This paper presents and evaluates a universal algorithm to improve the performance of MPI collective communication operations on hierarchical clusters with many-core nodes. This algorithm exploits shared-memory buffer...
详细信息
ISBN:
(纸本)9780769552071
This paper presents and evaluates a universal algorithm to improve the performance of MPI collective communication operations on hierarchical clusters with many-core nodes. This algorithm exploits shared-memory buffers for efficient intra-node communication while still allowing the use of unmodified, hierarchy-unaware traditional collectives for inter-node communication (including collectives like Alltoallv). This algorithm improves on past works that convert a specific collective algorithm into a hierarchical version and are generally restricted to fan-in, fan-out, and Allgather algorithms. Experimental results show impressive performance improvements utilizing a variety of collectives from MPICH as well as the closed-source Cray MPT for the inter-node communication. The experimental evaluation tests the new algorithms with as many as 65536 cores and sees speedups over the baseline averaging 14.2x for Alltoallv, 26x for Allgather, and 32.7x for Reduce-Scatter. The paper further improves inter-node communication by utilizing multiple senders from the same shared memory buffer, achieving additional speedups averaging 2.5x. The discussion also evaluates special-purpose extensions to improve intra-node communication by returning shared memory or copy-on-write protected buffers from the collective.
generalized algorithm for determining the I-pv-V-pv and P-pv-V-pv curves at given insolation and temperature of solar photovoltaic (SPV) array, under partial shadow conditions is proposed in this paper. The algorithm ...
详细信息
ISBN:
(纸本)9781665459303
generalized algorithm for determining the I-pv-V-pv and P-pv-V-pv curves at given insolation and temperature of solar photovoltaic (SPV) array, under partial shadow conditions is proposed in this paper. The algorithm is proficient of developing the characteristics of SPV array of any configuration with any kind of shadow pattern. Individual module specifications are also inputted. A MATLAB code is developed for the algorithm and I-pv-V-pv and P-pv-V-pv curves of a sample system of [4*3] PV array are identified for ten shadow patterns. For each shadow pattern values of Vmp and Pmp (voltage and peak power) are also identified using the algorithm, which are further required for peak power point tracking (PPPT).
We present a general algorithm for learning the structure of discrete Markov random fields from i.i.d. samples. The algorithm either achieves the same computational complexity or lowers the computational complexity of...
详细信息
ISBN:
(纸本)9781467310161
We present a general algorithm for learning the structure of discrete Markov random fields from i.i.d. samples. The algorithm either achieves the same computational complexity or lowers the computational complexity of earlier algorithms for several cases, and provides a new low-computational complexity algorithm for the case of Ising models where the underlying graph is the Erdos-Renyi random graph G similar to G (p, c/p).
We present a new algorithm for decoding AG-codes significantly beyond the error-correction bound. Specifically, given a word y whose distance to the AG-code is at most e, where e is a parameter depending on the block ...
详细信息
ISBN:
(纸本)0780344081
We present a new algorithm for decoding AG-codes significantly beyond the error-correction bound. Specifically, given a word y whose distance to the AG-code is at most e, where e is a parameter depending on the block length and the dimension of the code, our algorithm produces all codewords that have distance less than or equal to e from y. We also discuss modifications of our general algorithm and show how to obtain similar algorithms for binary codes using concatenated codes.
In recent years, researchers come up with lots of path following algorithms. One of their basic assumptions is that the robot's initial position is a single certain point. Actually, as robots rely on sensor data t...
详细信息
ISBN:
(纸本)9781479974894
In recent years, researchers come up with lots of path following algorithms. One of their basic assumptions is that the robot's initial position is a single certain point. Actually, as robots rely on sensor data to locate by Kalman filter or Partical filter, it outputs a probability distribution. In this paper, F* describes a new path following algorithm with the robot's initial position is uncertain. It depends on Monte Carlo method and A* algorithm. Some important concepts are introduced and pseudo code is given to show how it works. Experiment results show that it is more effective than general algorithm.
In this paper, we propose a class of generalized flocking algorithms based on the Olfati-Saber's results for self-driving multi-agent networks by introducing some metric weighting matrix and leader-agent-related n...
详细信息
ISBN:
(纸本)9781509035502
In this paper, we propose a class of generalized flocking algorithms based on the Olfati-Saber's results for self-driving multi-agent networks by introducing some metric weighting matrix and leader-agent-related navigation-weighting coefficients, in which the multi-agents are also driven by distributed continuous-time control algorithms that are defined merely by local position and velocity measurements of neighboring multi-agents as usual. It is shown that the concerned multi-agent systems eventually run into what we call the generalized α-lattice flocking by means of the generalized algorithms, while flocking lattice geometries such as direction variation and lattice deformation (say, shrinkage and enlargement) can be achieved by choosing the weighting matrix and the leader agent coefficients. In the process of flocking, the whole multi-agent system satisfies the Reynold's principles of flocking aggregation, collision avoidance and velocity matching, while possible inter-agent collision is also verified by means of the variation-of-constants formula when initially the agents are aggregative too densely.
In this study the scheduling supply chain models combination in terms of optimal program control is considered. The SC schedule execution problem is formulated. It is consist of the problem of optimal program control ...
详细信息
ISBN:
(纸本)9781467376983
In this study the scheduling supply chain models combination in terms of optimal program control is considered. The SC schedule execution problem is formulated. It is consist of the problem of optimal program control and the problem of positional control. Then for solving considered problem we propose to use mathematical methods of optimal control theory and linear programming, adaptive method. The generalized algorithm is represented and verified on numerical examples.
This paper proposes a generalized algorithm of detecting six-degree-of-freedom (6-DOF) motions of an object with unknown surface curvature. The proposed system is a 6-DOF motion detection system for long distance rang...
详细信息
ISBN:
(纸本)9781479957361
This paper proposes a generalized algorithm of detecting six-degree-of-freedom (6-DOF) motions of an object with unknown surface curvature. The proposed system is a 6-DOF motion detection system for long distance range based on low cost laser distance sensors and a vision sensor. This does not require prior information on the geometric configurations of the feature points nor the surface curvature of the object. Simulations are conducted to validate the proposed algorithm on a regularly curved object.
We propose a general algorithm for the systematic design of feedback strategies to stabilize the evolutionary dynamics of a generic disease model using an H-infinity approach. We show that designing therapy concentrat...
详细信息
ISBN:
(纸本)9781467357173
We propose a general algorithm for the systematic design of feedback strategies to stabilize the evolutionary dynamics of a generic disease model using an H-infinity approach. We show that designing therapy concentrations can be cast as an H-infinity state feedack synthesis problem, where the feedback gain is constrained to not only be strictly diagonal, but also that its diagonal elements satisfy an overdetermined set of linear equations. Leveraging recent results in positive systems, we develop an algorithm that always yields a stabilizing controller.
Following Oriols (2007 Phys. Rev. Lett. 98 066803), an algorithm to deal with the exchange interaction in non-separable quantum systems is presented. The algorithm can be applied to fermions or bosons and, by construc...
详细信息
Following Oriols (2007 Phys. Rev. Lett. 98 066803), an algorithm to deal with the exchange interaction in non-separable quantum systems is presented. The algorithm can be applied to fermions or bosons and, by construction, it exactly ensures that any observable is totally independent of the interchange of particles. It is based on the use of conditional Bohmian wave functions which are solutions of single-particle pseudo-Schrodinger equations. The exchange symmetry is directly defined by demanding symmetry properties of the quantum trajectories in the configuration space with a universal algorithm, rather than through a particular exchange-correlation functional introduced into the single-particle pseudo-Schrodinger equation. It requires the computation of N-2 conditional wave functions to deal with N identical particles. For separable Hamiltonians, the algorithm reduces to the standard Slater determinant for fermions (or permanent for bosons). A numerical test for a two-particle system, where exact solutions for non-separable Hamiltonians are computationally accessible, is presented. The numerical viability of the algorithm for quantum electron transport (in a far-from-equilibrium time-dependent open system) is demonstrated by computing the current and fluctuations in a nano-resistor, with exchange and Coulomb interactions among electrons.
暂无评论