Distributed allocation and scheduling of applications on HPC platforms

High Performance Computing e una tecnologia usata dai cluster computazionali per creare sistemi di elaborazione che sono in grado di fornire servizi molto piu potenti rispetto ai computer tradizionali. Di conseguenza la tecnologia HPC e diventata un fattore determinante nella competizione industriale e nella ricerca. I sistemi HPC continuano a crescere in termini di nodi e core. Le previsioni indicano che il numero dei nodi arrivera a un milione a breve. Questo tipo di architettura presenta anche dei costi molto alti in termini del consumo delle risorse, che diventano insostenibili per il mercato industriale. Un scheduler centralizzato non e in grado di gestire un numero di risorse cosi alto, mantenendo un tempo di risposta ragionevole. In questa tesi viene presentato un modello di scheduling distribuito che si basa sulla programmazione a vincoli e che modella il problema dello scheduling grazie a una serie di vincoli temporali e vincoli sulle risorse che devono essere soddisfatti. Lo scheduler cerca di ottimizzare le performance delle risorse e tende ad avvicinarsi a un profilo di consumo desiderato, considerato ottimale. Vengono analizzati vari modelli diversi e ognuno di questi viene testato in vari ambienti.

Design and Performance Analysis of the Electrical Power System of a Petrochemical Facility

This thesis deals with the sizing and analysis of the electrical power system of a petrochemical plant. The activity was carried out in the framework of an electrical engineering internship. The sizing and electrical calculations, as well as the study of the dynamic behavior of network quantities, are accomplished by using the ETAP (Electrical Transient Analyzer Program) software. After determining the type and size of the loads, the calculation of power flows is carried out for all possible network layout and different power supply configurations. The network is normally operated in a double radial configuration. However, the sizing must be carried out taking into account the most critical configuration, i.e., the temporary one of single radial operation, and also considering the most unfavorable power supply conditions. The calculation of shortcircuit currents is then carried out and the appropriate circuit breakers are selected. Where necessary, capacitor banks are sized in order to keep power factor at the point of common coupling within the preset limits. The grounding system is sized by using the finite element method. For loads with the highest level of criticality, UPS are sized in order to ensure their operation even in the absence of the main power supply. The main faults that can occur in the plant are examined, determining the intervention times of the protections to ensure that, in case of failure on one component, the others can still properly operate. The report concludes with the dynamic and stability analysis of the power system during island operation, in order to ensure that the two gas turbines are able to support the load even during transient conditions.

Machine Learning in Automatic Tabulation for Constraint Model Reformulation

Combinatorial decision and optimization problems belong to numerous applications, such as logistics and scheduling, and can be solved with various approaches. Boolean Satisfiability and Constraint Programming solvers are some of the most used ones and their performance is significantly influenced by the model chosen to represent a given problem. This has led to the study of model reformulation methods, one of which is tabulation, that consists in rewriting the expression of a constraint in terms of a table constraint. To apply it, one should identify which constraints can help and which can hinder the solving process. So far this has been performed by hand, for example in MiniZinc, or automatically with manually designed heuristics, in Savile Row. Though, it has been shown that the performances of these heuristics differ across problems and solvers, in some cases helping and in others hindering the solving procedure. However, recent works in the field of combinatorial optimization have shown that Machine Learning (ML) can be increasingly useful in the model reformulation steps. This thesis aims to design a ML approach to identify the instances for which Savile Row’s heuristics should be activated. Additionally, it is possible that the heuristics miss some good tabulation opportunities, so we perform an exploratory analysis for the creation of a ML classifier able to predict whether or not a constraint should be tabulated. The results reached towards the first goal show that a random forest classifier leads to an increase in the performances of 4 different solvers. The experimental results in the second task show that a ML approach could improve the performance of a solver for some problem classes.

Feasibility study and flow scheduling optimization of a multi-product pipeline system

The first goal of this study is to analyse a real-world multiproduct onshore pipeline system in order to verify its hydraulic configuration and operational feasibility by constructing a simulation model step by step from its elementary building blocks that permits to copy the operation of the real system as precisely as possible. The second goal is to develop this simulation model into a user-friendly tool that one could use to find an “optimal” or “best” product batch schedule for a one year time period. Such a batch schedule could change dynamically as perturbations occur during operation that influence the behaviour of the entire system. The result of the simulation, the ‘best’ batch schedule is the one that minimizes the operational costs in the system. The costs involved in the simulation are inventory costs, interface costs, pumping costs, and penalty costs assigned to any unforeseen situations. The key factor to determine the performance of the simulation model is the way time is represented. In our model an event based discrete time representation is selected as most appropriate for our purposes. This means that the time horizon is divided into intervals of unequal lengths based on events that change the state of the system. These events are the arrival/departure of the tanker ships, the openings and closures of loading/unloading valves of storage tanks at both terminals, and the arrivals/departures of trains/trucks at the Delivery Terminal. In the feasibility study we analyse the system’s operational performance with different Head Terminal storage capacity configurations. For these alternative configurations we evaluated the effect of different tanker ship delay magnitudes on the number of critical events and product interfaces generated, on the duration of pipeline stoppages, the satisfaction of the product demand and on the operative costs. Based on the results and the bottlenecks identified, we propose modifications in the original setup.