Luis Díaz de Cerio

Communication pipelining in hypercubes

Many parallel algorithms exhibit a hypercube communication topology. Such algorithms can easily be executed on a multicomputer with a hypercube interconnection topology. However, in most cases these parallel algorithms only make use of a small fraction of the interconnection bandwidth offered by the multicomputer. In particular, each processor of a hypercube multicomputer is connected to d different neighbors by d different links. Nevertheless, hypercube algorithms usually do not use more than one of these d links at the same time. This paper presents a technique called communication pipelining that enables a more efficient use of the interconnection network and, in consequence, a significant reduction in the execution time. This technique is based on a transformation of the original algorithm. The resulting equivalent algorithm makes use of several links of each node simultaneously. Given a particular problem and a particular architecture. the degree of pipelining to be applied is a design parameter that must be decided when transforming the original algorithm. The paper presents analytical models that allow for an optimal choice of the degree of pipelining for each problem and a given architecture. To illustrate the performance of the communication pipelining technique, its application to the FFT computation is presented as an example. It is shown that an optimal choice of the degree of pipelining can achieve a reduction by a factor of d in the communication overhead of the algorithm.

A method for exploiting communication/computation overlap in hypercubes

Abstract This paper presents a method to derive efficient algorithms for hypercubes. The method exploits two features of the underlying hardware: a) the parallelism provided by the multiple communication links of each node and b) the possibility of overlapping computations and communications which is a feature of machines supporting an asynchronous communication protocol. The method can be applied to a generic class of hypercube algorithms whose distinguishing features are quite frequent in common algorithms for hypercubes. Many examples of this class of algorithms are found in the literature for different problems. The paper shows the efficiency of the method for two case studies. The results show that the reduction in communication overhead is very significant in many cases. They also show that the algorithms produced by our method are always very close to the optimum in terms of execution time.

HGRID: An Adaptive Grid Resource Discovery

Grid resource discovery service is a fundamental problem that has been the focus of research in the recent past. We propose a scheme that has essential characteristics for efficient, self-configuring and fault-tolerant resource discovery and is able to handle dynamic attributes, such as memory capacity. Our approach consists of an overlay network with a hypercube topology connecting the grid nodes and a scalable, fault-tolerant, self-configuring and adaptive search algorithm. Every grid node keeps a small routing table of only log2N entries. The search algorithm is executed in less than (log2N +1) time steps and each grid node is queried only once. By design, the algorithm improves the probability of reaching all working nodes in the system even in the presence of non-alive nodes (inaccessible, crashed or heavily loaded nodes). We analyze the static resilience of the approach presented, which is the measure of how well the algorithm can discover resources without having to update the routing tables. This is done before the routing recovery is processed in order to reconfigure the overlay to avoid non-alive nodes. The results show that our approach has a significantly high static resilience for a grid environment.

Divide-and-Conquer Algorithms on Two-Dimensional Meshes

The Reflecting and Growing mappings have been proposed to map parallel divide-and-conquer algorithms onto two — dimensional meshes. The performance of these mappings has been previously analyzed under the assumption that the parallel algorithm is initiated always at the same fixed node of the mesh. In such scenario, the Reflecting mapping is optimal for meshes with wormhole routing and the Growing mapping is very close to the optimal for meshes with store-and-forward routing. In this paper we consider a more general scenario in which the parallel divide-and-conquer algorithm can be started at an arbitrary node of the mesh. We propose and approach that is simpler than both the Reflecting and Growing mappings, is optimal for wormhole meshes and better than the Growing mapping for store-and-forward meshes.

Design choices and their implications for 3d mid-air manipulation techniques

Manipulation is one of the most important tasks required in virtual environments and thus it has been thoroughly studied for widespread input devices such as mice or multi-touch screens. Nowadays, the Kinect sensor has turned mid-air interaction into another affordable and popular way of interacting. Mid-air interaction enables the possibility of interacting remotely without any physical contact and in a more natural manner. Nonetheless, although some scattered manipulation techniques have been proposed for mid-air interaction, there is a lack of evaluations and comparisons that hinders the selection and development of these techniques. To solve this issue, we gathered four design choices that can be used to classify mid-air manipulation techniques. Namely, choices are based on the required number of hands, separation of translation–rotation, decomposition of rotation, and interaction metaphors. Furthermore, we developed, adapted, and compared three manipulation techniques selected for studying the implications of the design choices. These implications are useful to select among already existing techniques as well as to inform technique developers.

Overlapping Communication and Computation in Hypercubes

This paper presents a method to derive efficient algorithms for hypercubes. The method exploits two features of the underlying hardware: a) the parallelism provided by the multiple communication links of each node and b) the possibility of overlapping computations and communications, which is a feature of machines supporting an asynchronous communication protocol. The method can be applied to a generic class of hypercube algorithms. Many examples of this class of algorithms are found in the literature for different problems. The paper shows the efficiency of the method using two of these problems as an example: FFT and Vector Add. The results show that the reduction in communication overhead is very significant in many cases and the algorithms produced by our method are always very close to the optimum in terms of execution time.

HGRID: Fault Tolerant, Log2N Resource Management for Grids

Grid Resource Discovery Service is currently a very important focus of research. We propose a scheme that presents essential characteristics for efficient, self-configuring and fault-tolerant resource discovery and is able to handle dynamic attributes, such as memory capacity. Our approach consists of an overlay network with a hypercube topology connecting the grid nodes and a scalable, fault-tolerant, self-configuring search algorithm. By design, the algorithm improves the probability of reaching all working nodes in the system even in the presence of failures (inaccessible, crashed or heavy loaded nodes). We analyze the static resilience of the presented approach, that is to say, how well the algorithm is able to find resources without having to update the routing tables. The results show that the presented approach has a high static resilience.

CALMANT: Un Método Sistemático para la Ejecución de Algoritmos Hipercubo en Sistemas Multiprocesador

EN ESTE TRABAJO PRESENTAMOS EL METODO CALMANT (CC CUBE ALGORITHMUS ON MESHE AND TORT) COMO UN METODO SISTEMATICO PARA L A EJECUCION DE UN CIERTO TIPO DE ALGORITMOS, QUE DENOMINAREMOS ALGORITMOS CC-CUBO, SOBRE MALLAS Y TOROS DE VARIAS DIMENSIONES. ES MUY FRECUENTE ENCONTRAR ALGORITMOS CC-CUBO EN LA LITERATURA (FFT, COMPLETE EXCHANGE, CALCULO DE VALORES Y VECTORES APROPIADOS, ETC.9 PERO LA APLICACION DIRECTA DE ESTOS ALGORITMOS NO PERMITE EXPLOTAR EFICIENTEMENTE EL ANCHO DE BANDO QUE NOS OFRECEN LAS REDES DE INTERCONEXION EN MALLA Y TORO. EL METODO CALMANT PERMITE REORGANIZAR LOS CALCULOS Y LAS COMUNICACIONES DE LOS ALGORITMOS CC-CUBE DE MANERA QU LA EFICIENCIA AUMENTE NOTABLEMENTE. LA IMPORTANCIA NO SOLO RADICA EN EL AUMENTO DE LA EFICIENCIAS SINO QUE ADEMAS PUEDE SER APLICADO DE FORMA SISTEMATICA SOBRE DIFERENTES TIPOS DE ARQUITECTURA.

Complete Exchange Algorithms for Meshes and Tori Using a Systematic Approach

Frequently, algorithms for a given multicomputer architecture cannot be used (or are not efficient) for a different architecture. The proposed method allows the systematic design of complete exchange algorithms for meshes and tori and it can be extended to some other architectures that may be interesting in the future.