Lucio Grandinetti

Parallel algorithms to solve two-stage stochastic linear programs with robustness constraints

Abstract In this paper we present a parallel method for solving two-stage stochastic linear programs with restricted recourse. The mathematical model considered here can be used to represent several real-world applications, including financial and production planning problems, for which significant changes in the recourse solutions should be avoided because of their difficulty to be implemented. Our parallel method is based on a primal-dual path-following interior point algorithm, and exploits fruitfully the dual block-angular structure of the constraint matrix and the special block structure of the matrices involved in the restricted recourse model. We describe and discuss both message-passing and shared-memory implementations and we present the numerical results collected on the Origin2000.

Auction algorithms for decentralized parallel machine scheduling

Computational grids are highly complex distributed systems (involving multiple organizations with different goals and policies) which aim at providing computing services without the users need to know the location and features of the required resources. A key issue in managing and scheduling grid resources is the coordination among multiple administrative domains. In this paper, we present a preliminary study which aims at developing auction mechanisms for decentralized scheduling which exhibit minimal communication overhead and an efficient usage of resources.

Software for parallel computation

1 Introduction.- Software for Parallel Computing: Key Issues and Research Directions.- 2 Tools and Methods for Parallel Computing.- Learning From Our Successes.- Software Development for Parallel Processing.- Software Tools for Developing and Porting Parallel Programs.- Scalable Software Tools for Parallel Computations.- PVM and HeNCE: Tools for Heterogeneous Network Computing.- Distributed Shared Memory: Principles and Implementation.- FORGE 90 and High Performance Fortran (HPF).- The Bird-Meertens Formalism as a Parallel Model.- Software Issues for the PASM Parallel Processing System.- Data Migrations on the Hypercube.- Automatic Determination of Parallelism in Programs.- Are Object Oriented Programming Methods Suitable for Numerical Computing on Parallel Machines.- Parallel Relational Data Base Management System Design Aspects.- 3 Graphics.- MUDI3: a Tool for the Interactive Visual Analysis of Multidimensional Fields.- Graphical Support for Parallel Debugging.- 4 Algorithms and Applications.- Backpropagation on Distributed Memory Systems.- Mapping Algorithms on Hypercube.- Software Development for Finite Element Fluid Dynamic Applications.- Asynchronous Communication on Shared-Memory Parallel Computers.- Two Different Data-Parallel Implementations of the BLAS.- 5 Performance of Parallel Programs and Systems.- Examples of Scalable Performance.- Performance Prediction from Data Transport.- Dynamic Load Balancing in Adaptive Parallel Applications.- Performance Analysis of Dynamic Scheduling Techniques for Irregularly Structured Computation.

An optimization-based heuristic for the Multi-objective Undirected Capacitated Arc Routing Problem

The Multi-objective Undirected Capacitated Arc Routing Problem (MUCARP) is the optimization problem aimed at finding the best strategy for servicing a subset of clients localized along the links of a logistic network, by using a fleet of vehicles and optimizing more than one objective. In general, the first goal consists in minimizing the total transportation cost, and in this case the problem brings back to the well-known Undirected Capacitated Arc Routing Problem (UCARP). The motivation behind the study of the MUCARP lies in the study of real situations where companies working in the logistic distribution field deal with complex operational strategies, in which different actors (trucks, drivers, customers) have to be allocated within an unified framework by taking into account opposite needs and different employment contracts. All the previous considerations lead to the MUCARP as a benchmark optimization problem for modeling practical situations. In this paper, the MUCARP is heuristically tackled. In particular, three competitive objectives are minimized at the same time: the total transportation cost, the longest route cost (makespan) and the number of vehicles (i.e., the total number of routes). An approximation of the optimal Pareto front is determined through an optimization-based heuristic procedure, whose performances are tested and analyzed on classical benchmark instances.

Applications of neural-based spot market prediction for cloud computing

Advances in service-oriented architectures (SOA), virtualization, high-speed networks, and cloud computing has resulted in attractive pay-as-you-go services. Job scheduling on these systems results in commodity bidding for computing time. This bidding is institutionalized by Amazon for its Elastic Cloud Computing (EC2) environment and bidding methods exist for other cloud-computing vendors as well as multi-cloud and cluster computing brokers such as SpotCloud. Commodity bidding for computing has resulted in complex spot price models that have ad-hoc strategies to provide demand for excess capacity. In this paper we will discuss vendors who provide spot pricing and bidding and present a predictive model for future spot prices based on neural networking giving users a high confidence on future prices aiding bidding on commodity computing.

An approximate ϵ-constraint method for a multi-objective job scheduling in the cloud

Cloud computing is a hybrid model that provides both hardware and software resources through computer networks. Data services (hardware) together with their functionalities (software) are hosted on web servers rather than on single computers connected by networks. Through a device (e.g., either a computer or a smartphone), a browser and an Internet connection, each user accesses a cloud platform and asks for specific services. For example, a user can ask for executing some applications (jobs) on the machines (hosts) of a cloud infrastructure. Therefore, it becomes significant to provide optimized job scheduling approaches suitable to balance the workload distribution among hosts of the platform. In this paper, a multi-objective mathematical formulation of the job scheduling problem in a homogeneous cloud computing platform is proposed in order to optimize the total average waiting time of the jobs, the average waiting time of the jobs in the longest working schedule (such as the makespan) and the required number of hosts. The proposed approach is based on an approximate @e-constraint method, tested on a set of instances and compared with the weighted sum (WS) method. The computational results highlight that our approach outperforms the WS method in terms of a number of non-dominated solutions.

Some investigations in a new algorithm for nonlinear optimization based on conic models of the objective function

An appealing approach to the solution of nonlinear optimization problems based on conic models of the objective function has been recently introduced by Davidon. It leads to a broad class of algorithms which, in some sense, can be considered to generalize the existing quasi-Newton algorithms. One particular member of this class has been deeply examined by Sorensen, who has proved some interesting theoretical properties. A new interpretation of this algorithm is suggested in this paper from a more straightforward and somewhat familiar point of view. In addition, numerical experiments have been carried out to compare the Sorensen algorithm with a straightforward BFGS implementation of the classical quasi-Newton method with the final aim to assess the real merits and benefits of the new algorithm. Although some challenging test functions are used in computational experiments, the results are not particularly favorable to the new algorithm. As a matter of fact they do not exhibit any jump of quality, as it might be expected. Lastly, it is pointed out that a difficulty may affect the new method in situations in which it is necessary to exploit the special structure of large-scale problems.

A multi-dimensional job scheduling

With the advent of new computing technologies, such as cloud computing and contemporary parallel processing systems, the building blocks of computing systems have become multi-dimensional. Traditional scheduling systems based on a single-resource optimization, like processors, fail to provide near optimal solutions. The efficient use of new computing systems depends on the efficient use of several resource dimensions. Thus, the scheduling systems have to fully use all resources. In this paper, we address the problem of multi-resource scheduling via multi-capacity bin-packing. We propose the application of multi-capacity-aware resource scheduling at host selection layer and queuing mechanism layer of a scheduling system. The experimental results demonstrate performance improvements of scheduling in terms of waittime and slowdown metrics. A proposal for scheduling problem based on multi-capacity bin-packing algorithms.A proposal for host selection and queuing based on multi-resource scheduling.Getting better waittime and slowdown metrics than the state of the art scheduling.

Efficiency Analysis of Parallel Batch Pattern NN Training Algorithm on General-Purpose Supercomputer

The theoretic and algorithmic description of the parallel batch pattern back propagation (BP) training algorithm of multilayer perceptron is presented in this paper. The efficiency research of the developed parallel algorithm is fulfilled at progressive increasing of the dimension of parallelized problem on general-purpose parallel computer NEC TX-7.

Improvement of parallelization efficiency of batch pattern BP training algorithm using Open MPI

The use of tuned collective’s module of Open MPI to improve a parallelization efficiency of parallel batch pattern back propagation training algorithm of a multilayer perceptron is considered in this paper. The multilayer perceptron model and the usual sequential batch pattern training algorithm are theoretically described. An algorithmic description of a parallel version of the batch pattern training method is introduced. The obtained parallelization efficiency results using Open MPI tuned collective’s module and MPICH2 are compared. Our results show that (i) Open MPI tuned collective’s module outperforms MPICH2 implementation both on SMP computer and computational cluster and (ii) different internal algorithms of MPI_Allreduce() collective operation give better results on different scenarios and different parallel systems. Therefore the properties of the communication network and user application should be taken into account when a specific collective algorithm is used.