Vol. 37 · No. 1

The recent years have significantly changed the perception of web services and data storages, as clouds became a big part of IT market. New challenges appear in the field of scalable web systems, which become bigger and more complex. One of them is designing load balancing algorithms that could allow for optimal utilization of servers' resources in large, distributed systems. This paper presents an algorithm called Two-Level Load Balancing, which has been implemented and evaluated in onedata – a global data access system. A study of onedata architecture, request types and use cases has been performed to determine the requirements of load balancing set by similar, highly scalable distributed systems. The algorithm was designed to match these requirements, and it was achieved by using a synergy of DNS and internal dispatcher load balancing. Test results show that the algorithm does not introduce considerable overheads and maintains the performance of the system on high level, even in cases when its servers are not equally loaded.

The progress of false information diffusion in the public crisis is harmful to the society. When the public crisis occurs, the public respond in different ways and the public also want to tell others what they think right. But what they think is right is not recognized by the government. Thus the false information forms and it begins to diffuse. As the false information spreads, the harm to society magnifies gradually. Particularly in network society, false information diffusion can easily cause secondary hazards and accelerate public crises to a devastating degree. Thus intervening and controlling the false information diffusion is an important aspect of the public crisis management. From the perspective of the social network theory, this study analyzes the progress of false information diffusion in terms of different public crisis management strategies and presents the result of false information diffusion through simulation on cellular automaton of different public crisis management strategies. In simulations on cellular automaton, interventions are also carried to control false information diffusion and alternatives are proposed to help reduce public crises. This study also extends the theory of false information management, which is significant for the government to improve the ability to evaluate the false information and carry out interventions effectively to control the false information when it begins to diffuse.

Large-scale continuous optimization has gained considerable attention in recent years. Differential evolution (DE) is a simple yet efficient global numerical optimization algorithm, which has been successfully used in diverse fields. Generally, the vectors in the DE mutation operators are chosen randomly from the population. In this paper, we employ the ranking-based mutation operators for the DE algorithm to improve DE's performance. In the ranking-based mutation operators, the vectors are selected according to their rankings in the current population. The ranking-based mutation operators are general, and they are integrated into the original DE algorithm, GODE, and GaDE to verify the enhanced performance. Experiments have been conducted on the large-scale continuous optimization problems. The results indicate that the ranking-based mutation operators are able to enhance the overall performance of DE, GODE, and GaDE in the large-scale continuous optimization problems.

In cloud environment resources are dynamically allocated, adjusted, and deallocated. When to allocate and how many resources to allocate is a challenging task. Resources allocated optimally and at the right time not only improve the utilization of resources but also increase energy efficiency, provider's profit and customers' satisfaction. This paper presents ant colony optimization (ACO) based energy aware solution for resource allocation problem. The proposed energy aware resource allocation (EARA) methodology strives to optimize allocation of resources in order to improve energy efficiency of the cloud infrastructure while satisfying quality of service (QoS) requirements of the end users. Resources are allocated to jobs according to their QoS requirements. For energy efficient and QoS aware allocation of resources, EARA uses ACO at two levels. First level ACO allocates Virtual Machines (VMs) resources to jobs whereas second level ACO allocates Physical Machines (PMs) resources to VMs. Server consolidation and dynamic performance scaling of PMs are employed to conserve energy. The proposed methodology is implemented in CloudSim and the results are compared with existing popular resource allocation methods. Simulation results demonstrate that EARA achieves desired QoS and superior energy gains through better utilization of resources. EARA outperforms major existing resource allocation methods and achieves up to 10.56

Software Engineering is considered a knowledge-intensive discipline, in which knowledge creation, collection and sharing is an uninterrupted process. However, a large part of this knowledge exists in a tacit form and depends on practitioners. Therefore defining a mechanism to transform tacit knowledge into explicit one is of upmost importance. This paper presents a formalization approach to represent Software Engineering practitioners' tacit knowledge, which is related to their ways of working, as a set of explicit statements. The formalization is based on KUALI-BEH, which is a normative kernel extension of ESSENCE formal specification, and consists of three parts: an ontology to share a common representation of knowledge as a set of concepts; a Situational Method Engineering based algebra that represents well-defined method properties and operations; and a knowledge representation of the ontology and algebra using Description Logics. The main objectives of this initial formalization are to improve communication among humans and machines, computational inference and reuse of knowledge.

This paper explores the problem of processing the immense volume of measurement data arising during network traffic monitoring. Due to the ever-increasing demands of current networks, observing accurate information about every single flow is virtually infeasible. In many cases the existing methods for the reduction of flow records are still not sufficient enough. Since the accurate knowledge of flows termed as ``heavy-hitters'' suffices to fulfill most of the monitoring purposes, we decided to aggregate the flow records pertaining to non-heavy-hitters. However, due to the ever-changing nature of traffic, their identification is a challenge. To overcome this challenge, our proposed approach – the adaptive aggregation of flow records – automatically adjusts its operation to the actual traffic load and to the monitoring requirements. Preliminary experiments in existing network topologies showed that adaptive aggregation efficiently reduces the number of flow records, while a significant proportion of traffic details is preserved.

Finding a program-specific compiler optimization sequence is a challenge, due to the large number of optimizations provided by optimizing compilers. As a result, researchers have proposed design-space exploration schemes. This paper also presents a design-space exploration scheme, which aims to search for a compiler optimization sequence. Our hybrid approach relies on sequences previously generated for a set of training programs, with the purpose of finding optimizations and their order of application. In the first step, a clustering algorithm chooses optimizations, and in the second step, a metaheuristic algorithm discovers the sequence, in which the compiler will apply each optimization. We evaluate our approach using the LLVM compiler, and an I7 processor, respectively. The results show that we can find optimization sequences that result in target codes that, when executed on the I7 processor, outperform the standard optimization level O3, by an average improvement of 8.01

Multithreaded programs are prone to concurrency errors such as deadlocks, race conditions and atomicity violations. These errors are notoriously difficult to detect due to the non-deterministic nature of concurrent software running on multicore hardware. Data races result from the concurrent access of shared data by multiple threads and can result in unexpected program behaviors. Main dynamic data race detection techniques in the literature are happens-before and lockset algorithms which suffer from high execution time and memory overhead, miss many data races or produce a high number of false alarms. Our goal is to improve the performance of dynamic data race detection, while at the same time improving its accuracy by generating fewer false alarms. We develop a hybrid data race detection algorithm that is a combination of the happens-before and lockset algorithms in a tool. Rather than focusing on individual memory accesses by each thread, we focus on sequence of memory accesses by each thread, called a segment. This allows us to improve the performance of data race detection. We implement several optimizations on our hybrid data race detector and compare our technique with traditional happens-before and lockset detectors. The experiments are performed with C/C++ multithreaded benchmarks using Pthreads library from PARSEC suite and large applications such as Apache web server. Our experiments showed that our hybrid detector is 15

This paper presents an algorithm for ellipse detection on stains of blood, which is directly suited for the needs of forensic analysis. The algorithm is of the edge-analyzing type. It performs convexity detection and is able to split contours of overlapping ellipses by concave regions analysis. A filtering is applied to the fit ellipses to reduce the results only to those results necessary for blood drop trajectory tracing. Results for running time and fitting quality tests performed on real-life and artificial data are presented. The solution answers to both quality and running time expectations of the field of application.

We study the parallelized ant system algorithm solving the traveling salesman problem on n cities. First, following the series of recent results for the graphics processing unit, we show that they translate to the PRAM (parallel random access machine) model. In addition, we develop a novel pheromone matrix update method under the PRAM CREW (concurrent-read exclusive-write) model and translate it to the graphics processing unit without atomic instructions. As a consequence, we give new asymptotic bounds for the parallel ant system, resulting in step complexities O(n łg łg n) on CRCW (concurrent-read concurrent-write) and O(n łg n) on CREW variants of PRAM using n2 processors in both cases. Finally, we present an experimental comparison with the currently known pheromone matrix update methods on the graphics processing unit and obtain encouraging results.

Medical image registration is one of the fundamental tasks in medical image processing. It has various applications in field of image guided surgery (IGS) and computer assisted diagnosis (CAD). A set of non-linear methods have been already developed for inter-subject and intra-subject 3D medical image registration. However, efficient registration in terms of accuracy and speed is one of the most demanded of today surgical navigation (SN) systems. This paper is a result of a series of experiments which utilizes Fast Radial Basis Function (RBF) technique to register one or more medical images non-rigidly. Initially, a set of curves are extracted using a combined watershed and active contours algorithm and then tiled and converted to a regular surface using a global parameterization algorithm. It is shown that the registration accuracy improves when higher number of salient features (i.e. anatomical point landmarks and surfaces) are used and it also has no impact on the speed of the algorithm. The results show that the target registration error is less than 2 mm and has sub-second performance on intra-subject registration of MR image real datasets. It is observed that the Fast RBF algorithm is relatively insensitive to the increasing number of point landmarks used as compared with the competing feature based algorithms.