implementation on CPU and the two GPUs is carried out for 100 iterations of the Gibbs sampler. We found 30-130 fold speedup on Tesla GPU as compared to single threaded CPU code.

We present a high-performance C++ library for high but fixed precision (128 to 512 bit) integer arithmetic and symbolic polynomial computations. While the large integer and polynomial computation parts of the library can be used independently optimized kernels for symbolic polynomials with large integer coefficients are provided. The kernels were manually optimized in assembly language for the x86-64 and power64 architectures. Our main target application is high-temperature series expansions which requires inner products of large vectors of polynomials with large integer coefficients. For this purpose we implemented a tunable hybrid CPU/GPU inner product function using OpenMP and NVIDIA CUDA with inline PTX assembly. This way we make optimal use of today’s and upcoming hybrid supercomputers and attain 49% of the peak performance of the current NVIDIA Kepler GPU. Compared to a pure CPU solution using the GNU Multiple Precision Arithmetic Library (GMP) we gain a speedup of 13x for a pure CPU inner product and 38x using a GPU accelerator.

A multi-server queueing system with a Markovian Arrival Process (MAP), a finite buffer and impatient customers operating in random environment as a model of a call center is investigated. The service time of a customer by a server has an exponential distribution. If all servers are busy at a customer arrival epoch, the customer may leave the system forever or move to the buffer with probability that depends on the number of customers in the buffer. During a waiting period, a customer can be impatient and can leave the system without the service. System parameters depend on the state of the random environment. An efficient algorithm for calculating the stationary probabilities of system states is proposed. Some key performance measures are calculated. The Laplace-Stieltjes transforms of the sojourn and waiting time distributions are derived.

A single-server queueing system of M/G/1-type with boun- ded total volume is considered. It is assumed that volumes of arriving packets are generally distributed random variables. The AQM-type mechanism is used to control the actual buffer state: each of arriving packets is dropped with probability depending on its volume and the occupied volume of the system at the pre-arrival epoch. The explicit formulae for the stationary queue-size distribution and loss probability are found.

We study with the use of fluid flow approximation the impact of a modified weighted moving average on the performance of RED mechanism. A model of TCP/UDP connection with RED implemented in an intermediate IP router is used, the weighted moving average is determined on the basis of a difference (recursive) equation. The fluid flow approximation technique is applied to model the interactions between the set of TCP/UDP flows and RED mechanism.

We consider a fully observable single-server Markovian queue with variable service rate, where the customers observe the queue length and the state of the server upon arrival. We assume that the arriving customers decide whether to join the system or balk based on a natural reward-cost structure. With considering waiting cost and reward, we study the balking behavior of the customers and derive the corresponding Nash equilibrium strategies for all customers.

We are focusing on an iterative solver for the three-dimensional Helmholtz equation on multi-GPU using CUDA (Compute Unified Device Architecture). The Helmholtz equation discretized by a second order finite difference scheme is solved with Bi-CGSTAB preconditioned by a shifted Laplace multigrid method. Two multi-GPU approaches are considered: data parallelism and split of the algorithm. Their implementations on multi-GPU architecture are compared to a multi-threaded CPU and single GPU implementation. The results show that the data parallel implementation is suffering from communication between GPUs and CPU, but is still a number of times faster compared to many-cores. The split of the algorithm across GPUs limits communication and delivers speedups comparable to a single GPU implementation.

In term of spares demanded and shared with different lifetime distributions, using the simulation method, simulation model of spares demanded and shared are build. Simulation model of spares demanded solves the problem of demand quantity of spare parts with non-exponent distribution of lifetime distribution, comparing with the tradition mathematics analytical model more simple; simulation model of spares shared solves the problem of demand quantity of spares shared with different lifetime distributions. At the same time the model solves the problem of demand quantity of spares of serial system with different types of parts. The work expands the range of calculating demand quantity of spares. Finally, the results of computer simulation and analytical method validate its effectiveness and correctness.

We are witnessing the consolidation of the GPUs streaming paradigm in parallel computing. This paper explores stencil operations in CUDA to optimize on GPUs the Jacobi method for solving Laplace's differential equation. The code keeps constant the access pattern through a large number of loop iterations, that way being representative of a wide set of iterative linear algebra algorithms. Optimizations are focused on data parallelism, threads deployment and the GPU memory hierarchy, whose management is explicit by the CUDA programmer. Experimental results are shown on Nvidia Teslas C870 and C1060 GPUs and compared to a counterpart version optimized on a quadcore Intel CPU. The speed-up factor for our set of GPU optimizations reaches 3-4x and the execution times defeat those of the CPU by a wide margin, also showing great scalability when moving towards a more sophisticated GPU architecture and/or more demanding problem sizes.

The efficient transmission of TCP traffic over OBS networks is a challenging problem, due to the high sensitivity of TCP congestion control mechanism to losses. In this paper, a traffic profiling scheme is proposed for gathering TCP flow statistics, which are used to estimate steady-state performance of TCP traffic over OBS networks. The burst assembly unit, in parallel to the assembly process, can inspect TCP traffic, keeping traffic statistics that can be used for throughput estimations. In this paper, we detail the traffic profiling and estimation mechanism and also provide analytical and simulation results to assess its performance.

In INFOCOM 2009, Yu, Wei, Ramkumar and Guan have proposed the novel mechanism (called Yu's scheme), in which a forwarder can filter polluted messages before spreading the pollution in the XOR network coding systems. In order to perform such filtering, two or more message authentication codes (MACs) are used for this scheme. However, Yu's scheme has a problem that the number of MACs increases at every coding point, since it cannot operate MACs with the XOR network coding. This means that the MAC of Yu's scheme does not have homomorphic property. In this paper, we propose the first symmetric-key-based scheme not only to filter polluted messages but also to operate MACs with the XOR network coding on a forwarder. The XOR network coding of MACs produces improvement which does not increase the number of MACs at a coding point. Our scheme uses the UHFs-based MAC with a homomorphic property to hold homomorphic MAC, and hence it can aggregate MACs in our XOR network coding systems. We emphasize that a forwarder cannot straightforward filter polluted messages even if our scheme uses the UHFs-based MACs.

The Quiet DDoS attack was proposed in 2009 by Amey and Nirwan. It launches certain number of TCP flows to reduce the transferring ability of normal TCP flows by using botnet. By the simulation experiments on the NS2 platform, we analyze the factors and their impacts on the attack in detail. According to the results of simulation, we give some suggestion to counter the Quiet DDoS attack.