Algorithm-oriented design of efficient many-core architectures applied to dense matrix multiplication

Recent integrated circuit technologies have opened the possibility to design parallel architectures with hundreds of cores on a single chip. The design space of these parallel architectures is huge with many architectural options. Exploring the design space gets even more difficult if, beyond performance and area, we also consider extra metrics like performance and area efficiency, where the designer tries to design the architecture with the best performance per chip area and the best sustainable performance. In this paper we present an algorithm-oriented approach to design a many-core architecture. Instead of doing the design space exploration of the many core architecture based on the experimental execution results of a particular benchmark of algorithms, our approach is to make a formal analysis of the algorithms considering the main architectural aspects and to determine how each particular architectural aspect is related to the performance of the architecture when running an algorithm or set of algorithms. The architectural aspects considered include the number of cores, the local memory available in each core, the communication bandwidth between the many-core architecture and the external memory and the memory hierarchy. To exemplify the approach we did a theoretical analysis of a dense matrix multiplication algorithm and determined an equation that relates the number of execution cycles with the architectural parameters. Based on this equation a many-core architecture has been designed. The results obtained indicate that a 100 mm(2) integrated circuit design of the proposed architecture, using a 65 nm technology, is able to achieve 464 GFLOPs (double precision floating-point) for a memory bandwidth of 16 GB/s. This corresponds to a performance efficiency of 71 %. Considering a 45 nm technology, a 100 mm(2) chip attains 833 GFLOPs which corresponds to 84 % of peak performance These figures are better than those obtained by previous many-core architectures, except for the area efficiency which is limited by the lower memory bandwidth considered. The results achieved are also better than those of previous state-of-the-art many-cores architectures designed specifically to achieve high performance for matrix multiplication.

Task assignment algorithms for two-type heterogeneous multiprocessors

Consider the problem of assigning implicit-deadline sporadic tasks on a heterogeneous multiprocessor platform comprising two different types of processors—such a platform is referred to as two-type platform. We present two low degree polynomial time-complexity algorithms, SA and SA-P, each providing the following guarantee. For a given two-type platform and a task set, if there exists a task assignment such that tasks can be scheduled to meet deadlines by allowing them to migrate only between processors of the same type (intra-migrative), then (i) using SA, it is guaranteed to find such an assignment where the same restriction on task migration applies but given a platform in which processors are 1+α/2 times faster and (ii) SA-P succeeds in finding a task assignment where tasks are not allowed to migrate between processors (non-migrative) but given a platform in which processors are 1+α times faster. The parameter 0<α≤1 is a property of the task set; it is the maximum of all the task utilizations that are no greater than 1. We evaluate average-case performance of both the algorithms by generating task sets randomly and measuring how much faster processors the algorithms need (which is upper bounded by 1+α/2 for SA and 1+α for SA-P) in order to output a feasible task assignment (intra-migrative for SA and non-migrative for SA-P). In our evaluations, for the vast majority of task sets, these algorithms require significantly smaller processor speedup than indicated by their theoretical bounds. Finally, we consider a special case where no task utilization in the given task set can exceed one and for this case, we (re-)prove the performance guarantees of SA and SA-P. We show, for both of the algorithms, that changing the adversary from intra-migrative to a more powerful one, namely fully-migrative, in which tasks can migrate between processors of any type, does not deteriorate the performance guarantees. For this special case, we compare the average-case performance of SA-P and a state-of-the-art algorithm by generating task sets randomly. In our evaluations, SA-P outperforms the state-of-the-art by requiring much smaller processor speedup and by running orders of magnitude faster.

Task Assignment Algorithms for Heterogeneous Multiprocessors

Consider the problem of assigning implicit-deadline sporadic tasks on a heterogeneous multiprocessor platform comprising a constant number (denoted by t) of distinct types of processors—such a platform is referred to as a t-type platform. We present two algorithms, LPGIM and LPGNM, each providing the following guarantee. For a given t-type platform and a task set, if there exists a task assignment such that tasks can be scheduled to meet their deadlines by allowing them to migrate only between processors of the same type (intra-migrative), then: (i) LPGIM succeeds in finding such an assignment where the same restriction on task migration applies (intra-migrative) but given a platform in which only one processor of each type is 1 + α × t-1/t times faster and (ii) LPGNM succeeds in finding a task assignment where tasks are not allowed to migrate between processors (non-migrative) but given a platform in which every processor is 1 + α times faster. The parameter α is a property of the task set; it is the maximum of all the task utilizations that are no greater than one. To the best of our knowledge, for t-type heterogeneous multiprocessors: (i) for the problem of intra-migrative task assignment, no previous algorithm exists with a proven bound and hence our algorithm, LPGIM, is the first of its kind and (ii) for the problem of non-migrative task assignment, our algorithm, LPGNM, has superior performance compared to state-of-the-art.

Unified overhead-aware schedulability analysis for slot-based task-splitting

Hard real- time multiprocessor scheduling has seen, in recent years, the flourishing of semi-partitioned scheduling algorithms. This category of scheduling schemes combines elements of partitioned and global scheduling for the purposes of achieving efficient utilization of the system’s processing resources with strong schedulability guarantees and with low dispatching overheads. The sub-class of slot-based “task-splitting” scheduling algorithms, in particular, offers very good trade-offs between schedulability guarantees (in the form of high utilization bounds) and the number of preemptions/migrations involved. However, so far there did not exist unified scheduling theory for such algorithms; each one was formulated in its own accompanying analysis. This article changes this fragmented landscape by formulating a more unified schedulability theory covering the two state-of-the-art slot-based semi-partitioned algorithms, S-EKG and NPS-F (both fixed job-priority based). This new theory is based on exact schedulability tests, thus also overcoming many sources of pessimism in existing analysis. In turn, since schedulability testing guides the task assignment under the schemes in consideration, we also formulate an improved task assignment procedure. As the other main contribution of this article, and as a response to the fact that many unrealistic assumptions, present in the original theory, tend to undermine the theoretical potential of such scheduling schemes, we identified and modelled into the new analysis all overheads incurred by the algorithms in consideration. The outcome is a new overhead-aware schedulability analysis that permits increased efficiency and reliability. The merits of this new theory are evaluated by an extensive set of experiments.

CPMD-mindful task assignment for NPS-F

The multiprocessor scheduling scheme NPS-F for sporadic tasks has a high utilisation bound and an overall number of preemptions bounded at design time. NPS-F binpacks tasks offline to as many servers as needed. At runtime, the scheduler ensures that each server is mapped to at most one of the m processors, at any instant. When scheduled, servers use EDF to select which of their tasks to run. Yet, unlike the overall number of preemptions, the migrations per se are not tightly bounded. Moreover, we cannot know a priori which task a server will be currently executing at the instant when it migrates. This uncertainty complicates the estimation of cache-related preemption and migration costs (CPMD), potentially resulting in their overestimation. Therefore, to simplify the CPMD estimation, we propose an amended bin-packing scheme for NPS-F allowing us (i) to identify at design time, which task migrates at which instant and (ii) bound a priori the number of migrating tasks, while preserving the utilisation bound of NPS-F.

Provably Good Task Assignment for Two-Type Heterogeneous Multiprocessors Using Cutting Planes

Consider scheduling of real-time tasks on a multiprocessor where migration is forbidden. Specifically, consider the problem of determining a task-to-processor assignment for a given collection of implicit-deadline sporadic tasks upon a multiprocessor platform in which there are two distinct types of processors. For this problem, we propose a new algorithm, LPC (task assignment based on solving a Linear Program with Cutting planes). The algorithm offers the following guarantee: for a given task set and a platform, if there exists a feasible task-to-processor assignment, then LPC succeeds in finding such a feasible task-to-processor assignment as well but on a platform in which each processor is 1.5 × faster and has three additional processors. For systems with a large number of processors, LPC has a better approximation ratio than state-of-the-art algorithms. To the best of our knowledge, this is the first work that develops a provably good real-time task assignment algorithm using cutting planes.

Functional studies on magnetic resonance

Dissertation submitted in Faculdade de Ciências e Tecnologia of Universidade Nova de Lisboa for the degree of Master in Biomedical Engineering

Time-domain optimization of amplifiers based on distributed genetic algorithms

Thesis presented in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the subject of Electrical and Computer Engineering

Service-oriented mobility of java code in web services-based architectures

Dissertação apresentada na Faculdade de Ciências e Tecnologias da Universidade Nova de Lisboa para a obtenção do Grau de Mestre em Engenharia Informática

Balanced Scorecard

Mestrado Em Engenharia Mecânica - Ramo Gestão Industrial

Vertex component analysis: a geometric-based approach to unmix hyperspectral data

Hyperspectral remote sensing exploits the electromagnetic scattering patterns of the different materials at specific wavelengths [2, 3]. Hyperspectral sensors have been developed to sample the scattered portion of the electromagnetic spectrum extending from the visible region through the near-infrared and mid-infrared, in hundreds of narrow contiguous bands [4, 5]. The number and variety of potential civilian and military applications of hyperspectral remote sensing is enormous [6, 7]. Very often, the resolution cell corresponding to a single pixel in an image contains several substances (endmembers) [4]. In this situation, the scattered energy is a mixing of the endmember spectra. A challenging task underlying many hyperspectral imagery applications is then decomposing a mixed pixel into a collection of reflectance spectra, called endmember signatures, and the corresponding abundance fractions [8–10]. Depending on the mixing scales at each pixel, the observed mixture is either linear or nonlinear [11, 12]. Linear mixing model holds approximately when the mixing scale is macroscopic [13] and there is negligible interaction among distinct endmembers [3, 14]. If, however, the mixing scale is microscopic (or intimate mixtures) [15, 16] and the incident solar radiation is scattered by the scene through multiple bounces involving several endmembers [17], the linear model is no longer accurate. Linear spectral unmixing has been intensively researched in the last years [9, 10, 12, 18–21]. It considers that a mixed pixel is a linear combination of endmember signatures weighted by the correspondent abundance fractions. Under this model, and assuming that the number of substances and their reflectance spectra are known, hyperspectral unmixing is a linear problem for which many solutions have been proposed (e.g., maximum likelihood estimation [8], spectral signature matching [22], spectral angle mapper [23], subspace projection methods [24,25], and constrained least squares [26]). In most cases, the number of substances and their reflectances are not known and, then, hyperspectral unmixing falls into the class of blind source separation problems [27]. Independent component analysis (ICA) has recently been proposed as a tool to blindly unmix hyperspectral data [28–31]. ICA is based on the assumption of mutually independent sources (abundance fractions), which is not the case of hyperspectral data, since the sum of abundance fractions is constant, implying statistical dependence among them. This dependence compromises ICA applicability to hyperspectral images as shown in Refs. [21, 32]. In fact, ICA finds the endmember signatures by multiplying the spectral vectors with an unmixing matrix, which minimizes the mutual information among sources. If sources are independent, ICA provides the correct unmixing, since the minimum of the mutual information is obtained only when sources are independent. This is no longer true for dependent abundance fractions. Nevertheless, some endmembers may be approximately unmixed. These aspects are addressed in Ref. [33]. Under the linear mixing model, the observations from a scene are in a simplex whose vertices correspond to the endmembers. Several approaches [34–36] have exploited this geometric feature of hyperspectral mixtures [35]. Minimum volume transform (MVT) algorithm [36] determines the simplex of minimum volume containing the data. The method presented in Ref. [37] is also of MVT type but, by introducing the notion of bundles, it takes into account the endmember variability usually present in hyperspectral mixtures. The MVT type approaches are complex from the computational point of view. Usually, these algorithms find in the first place the convex hull defined by the observed data and then fit a minimum volume simplex to it. For example, the gift wrapping algorithm [38] computes the convex hull of n data points in a d-dimensional space with a computational complexity of O(nbd=2cþ1), where bxc is the highest integer lower or equal than x and n is the number of samples. The complexity of the method presented in Ref. [37] is even higher, since the temperature of the simulated annealing algorithm used shall follow a log( ) law [39] to assure convergence (in probability) to the desired solution. Aiming at a lower computational complexity, some algorithms such as the pixel purity index (PPI) [35] and the N-FINDR [40] still find the minimum volume simplex containing the data cloud, but they assume the presence of at least one pure pixel of each endmember in the data. This is a strong requisite that may not hold in some data sets. In any case, these algorithms find the set of most pure pixels in the data. PPI algorithm uses the minimum noise fraction (MNF) [41] as a preprocessing step to reduce dimensionality and to improve the signal-to-noise ratio (SNR). The algorithm then projects every spectral vector onto skewers (large number of random vectors) [35, 42,43]. The points corresponding to extremes, for each skewer direction, are stored. A cumulative account records the number of times each pixel (i.e., a given spectral vector) is found to be an extreme. The pixels with the highest scores are the purest ones. N-FINDR algorithm [40] is based on the fact that in p spectral dimensions, the p-volume defined by a simplex formed by the purest pixels is larger than any other volume defined by any other combination of pixels. This algorithm finds the set of pixels defining the largest volume by inflating a simplex inside the data. ORA SIS [44, 45] is a hyperspectral framework developed by the U.S. Naval Research Laboratory consisting of several algorithms organized in six modules: exemplar selector, adaptative learner, demixer, knowledge base or spectral library, and spatial postrocessor. The first step consists in flat-fielding the spectra. Next, the exemplar selection module is used to select spectral vectors that best represent the smaller convex cone containing the data. The other pixels are rejected when the spectral angle distance (SAD) is less than a given thresh old. The procedure finds the basis for a subspace of a lower dimension using a modified Gram–Schmidt orthogonalizati on. The selected vectors are then projected onto this subspace and a simplex is found by an MV T pro cess. ORA SIS is oriented to real-time target detection from uncrewed air vehicles using hyperspectral data [46]. In this chapter we develop a new algorithm to unmix linear mixtures of endmember spectra. First, the algorithm determines the number of endmembers and the signal subspace using a newly developed concept [47, 48]. Second, the algorithm extracts the most pure pixels present in the data. Unlike other methods, this algorithm is completely automatic and unsupervised. To estimate the number of endmembers and the signal subspace in hyperspectral linear mixtures, the proposed scheme begins by estimating sign al and noise correlation matrices. The latter is based on multiple regression theory. The signal subspace is then identified by selectin g the set of signal eigenvalue s that best represents the data, in the least-square sense [48,49 ], we note, however, that VCA works with projected and with unprojected data. The extraction of the end members exploits two facts: (1) the endmembers are the vertices of a simplex and (2) the affine transformation of a simplex is also a simplex. As PPI and N-FIND R algorithms, VCA also assumes the presence of pure pixels in the data. The algorithm iteratively projects data on to a direction orthogonal to the subspace spanned by the endmembers already determined. The new end member signature corresponds to the extreme of the projection. The algorithm iterates until all end members are exhausted. VCA performs much better than PPI and better than or comparable to N-FI NDR; yet it has a computational complexity between on e and two orders of magnitude lower than N-FINDR. The chapter is structure d as follows. Section 19.2 describes the fundamentals of the proposed method. Section 19.3 and Section 19.4 evaluate the proposed algorithm using simulated and real data, respectively. Section 19.5 presents some concluding remarks.

Multy-dimensional composition by objectives in aspect-oriented requirements analysis

Thesis submitted to Faculdade de Ciências e Tecnologia of Universidade Nova de Lisboa in partial fulfilment of the requirements for the degree of Master in Computer Science

Aplicação da metodologia de “lean management” no processo de desenvolvimento de produto

O presente trabalho de dissertação teve como objetivo a implementação de metodologias de Lean Management e avaliação do seu impacto no processo de Desenvolvimento de Produto. A abordagem utilizada consistiu em efetuar uma revisão da literatura e levantamento do Estado da Arte para obter a fundamentação teórica necessária à implementação de metodologias Lean. Prosseguiu com o levantamento da situação inicial da organização em estudo ao nível das atividades de desenvolvimento de produto, práticas de gestão documental e operacional e ainda de atividades de suporte através da realização de inquéritos e medições experimentais. Este conhecimento permitiu criar um modelo de referência para a implementação de Lean Management nesta área específica do desenvolvimento de produto. Após implementado, este modelo foi validado pela sua experimentação prática e recolha de indicadores. A implementação deste modelo de referência permitiu introduzir na Unidade de Desenvolvimento de Produto e Sistemas (DPS) da organização INEGI, as bases do pensamento Lean, contribuindo para a criação de um ambiente de Respeito pela Humanidade e de Melhoria Contínua. Neste ambiente foi possível obter ganhos qualitativos e quantitativos nas várias áreas em estudo, contribuindo de forma global para um aumento da eficiência e eficácia da DPS. Prevê-se que este aumento de eficiência represente um aumento da capacidade instalada na Organização, pela redução anual de 2290 horas de desperdício (6.5% da capacidade total da unidade) e pela redução significativa em custos operacionais. Algumas das implementações de melhoria propostas no decorrer deste trabalho, após verificado o seu sucesso, extravasaram a unidade em estudo e foram aplicadas transversalmente à da organização. Foram também obtidos ganhos qualitativos, tais como a normalização de práticas de gestão documental e a centralização e agilização de fluxos de informação. Isso permitiu um aumento de qualidade dos serviços prestados pela redução de correções e retrabalho. Adicionalmente, com o desenvolvimento de uma nova ferramenta que permite a monitorização do estado atual dos projetos a nível da sua percentagem de execução (cumprimento de objetivos), prazos e custos, bem como a estimação das datas de conclusão dos projetos possibilitando o replaneamento do projeto bem como a detecção atempada de desvios. A ferramenta permite também a criação de um histórico que identifica o esforço horário associado à realização das atividades/tarefas das várias áreas de Desenvolvimento de Produto e desta forma pode ser usada como suporte à orçamentação futura de atividades similares. No decorrer do projeto, foram também criados os mecanismos que permitem o cálculo de indicadores das competências técnicas e motivações intrínsecas individuais da equipa DPS. Estes indicadores podem ser usados na definição por parte dos gestores dos projetos da composição das equipas de trabalho, dos executantes de tarefas individuais do projeto e dos destinatários de ações de formação. Com esta informação é expectável que se consiga um maior aproveitamento do potencial humano e como consequência um aumento do desempenho e da satisfação pessoal dos recursos humanos da organização. Este caso de estudo veio demonstrar que o potencial de melhoria dos processos associados ao desenvolvimento de produto através de metodologias de Lean Management é muito significativo, e que estes resultam em ganhos visíveis para a organização bem como para os seus elementos individualmente.

Task-based learning : fonte de sucesso e motivação na apredizagem de uma língua estrangeira

Relatório apresentado para cumprimento dos requisitos necessários à obtenção do grau Mestre em Ensino de Inglês e de Língua Estrangeira (Espanhol) no 3º Ciclo do Ensino Básico e no Ensino Secundário

A middleware for service oriented computing in dynamic environments

Dissertação apresentada na Faculdade de Ciências e Tecnologia da Universidade Nova de Lisboa para a obtenção do Grau de Mestre em Engenharia Informática