Suchergebnisse

Introduction -- Modeling for Energy Demand Forecasting -- Data Pre-processing Methods -- Hybridizing Meta-heuristic Algorithms with CMM and QCM for SVR's Parameters Determination -- Hybridizing QCM with Dragonfly algorithm to Enrich the Solution Searching Be-haviors -- Phase Space Reconstruction and Recurrence Plot Theory .

Intro -- Preface -- Contents -- 1 The Importance of Agricultural and Meteorological Predictions Using Machine Learning Models -- 1.1 Introduction -- 1.2 The Necessity of Meteorological Variables Prediction -- 1.3 The Necessity of Agricultural Factors Prediction -- 1.4 Conclusion -- References -- 2 Structure of Particle Swarm Optimization (PSO) -- 2.1 Introduction -- 2.2 Structure of Particle Swarm Optimization -- 2.3 The Application of PSO in Meteorological Field -- 2.4 The Application of PSO in Agricultural Studies -- 2.5 The Application of PSO in Other Related Studies -- 2.6 Conclusion -- References -- 3 Structure of Shark Optimization Algorithm -- 3.1 Introduction -- 3.2 The Structure of Shark Algorithm -- 3.3 Application of SSO in Climate Studies -- 3.4 Application of SSO in Agricultural Studies -- 3.5 Application of SSO in Other Studies -- 3.6 Conclusion -- References -- 4 Sunflower Optimization Algorithm -- 4.1 Introduction -- 4.2 Applications of SFO in the Different Fields -- 4.3 Structure of Sunflower Optimization Algorithm -- References -- 5 Henry Gas Solubility Optimizer -- 5.1 Introduction -- 5.2 Application of HGSO in Different Fields -- 5.3 Structure of Henry Gas Solubility -- References -- 6 Structure of Crow Optimization Algorithm -- 6.1 Introduction -- 6.2 The Application of the COA -- 6.3 Mathematical Model of COA -- References -- 7 Structure of Salp Swarm Algorithm -- 7.1 Introduction -- 7.2 The Application of the Salp Swarm Algorithm in Different Fields -- 7.3 Structure of Salp Swarm Algorithm -- References -- 8 Structure of Dragonfly Optimization Algorithm -- 8.1 Introduction -- 8.2 Application of Dragonfly Optimization Algorithm -- 8.3 Structure of Dragonfly Optimization Algorithm -- References -- 9 Rat Swarm Optimization Algorithm -- 9.1 Introduction -- 9.2 Applications of Rat Swarm Algorithm.

RESUMEN: La red de interconexión es un concepto clave de los sistemas de computación paralelos. El primer aspecto que define una red de interconexión es su topología. Habitualmente, las redes escalables y eficientes en términos de coste y consumo energético tienen bajo diámetro y se basan en topologías que encaran el límite de Moore y en las que no hay diversidad de caminos mínimos. Una vez definida la topología, quedando implícitamente definidos los límites de rendimiento de la red, es necesario diseñar un algoritmo de enrutamiento que se acerque lo máximo posible a esos límites y debido a la ausencia de caminos mínimos, este además debe explotar los caminos no mínimos cuando el tráfico es adverso. Estos algoritmos de enrutamiento habitualmente seleccionan entre rutas mínimas y no mínimas en base a las condiciones de la red. Las rutas no mínimas habitualmente se basan en el algoritmo de balanceo de carga propuesto por Valiant, esto implica que doblan la longitud de las rutas mínimas y por lo tanto, la latencia soportada por los paquetes se incrementa. En cuanto a la tecnología, desde su introducción en entornos HPC a principios de los años 2000, Ethernet ha sido usado en un porcentaje representativo de los sistemas. Esta tesis introduce una implementación realista y competitiva de una red escalable y sin pérdidas basada en dispositivos de red Ethernet commodity, considerando topologías de bajo diámetro y bajo consumo energético y logrando un ahorro energético de hasta un 54%. Además, propone un enrutamiento sobre la citada arquitectura, en adelante QCN-Switch, el cual selecciona entre rutas mínimas y no mínimas basado en notificaciones de congestión explícitas. Una vez implementada la decisión de enrutar siguiendo rutas no mínimas, se introduce un enrutamiento adaptativo en fuente capaz de adaptar el número de saltos en las rutas no mínimas. Este enrutamiento, en adelante ACOR, es agnóstico de la topología y mejora la latencia en hasta un 28%. Finalmente, se introduce un enrutamiento dependiente de la topología, en adelante LIAN, que optimiza el número de saltos de las rutas no mínimas basado en las condiciones de la red. Los resultados de su evaluación muestran que obtiene una latencia cuasi óptima y mejora el rendimiento de algoritmos de enrutamiento actuales reduciendo la latencia en hasta un 30% y obteniendo un rendimiento estable y equitativo. ; ABSTRACT: Interconnection network is a key concept of any parallel computing system. The first aspect to define an interconnection network is its topology. Typically, power and cost-efficient scalable networks with low diameter rely on topologies that approach the Moore bound in which there is no minimal path diversity. Once the topology is defined, the performance bounds of the network are determined consequently, so a suitable routing algorithm should be designed to accomplish as much as possible of those limits and, due to the lack of minimal path diversity, it must exploit non-minimal paths when the traffic pattern is adversarial. These routing algorithms usually select between minimal and non-minimal paths based on the network conditions, where the non-minimal paths are built according to Valiant load-balancing algorithm. This implies that these paths double the length of minimal ones and then the latency supported by packets increases. Regarding the technology, from its introduction in HPC systems in the early 2000s, Ethernet has been used in a significant fraction of the systems. This dissertation introduces a realistic and competitive implementation of a scalable lossless Ethernet network for HPC environments considering low-diameter and low-power topologies. This allows for up to 54% power savings. Furthermore, it proposes a routing upon the cited architecture, hereon QCN-Switch, which selects between minimal and non-minimal paths per packet based on explicit congestion notifications instead of credits. Once the miss-routing decision is implemented, it introduces two mechanisms regarding the selection of the intermediate switch to develop a source adaptive routing algorithm capable of adapting the number of hops in the non-minimal paths. This routing, hereon ACOR, is topology-agnostic and improves average latency in all cases up to 28%. Finally, a topology-dependent routing, hereon LIAN, is introduced to optimize the number of hops in the non-minimal paths based on the network live conditions. Evaluations show that LIAN obtains almost-optimal latency and outperforms state-of-the-art adaptive routing algorithms, reducing latency by up to 30.0% and providing stable throughput and fairness. ; This work has been supported by the Spanish Ministry of Education, Culture and Sports under grant FPU14/02253, the Spanish Ministry of Economy, Industry and Competitiveness under contracts TIN2010-21291-C02-02, TIN2013-46957-C2-2-P, and TIN2013-46957-C2-2-P (AEI/FEDER, UE), the Spanish Research Agency under contract PID2019-105660RBC22/AEI/10.13039/501100011033, the European Union under agreements FP7-ICT-2011- 7-288777 (Mont-Blanc 1) and FP7-ICT-2013-10-610402 (Mont-Blanc 2), the University of Cantabria under project PAR.30.P072.64004, and by the European HiPEAC Network of Excellence through an internship grant supported by the European Union's Horizon 2020 research and innovation program under grant agreement No. H2020-ICT-2015-687689.