R has gained explicit text mining support with the tm package enabling statisticians to answer many interesting research questions via statistical analysis or modeling of (text) corpora. However, we typically face two challenges when analyzing large corpora: (1) the amount of data to be processed in a single machine is usually limited by the available main memory (i.e., RAM), and (2) the more data to be analyzed the higher the need for efficient procedures for calculating valuable results. Fortunately, adequate programming models like MapReduce facilitate parallelization of text mining tasks and allow for processing data sets beyond what would fit into memory by using a distributed file system possibly spanning over several machines, e.g., in a cluster of workstations. In this paper we present a plug-in package to tm called tm.plugin.dc implementing a distributed corpus class which can take advantage of the Hadoop MapReduce library for large scale text mining tasks. We show on the basis of an application in culturomics that we can efficiently handle data sets of signifficant size. (authors' abstract)

Since the turn of the century interest has grown in alternative models of leadership to reflect increased complexity and ambiguity, the need to respond faster to complex market conditions, and new patterns of accountability, inter- dependency and co-ordination within organisations of all types. This has led to the emergence of alternative models of leadership including shared and distributed leadership. In many organisations, such as those with matrix structures, many leaders need to accomplish organisational goals without formal line management authority over employees. This is also the case in many professional services (e.g. law and consultancy) that operate partnership models whereby individuals have little direct authority over their peers. In University settings the governance structure also impedes traditional hierarchical leadership. The tenure system, operated by many universities and colleges in the United States and Canada, provides intellectual autonomy, protects academics from external pressure and offers job security. Despite a growing literature on shared and distributed leadership, few studies have empirically examined the nature of leadership distribution, the contextual factors that impact leadership, and how those in senior positions (e.g. university department chairs) achieve organisational goals when employees (e.g. faculty members) possess significant authority and autonomy. This study addresses this gap. In so doing the study aims to contribute to the literature on shared and distributed leadership and provide important insight to assist positional leaders who possess limited direct authority in more effectively accomplishing their leadership goals.

Climate change has brought about legally-binding targets for Scotland, the U.K. and the E.U. to reduce greenhouse gas emissions and source a share of overall energy consumption from renewable energy resources by 2020. With severe limitations in the transport and heating sectors the onus is on the electricity sector to provide a significant reduction in greenhouse gas emissions and introduce a substantial increase in renewable energy production. The most attractive renewable energy resources are located in the geographic extremes of the country, far from the large population densities and high voltage, high capacity transmission networks. This means that the majority of renewable generation technologies will need to connect to the conventionally passive, lower voltage distribution networks. The integration of Distributed Generation (DG) is severely restricted by the technical limitations of the passively managed lower voltage infrastructure. Long lead times and the capital expenditure of traditional electricity network reinforcement can significantly delay or make the economics of some renewable generation schemes unviable. To be able to quickly and cost-effectively integrate significant levels of DG, the conventional fit-and-forget approach will have to be evolved into a ‘connect-and-manage’ system using active network management (ANM) techniques. ANM considers the real-time variation in generation and demand levels and schedules electricity network control settings to alleviate system constraints and increase connectable capacity of DG. This thesis explores the extent to which real time adjustments to DG and network asset controller set-points could allow existing networks to accommodate more DG. This thesis investigates the use of a full AC OPF technique to operate and schedule in real time variables of ANM control in distribution networks. These include; DG real and reactive power output and on-load-tap-changing transformer set-points. New formulations of the full AC OPF problem including multi-objective functions, penalising unnecessary deviation of variable control settings, and a Receding-Horizon formulation are assessed. This thesis also presents a methodology and modelling environment to explore the new and innovative formulations of OPF and to assess the interactions of various control practices in real time. Continuous time sequential, single scenario, OPF analyses at a very short control cycle can lead to the discontinuous and unnecessary switching of network control set-points, particularly during the less onerous network operating conditions. Furthermore, residual current flow and voltage variation can also gave rise to undesirable network effects including over and under voltage excursion and thermal overloading of network components. For the majority of instances, the magnitude of constraint violation was not significant but the levels of occurrence gave occasional cause for concern. The new formulations of the OPF problem were successful in deterring any extreme and unsatisfactory effects. Results have shown significant improvements in the energy yield from non-firm renewable energy resources. Initial testing of the real time OPF techniques in a simple demonstration network where voltage rise restricted the headroom for installed DG capacity and energy yield, showed that the energy yield for a single DG increased by 200% from the fit-and-forget scenario. Extrapolation of the OPF technique to a network with multiple DGs from different types of renewable energy resources showed an increase of 216% from the fit-and-forget energy yield. In a much larger network case study, where thermal loading limits constrained further DG capacity and energy yield, the increase in energy yield was more modest with an average increase of 45% over the fit-and-forget approach. In the large network where thermal overloading prevailed there was no immediate alternative to real power curtailment. This work has demonstrated that the proposed ANM OPF schemes can provide an intelligent, more cost effective and quicker alternative to network upgrades. As a result, DNOs can have a better knowledge and understanding of the capabilities and technical limitations of their networks to absorb DG safely and securely, without the expense of conventional network reinforcement.

In this thesis, we propose optimization techniques for distributed graph processing. First, we describe a data processing pipeline that leverages an iterative graph algorithm for automatic classification of web trackers. Using this application as a motivating example, we examine how asymmetrical convergence of iterative graph algorithms can be used to reduce the amount of computation and communication in large-scale graph analysis. We propose an optimization framework for fixpoint algorithms and a declarative API for writing fixpoint applications. Our framework uses a cost model to automatically exploit asymmetrical convergence and evaluate execution strategies during runtime. We show that our cost model achieves speedup of up to 1.7x and communication savings of up to 54%. Next, we propose to use the concepts of semi-metricity and the metric backbone to reduce the amount of data that needs to be processed in large-scale graph analysis. We provide a distributed algorithm for computing the metric backbone using the vertex-centric programming model. Using the backbone, we can reduce graph sizes up to 88% and achieve speedup of up to 6.7x. / <p>QC 20160919</p>

M.Sc. (Computer Science) / Please refer to full text to view abstract

Bush and Glover (2003) argue that, in order for principals and other leaders to focus on the management of teaching and learning, they need to be instructional leaders. Instructional leadership focuses on teaching and learning in a school, with a major emphasis on the man-agement of teaching and learning as a key activity of the principal. This research is, however, underpinned by Lambert‟s argument (2000), cited in MacNiel and McClanahan (2005:1), that one administrator cannot serve as the instructional leader for an entire school and that the par-ticipation of all other educators is necessary. This sharing and participation is necessitated by the fact that the task of management of teaching and learning is too huge a task for one per-son to accomplish. Furthermore, Elmore, cited in Harris (2004), points out that teaching and learning is a „knowledge-intensive enterprise‟ involving many complex tasks that cannot be performed without distributing the responsibility for leadership amongst others in the school. The inception of the new democratic dispensation in South Africa in 1994 has been associ-ated with a move to a decentralized system of schooling – a site-based education system. Thurlow (2003:27) has argued that inherent in this new model is a move towards institutional autonomy, to a more school - based management system (SBM). SBM involves the devolu-tion of power and responsibilities to principals, the empowerment of educators and increased participation of parents in the decision making process. This new system is vastly different from the „control‟ model of school leadership during the Apartheid era (Chisholm 1999), cited in Moloi (2007:466). Within the SBM system it becomes necessary for the school prin-cipal to share and distribute his/her leadership role in order to cope. Rutherford (2006), cited in Khumalo and Grant (2008:3) points out that the decentralizing of management in schools lends itself to the distribution of leadership throughout a school. Distributive leadership in-volves the view that leadership can be distributed or shared among those not only in formal leadership positions but those members of staff not in leadership positions. The movement towards a decentralized or school based management system falls within the transformation agenda of education in the new South Africa, which is committed to building democracy in schools by emphasizing the sharing or distribution of school management. This study entailed a small scale purposeful case study of two secondary schools in Gauteng, investigating the role of the principal in managing teaching and learning. It examined how vi and to whom principals distributed the management of teaching and learning in schools. Various qualitative research methods and approaches were used to collect relevant informa-tion on the role of the principal in the management of teaching and learning, and on how the management of teaching and learning was distributed in the two schools. Questionnaires and structured interviews were used to collect relevant data from Principals, deputies and HODs. The research findings revealed that the principal‟s role in managing teaching and learning is to create the necessary environment that will enable effective teaching and learning to take place. Principals therefore ensure that educators have all the necessary resources to teach, that educators are in class, that discipline is maintained and that educators are prepared to teach. It can therefore be argued, as Kruger did (2003:209), that the principal‟s role in managing teaching and learning is an indirect and supportive role. The study did, however, show that the task of managing teaching and learning was predomi-nantly that of the HODs. They were involved in monitoring of educators‟ work through learner‟s books, conducting class visits on a regular basis to observe educators teaching, en-suring that educators planned for lessons and had the necessary resources to teach. The study revealed that the leadership distribution in both schools was based either on the hierarchy of the school or distributed among staff as a whole. An important criterion for choosing who would be involved in sharing tasks was the skills and knowledge of the staff. Both schools tended to distribute tasks among those staff that had the necessary skills and knowledge to complete tasks successfully.

Thesis advisor: Lauri Johnson / This qualitative case study examined the perceived influence of distributed leadership on the job satisfaction of novice teachers in one Massachusetts school district. Dinham and Scott’s “Three Factor Theory” (1998), which emphasizes the role of school-based factors on job satisfaction, was utilized as a conceptual framework to guide the study. Data were gathered from interviews with novice teachers, school leaders, and district leaders and document review. The results of the study found that novice teachers involved in leadership activities via a distributed model perceived increased job satisfaction. Novice teachers who did not participate in leadership activities expressed a neutral or negative feeling towards their lack of involvement. A key determining factor in whether novice teachers participated in leadership activities was having a principal who encouraged this practice. However, all novice teachers interviewed reported that collegial staff relationships, regardless of their direct participation in leadership activities, positively influenced their job satisfaction. This study confirmed many of Dinham and Scott’s findings about the influence of school-based factors on job satisfaction; however, a conflicting finding about the relationship between school-based factors and intrinsic factors was identified. Recommendations include aligning school leaders’ practices in supporting the participation of novice teachers in leadership activities and mapping distributed leadership structures using a tool such as social network analysis. / Thesis (EdD) — Boston College, 2016. / Submitted to: Boston College. Lynch School of Education. / Discipline: Educational Leadership and Higher Education.

Thesis advisor: Diana C. Pullin / As district- and school-level leaders face increasing pressure from federal, state, and local accountability mandates there has been increased dependence on using and analyzing student data to help improve student performance. While the reporting of disaggregated data by student subgroup confirms that achievement gaps exist, it does not provide district- and school-level leaders with the diagnostic data needed to identify key factors inhibiting student performance. Identifying and understanding factors hindering student performance is critical knowledge for leaders to cultivate as they work to address elements within their school or district that may need to change if student learning is to improve. This research study examined specific ways district- and school-level leaders go about challenging and helping their community to face the problem of student performance disparities, as well as specific aspects of the situation that may be contributing to the community's collective capacity, to address student performance disparities. Without proper district-level leadership, effectively addressing operational conditions that may lead to disparities in student learning is unlikely. Yet, little is known about which structures and routines district- and school-level leaders perceive to be important when analyzing student data. This single case study presents the results of an examination of student data analysis structures and routines within a small diverse urban Massachusetts district designated by the state as low-performing based on state indicators. In order to further understand structures and routines, interview and document data were reviewed. Four primary findings identified the district leadership's response to educational inequities: (1) a mandate for using data war-rooms and student data walls; (2) a traveling cabinet to ensure uniform review of student data across the schools in the district; (3) a mandate for individual school improvement plans; and (4) the use of school-based instructional coaches. / Thesis (EdD) — Boston College, 2014. / Submitted to: Boston College. Lynch School of Education. / Discipline: Educational Leadership and Higher Education.

by Sau-ming Lau. / Thesis (M.Phil.)--Chinese University of Hong Kong, 1995. / Includes bibliographical references (leaves 131-134). / Abstract --- p.i / Acknowledgement --- p.iii / List of Tables --- p.viii / List of Figures --- p.x / Chapter 1 --- Introduction --- p.1 / Chapter 2 --- Basic Concepts and Related Work --- p.9 / Chapter 2.1 --- Components of Dynamic Load Balancing Algorithms --- p.10 / Chapter 2.2 --- Classification of Load Balancing Algorithms --- p.11 / Chapter 2.2.1 --- Casavant and Kuhl's Taxonomy --- p.12 / Chapter 3 --- System Model and Assumptions --- p.19 / Chapter 3.1 --- The System Model and Assumptions --- p.19 / Chapter 3.2 --- Survey on Cost Models --- p.21 / Chapter 3.2.1 --- "Eager, Lazowska, and Zahorjan's Model" --- p.22 / Chapter 3.2.2 --- "Shivaratri, Krueger, and Singhal's Model" --- p.23 / Chapter 3.3 --- Our Cost Model --- p.24 / Chapter 3.3.1 --- Design Philosophy --- p.24 / Chapter 3.3.2 --- Polling Query Cost Model --- p.25 / Chapter 3.3.3 --- Load State Broadcasting Cost Model --- p.26 / Chapter 3.3.4 --- Task Assignment Cost Model --- p.27 / Chapter 3.3.5 --- Task Migration Cost Model --- p.28 / Chapter 3.3.6 --- Execution Priority --- p.29 / Chapter 3.3.7 --- Simulation Parameter Values --- p.31 / Chapter 3.4 --- Performance Metrics --- p.33 / Chapter 4 --- A Performance Study on Load Information Dissemination Strategies --- p.36 / Chapter 4.1 --- Algorithm Descriptions --- p.37 / Chapter 4.1.1 --- Transfer Policy --- p.37 / Chapter 4.1.2 --- Information Policy --- p.40 / Chapter 4.1.3 --- Location Policy --- p.40 / Chapter 4.1.4 --- Categorization of the Algorithms --- p.43 / Chapter 4.2 --- Simulations and Analysis of Results --- p.43 / Chapter 4.2.1 --- Performance Comparisons --- p.44 / Chapter 4.2.2 --- Effect of Imbalance Factor on AWLT Algorithms --- p.49 / Chapter 4.2.3 --- Comparison of Average Performance --- p.52 / Chapter 4.2.4 --- Raw Simulation Results --- p.54 / Chapter 4.3 --- Discussions --- p.55 / Chapter 5 --- Resolving Processor Thrashing with Batch Assignment --- p.56 / Chapter 5.1 --- The GR.batch Algorithm --- p.57 / Chapter 5.1.1 --- The Guarantee and Reservation Protocol --- p.57 / Chapter 5.1.2 --- The Location Policy --- p.58 / Chapter 5.1.3 --- Batch Size Determination --- p.60 / Chapter 5.1.4 --- The Complete GR.batch Description --- p.62 / Chapter 5.2 --- Additional Performance Metrics --- p.66 / Chapter 5.3 --- Simulations and Analysis of Results --- p.67 / Chapter 5.4 --- Discussions --- p.73 / Chapter 6 --- Applying Batch Assignment to Systems with Bursty Task Arrival Patterns --- p.75 / Chapter 6.1 --- Bursty Workload Pattern Characterization Model --- p.76 / Chapter 6.2 --- Algorithm Descriptions --- p.77 / Chapter 6.2.1 --- The GR.batch Algorithm --- p.77 / Chapter 6.2.2 --- The SK .single Algorithm --- p.77 / Chapter 6.2.3 --- Summary of Algorithm Properties --- p.77 / Chapter 6.3 --- Analysis of Simulation Results --- p.77 / Chapter 6.3.1 --- Performance Comparison --- p.79 / Chapter 6.3.2 --- Time Trace --- p.80 / Chapter 6.4 --- Discussions --- p.80 / Chapter 7 --- A Preliminary Study on Task Assignment Augmented with Migration --- p.87 / Chapter 7.1 --- Algorithm Descriptions --- p.87 / Chapter 7.1.1 --- Information Policy --- p.88 / Chapter 7.1.2 --- Location Policy --- p.88 / Chapter 7.1.3 --- Transfer Policy --- p.88 / Chapter 7.1.4 --- The Three Load Balancing Algorithms --- p.89 / Chapter 7.2 --- Simulations and Analysis of Results --- p.90 / Chapter 7.2.1 --- Even Task Service Time --- p.90 / Chapter 7.2.2 --- Uneven Task Service Time --- p.94 / Chapter 7.3 --- Discussions --- p.99 / Chapter 8 --- Assignment Augmented with Migration Revisited --- p.100 / Chapter 8.1 --- Algorithm Descriptions --- p.100 / Chapter 8.1.1 --- The GR.BATCH.A Algorithm --- p.101 / Chapter 8.1.2 --- The SK.SINGLE.AM Algorithm --- p.101 / Chapter 8.1.3 --- Summary of Algorithm Properties --- p.101 / Chapter 8.2 --- Simulations and Analysis of Results --- p.101 / Chapter 8.2.1 --- Performance Comparisons --- p.102 / Chapter 8.2.2 --- Effect of Workload Imbalance --- p.105 / Chapter 8.3 --- Discussions --- p.106 / Chapter 9 --- Applying Batch Transfer to Heterogeneous Systems with Many Task Classes --- p.108 / Chapter 9.1 --- Heterogeneous System Model --- p.109 / Chapter 9.1.1 --- Processing Node Specification --- p.110 / Chapter 9.1.2 --- Task Type Specification --- p.111 / Chapter 9.1.3 --- Workload State Measurement --- p.112 / Chapter 9.1.4 --- Task Selection Candidates --- p.113 / Chapter 9.2 --- Algorithm Descriptions --- p.115 / Chapter 9.2.1 --- First Category ´ؤ The Sk .single Variations --- p.115 / Chapter 9.2.2 --- Second Category ´ؤ The GR. batch Variation Modeled with SSP --- p.117 / Chapter 9.3 --- Analysis of Simulation Results --- p.123 / Chapter 10 --- Conclusions and Future Work --- p.127 / Bibliography --- p.131 / Appendix A System Model Notations and Definitions --- p.131 / Appendix A.1 Processing Node Model --- p.131 / Appendix A.2 Cost Models --- p.132 / Appendix A.3 Load Measurement --- p.134 / Appendix A.4 Batch Size Determination Rules --- p.135 / Appendix A.5 Bursty Arrivals Modeling --- p.135 / Appendix A.6 Heterogeneous Systems Modeling --- p.135 / Appendix B Shivaratri and Krueger's Location Policy --- p.137

Large-scale distributed systems and distributed computing are the pillars of IT infrastructure and society nowadays. Robust theoretical principles for designing, building, managing and understanding the interactive behaviours of such systems need to be explored. A promising approach for establishing such principles is to view the session as the key unit for design, execution and verification. Governance is a general term for verifying whether activities meet the specified requirements and for enforcing safe behaviours among processes. This thesis, based on the asynchronous -calculus and the theory of session types, provides a monitoring framework and a theory for validating specifications, verifying mutual behaviours during runtime, and taking actions when noncompliant behaviours are detected. We explore properties and principles for governing large-scale distributed systems, in which autonomous and heterogeneous system components interact with each other in the network to accomplish application goals. This thesis, incorporating lessons from my participation in a substantial practical project, the Ocean Observatories Initiative (OOI), proposes an asynchronous monitoring framework and the process calculus for dynamically governing the asynchronous interactions among distributed multiple applications. We prove that this monitoring model guarantees the satisfaction of global assertions, and state and prove theorems of local and global safety, transparency, and session fidelity. We also study and introduce the semantic mechanisms for runtime session-based governance and the principles of validation of stateful specifications through capturing the runtime asynchronous interactions.