Dynamic task allocation and coordination in cooperative multi-agent environments

La coordinació i assignació de tasques en entorns distribuïts ha estat un punt important de la recerca en els últims anys i aquests temes són el cor dels sistemes multi-agent. Els agents en aquests sistemes necessiten cooperar i considerar els altres agents en les seves accions i decisions. A més a més, els agents han de coordinar-se ells mateixos per complir tasques complexes que necessiten més d'un agent per ser complerta. Aquestes tasques poden ser tan complexes que els agents poden no saber la ubicació de les tasques o el temps que resta abans de que les tasques quedin obsoletes. Els agents poden necessitar utilitzar la comunicació amb l'objectiu de conèixer la tasca en l'entorn, en cas contrari, poden perdre molt de temps per trobar la tasca dins de l'escenari. De forma similar, el procés de presa de decisions distribuït pot ser encara més complexa si l'entorn és dinàmic, amb incertesa i en temps real. En aquesta dissertació, considerem entorns amb sistemes multi-agent amb restriccions i cooperatius (dinàmics, amb incertesa i en temps real). En aquest sentit es proposen dues aproximacions que permeten la coordinació dels agents. La primera és un mecanisme semi-centralitzat basat en tècniques de subhastes combinatòries i la idea principal es minimitzar el cost de les tasques assignades des de l'agent central cap als equips d'agents. Aquest algoritme té en compte les preferències dels agents sobre les tasques. Aquestes preferències estan incloses en el bid enviat per l'agent. La segona és un aproximació d'scheduling totalment descentralitzat. Això permet als agents assignar les seves tasques tenint en compte les preferències temporals sobre les tasques dels agents. En aquest cas, el rendiment del sistema no només depèn de la maximització o del criteri d'optimització, sinó que també depèn de la capacitat dels agents per adaptar les seves assignacions eficientment. Addicionalment, en un entorn dinàmic, els errors d'execució poden succeir a qualsevol pla degut a la incertesa i error de accions individuals. A més, una part indispensable d'un sistema de planificació és la capacitat de re-planificar. Aquesta dissertació també proveeix una aproximació amb re-planificació amb l'objectiu de permetre als agent re-coordinar els seus plans quan els problemes en l'entorn no permeti la execució del pla. Totes aquestes aproximacions s'han portat a terme per permetre als agents assignar i coordinar de forma eficient totes les tasques complexes en un entorn multi-agent cooperatiu, dinàmic i amb incertesa. Totes aquestes aproximacions han demostrat la seva eficiència en experiments duts a terme en l'entorn de simulació RoboCup Rescue.

Learning task specific distributed paragraph representations using a 2-tier convolutional neural network

We introduce a type of 2-tier convolutional neural network model for learning distributed paragraph representations for a special task (e.g. paragraph or short document level sentiment analysis and text topic categorization). We decompose the paragraph semantics into 3 cascaded constitutes: word representation, sentence composition and document composition. Specifically, we learn distributed word representations by a continuous bag-of-words model from a large unstructured text corpus. Then, using these word representations as pre-trained vectors, distributed task specific sentence representations are learned from a sentence level corpus with task-specific labels by the first tier of our model. Using these sentence representations as distributed paragraph representation vectors, distributed paragraph representations are learned from a paragraph-level corpus by the second tier of our model. It is evaluated on DBpedia ontology classification dataset and Amazon review dataset. Empirical results show the effectiveness of our proposed learning model for generating distributed paragraph representations.

MASDScheGATS: a prototype system for dynamic scheduling

A manufacturing system has a natural dynamic nature observed through several kinds of random occurrences and perturbations on working conditions and requirements over time. For this kind of environment it is important the ability to efficient and effectively adapt, on a continuous basis, existing schedules according to the referred disturbances, keeping performance levels. The application of Meta-Heuristics and Multi-Agent Systems to the resolution of this class of real world scheduling problems seems really promising. This paper presents a prototype for MASDScheGATS (Multi-Agent System for Distributed Manufacturing Scheduling with Genetic Algorithms and Tabu Search).

Scheduling Grid Applications with Software Requirements

The aim of task scheduling is to minimize the makespan of applications, exploiting the best possible way to use shared resources. Applications have requirements which call for customized environments for their execution. One way to provide such environments is to use virtualization on demand. This paper presents two schedulers based on integer linear programming which schedule virtual machines (VMs) in grid resources and tasks on these VMs. The schedulers differ from previous work by the joint scheduling of tasks and VMs and by considering the impact of the available bandwidth on the quality of the schedule. Experiments show the efficacy of the schedulers in scenarios with different network configurations.

BEE COLONIES APPLIED TO MULTIPROCESSOR SCHEDULING

In order to achieve the high performance, we need to have an efficient scheduling of a parallelprogram onto the processors in multiprocessor systems that minimizes the entire executiontime. This problem of multiprocessor scheduling can be stated as finding a schedule for ageneral task graph to be executed on a multiprocessor system so that the schedule length can be minimize [10]. This scheduling problem is known to be NP- Hard.In multi processor task scheduling, we have a number of CPU’s on which a number of tasksare to be scheduled that the program’s execution time is minimized. According to [10], thetasks scheduling problem is a key factor for a parallel multiprocessor system to gain betterperformance. A task can be partitioned into a group of subtasks and represented as a DAG(Directed Acyclic Graph), so the problem can be stated as finding a schedule for a DAG to beexecuted in a parallel multiprocessor system so that the schedule can be minimized. Thishelps to reduce processing time and increase processor utilization. The aim of this thesis workis to check and compare the results obtained by Bee Colony algorithm with already generatedbest known results in multi processor task scheduling domain.

Geração de algoritmos de escalonamento para simulação de grades computacionais

Pós-graduação em Ciência da Computação - IBILCE

Escalonamento de tarefas em ambiente de simulação de grid computacional

Pós-graduação em Ciência da Computação - IBILCE

Biblioteca de escalonamento de tarefas em grid computacional - LIBTS

The use of computational grid simulators is particularly important for studying the algorithms of task scheduling. Through the simulators it’s possible to assess and compare the performance of different algorithms in various scenarios. Despite the simulation tools provide basic features for simulation in distributed environments, they don’t offer internal policies of task scheduling, so that the implementation of the algorithms must be realized by the user himself. Therefore, this study aims to present the library of task scheduling LIBTS (LIBrary Tasks Scheduling) which is developed and adapted to the SimGrid simulator to provide the users with a tool to analyze the algorithms of task scheduling in the computational grid.

Evolution of competing strategies in a threshold model for task allocation

A nature inspired decentralised multi-agent algorithm is proposed to solve a problem of distributed task allocation in which cities produce and store batches of different mail types. Agents must collect and process the mail batches, without global knowledge of their environment or communication between agents. The problem is constrained so that agents are penalised for switching mail types. When an agent process a mail batch of different type to the previous one, it must undergo a change-over, with repeated change-overs rendering the agent inactive. The efficiency (average amount of mail retrieved), and the flexibility (ability of the agents to react to changes in the environment) are investigated both in static and dynamic environments and with respect to sudden changes. New rules for mail selection and specialisation are proposed and are shown to exhibit improved efficiency and flexibility compared to existing ones. We employ a evolutionary algorithm which allows the various rules to evolve and compete. Apart from obtaining optimised parameters for the various rules for any environment, we also observe extinction and speciation.

Improving scheduling techniques in heterogeneous systems with dynamic, on-line optimisations

Computational performance increasingly depends on parallelism, and many systems rely on heterogeneous resources such as GPUs and FPGAs to accelerate computationally intensive applications. However, implementations for such heterogeneous systems are often hand-crafted and optimised to one computation scenario, and it can be challenging to maintain high performance when application parameters change. In this paper, we demonstrate that machine learning can help to dynamically choose parameters for task scheduling and load-balancing based on changing characteristics of the incoming workload. We use a financial option pricing application as a case study. We propose a simulation of processing financial tasks on a heterogeneous system with GPUs and FPGAs, and show how dynamic, on-line optimisations could improve such a system. We compare on-line and batch processing algorithms, and we also consider cases with no dynamic optimisations.

Efficient schedulability tests for real-time embedded systems with urgent routines

Task scheduling is one of the key mechanisms to ensure timeliness in embedded real-time systems. Such systems have often the need to execute not only application tasks but also some urgent routines (e.g. error-detection actions, consistency checkers, interrupt handlers) with minimum latency. Although fixed-priority schedulers such as Rate-Monotonic (RM) are in line with this need, they usually make a low processor utilization available to the system. Moreover, this availability usually decreases with the number of considered tasks. If dynamic-priority schedulers such as Earliest Deadline First (EDF) are applied instead, high system utilization can be guaranteed but the minimum latency for executing urgent routines may not be ensured. In this paper we describe a scheduling model according to which urgent routines are executed at the highest priority level and all other system tasks are scheduled by EDF. We show that the guaranteed processor utilization for the assumed scheduling model is at least as high as the one provided by RM for two tasks, namely 2(2√−1). Seven polynomial time tests for checking the system timeliness are derived and proved correct. The proposed tests are compared against each other and to an exact but exponential running time test.

Evolutionary computing in search-based software engineering

This master’s thesis aims to study and represent from literature how evolutionary algorithms are used to solve different search and optimisation problems in the area of software engineering. Evolutionary algorithms are methods, which imitate the natural evolution process. An artificial evolution process evaluates fitness of each individual, which are solution candidates. The next population of candidate solutions is formed by using the good properties of the current population by applying different mutation and crossover operations. Different kinds of evolutionary algorithm applications related to software engineering were searched in the literature. Applications were classified and represented. Also the necessary basics about evolutionary algorithms were presented. It was concluded, that majority of evolutionary algorithm applications related to software engineering were about software design or testing. For example, there were applications about classifying software production data, project scheduling, static task scheduling related to parallel computing, allocating modules to subsystems, N-version programming, test data generation and generating an integration test order. Many applications were experimental testing rather than ready for real production use. There were also some Computer Aided Software Engineering tools based on evolutionary algorithms.

Using combined static analysis and profiling for logic program execution time estimation

Predicting statically the running time of programs has many applications ranging from task scheduling in parallel execution to proving the ability of a program to meet strict time constraints. A starting point in order to attack this problem is to infer the computational complexity of such programs (or fragments thereof). This is one of the reasons why the development of static analysis techniques for inferring cost-related properties of programs (usually upper and/or lower bounds of actual costs) has received considerable attention.

Agent Based Models of Competition and Collaboration

Swarm intelligence is a popular paradigm for algorithm design. Frequently drawing inspiration from natural systems, it assigns simple rules to a set of agents with the aim that, through local interactions, they collectively solve some global problem. Current variants of a popular swarm based optimization algorithm, particle swarm optimization (PSO), are investigated with a focus on premature convergence. A novel variant, dispersive PSO, is proposed to address this problem and is shown to lead to increased robustness and performance compared to current PSO algorithms. A nature inspired decentralised multi-agent algorithm is proposed to solve a constrained problem of distributed task allocation. Agents must collect and process the mail batches, without global knowledge of their environment or communication between agents. New rules for specialisation are proposed and are shown to exhibit improved eciency and exibility compared to existing ones. These new rules are compared with a market based approach to agent control. The eciency (average number of tasks performed), the exibility (ability to react to changes in the environment), and the sensitivity to load (ability to cope with differing demands) are investigated in both static and dynamic environments. A hybrid algorithm combining both approaches, is shown to exhibit improved eciency and robustness. Evolutionary algorithms are employed, both to optimize parameters and to allow the various rules to evolve and compete. We also observe extinction and speciation. In order to interpret algorithm performance we analyse the causes of eciency loss, derive theoretical upper bounds for the eciency, as well as a complete theoretical description of a non-trivial case, and compare these with the experimental results. Motivated by this work we introduce agent "memory" (the possibility for agents to develop preferences for certain cities) and show that not only does it lead to emergent cooperation between agents, but also to a signicant increase in efficiency.

Progettazione ed analisi di algoritmi di mobilità controllata per reti mesh aeree con vincoli di copertura e persistenza

In questo lavoro di tesi verrà presentata l’implementazione di due algoritmi di Deployment e gestione di uno sciame di dispositivi UAV (Unmanned Aerial Vehicles). L’interesse scientifico su cui si fonda quest'analisi ha origine nelle enormi potenzialità degli UAV che garantiscono un'esplorazione aerea di aree pericolose in contesti di emergenze quali ad esempio scenari post catastrofe. La problematica principale affrontata sarà quella della gestione continuativa dell'area disastrata fornendo un algoritmo di schedulazione della cooperazione degli UAV alternando periodi attivi con quelli di ricarica dei dispositivi.