Vol. 38 · No. 2

Meta-heuristics utilizing numerous parameters are more complicated than meta-heuristics with a couple of parameters for various reasons. In essence, the effort expected to tune the strategy-particular parameters is far more prominent as the quantity of parameters increases and furthermore, complex algorithms are liable for the presence of further parameter interactions. Jaya meta-heuristic does not involve any strategy-specific parameters and is a one-stage technique. It has demonstrated its effectiveness compared to major types of meta-heuristics and it introduces various points of interest, such as its easy deployment and set-up in industrial applications and its low complexity to be studied. In this work, a new meta-heuristic, Enhanced Jaya (EJaya) is proposed to overcome the inconsistency of Jaya in diverse situations, introducing coherent attraction and repulsion movements and restrained intensity for flight. Comparative results of EJaya in a set of benchmark problems including statistical tests show that it is feasible to increase the accuracy, scalability and exploitation capability of Jaya while keeping its specific parameter-free feature. EJaya is especially suitable for a priori undefined characteristics optimization functions or applications where the set-up time of the optimization process is critical and parameters tuning and interactions must be avoided.

One important benefit of servers' virtualization is the reduction of the maintenance complexity of infrastructures. A key feature is servers' live migration which allows virtual servers to be exchanged between physical machines without stopping their services. However, virtualization also has some drawbacks caused by the overhead generated. Our research evaluated live migration process overhead, on real and virtual environments, noticed from the client's side regarding two different services: web and database. YCSB and ab Benchmark were adopted as workloads. Almost all tests on real environment overcame those on virtual, with both benchmarks. The impact of the live migration in the services was evident, proving to be more effective on real machines than on virtual machines. We found the DB service accommodated better to the virtual environment and to migration than Web service. We also considered an environment with multiple migrations which presented a higher degradation than when only one migration is performed.

Existing histogram based reversible data hiding schemes use only absolute difference values between the neighboring pixels of a cover image. In these schemes, maxima and minima points at maximum distance are selected in all the blocks of the image which causes shifting of the large number of pixels to embed the secret data. This shifting produces more degradation in the visual quality of the marked image. In this work, the cover image is segmented into blocks, which are classified further into complex and smooth blocks using a threshold value. This threshold value is optimized using firefly algorithm. Simple difference values between the neighboring pixels of complex blocks have been utilized to embed the secret data bits. The closest maxima and minima points in the histogram of the difference blocks are selected so that number of shifted pixels get reduced, which further reduces the distortion in the marked image. Experimental results prove that the proposed scheme has better performance as compared to the existing schemes. The scheme shows minimum distortion and large embedding capacity. Novelty of work is the usage of negative difference values of complex blocks for secret data embedding with the minimal number of pixel shifting.

Steganography is considered to be not only a science, but also a craft of concealing ongoing communication by hiding messages in unsuspicious cover do-cuments, such as texts, digital images, audio and video sequences. Its essential feature is the constant search for – often exceptionally creative – possibilities of concealing information. In computers, steganography often uses secondary memory and exchangeable memory media utilising file systems. This paper deals with the current state of the issues related to information hiding by means of hard disks, being the most important source of forensic data. This paper focuses on information hiding using the File Allocation Table (FAT) file system. It also proposes a novel multi-carrier algorithm of hiding information in file fragmentation. The algorithm provides flexibility of encoding the information to be hidden and makes steps toward optimization that allows reduction of interference with the current state of the file system, represented by the statistical values of the file fragmentation parameters.

We aim to speed up approximate keyword matching with the use of a lightweight, fixed-size block of data for each string, called a fingerprint. These work in a similar way to hash values; however, they can be also used for matching with errors. They store information regarding symbol occurrences using individual bits, and they can be compared against each other with a constant number of bitwise operations. In this way, certain strings can be deduced to be at least within the distance k from each other (using Hamming or Levenshtein distance) without performing an explicit verification. We show experimentally that for a preprocessed collection of strings, fingerprints can provide substantial speedups for k = 1, namely over 2.5 times for the Hamming distance and over 30 times for the Levenshtein distance. Tests were conducted on synthetic and real-world English and URL data.

Over the past decade, the fast advance of network technologies, hardware and middleware, as well as software resource sophistication has contributed to the emergence of new computational models. Consequently, there was a capacity increasing for efficient and effective use of resources distributed aiming to integrate them, in order to provide a widely distributed environment, which computational capacity could be used to solve complex computer problems. The two most challenging aspects of distributed systems are resource management and task scheduling. This work contributes to minimize such problems by i) aiming to reduce this problem through the use of migration techniques; ii) implementing a multicluster simulation environment with mechanisms for load balancing; iii) plus, the gang scheduling implementation algorithms will be analyzed through the use of metrics, in order to measure the schedulers performance in different situations. Thus, the results showed a better use of resources, implying operating costs reduction.

In the present paper, we propose RDGC, a reuse distance-based performance analysis approach for GPU cache hierarchy. RDGC models the thread-level parallelism in GPUs to generate appropriate cache reference sequence. Further, reuse distance analysis is extended to model the multi-partition/multi-port parallel caches and employed by RDGC to analyze GPU cache memories. RDGC can be utilized for architectural space exploration and parallel application development through providing hit ratios and transaction counts. The results of the present study demonstrate that the proposed model has an average error of 3.72

A conceptual data persistence architecture and methodology to evaluate its performance is created. Results of the empirical research indicate that the architecture created is convenient for high intensity processing of large ECM data volumes. The synthesis of SQL and NoSQL technology allows to handle high volume transactions on current data and full-text search on large sets of history data. The history data is moved from SQL to NoSQL data store thus allowing to use a cluster of commodity hardware to store major volume of ECM data. Scaling out (increasing a number of cluster nodes) is less expensive than scaling up (buying a more powerful server hardware) that is normally needed to upgrade SQL database.

Materialized views are important for optimizing Business Intelligence (BI) systems when they are designed without data cubes. Selecting candidate queries from large number of queries for materialized views is a challenging task. Most of the work done in the past involves finding out frequent queries from the past workload and creating materialized views from such queries by either manually analyzing workload or using approximate string matching algorithms using query text. Most of the existing methods suggest complete queries but ignore query components such as sub queries for creation of materialized views. This paper presents a novel method to determine on which queries and query components materialized views can be created to optimize aggregate and join queries by mining database of query execution plans which are in the form of binary trees. The proposed algorithm showed significant improvement in terms of more number of optimized queries because it is using the execution plan tree of the query as a basis of selection of query to be optimized using materialized views rather than choosing query text which is used by traditional methods. For selecting a correct set of queries to be optimized using materialized views, the paper proposes efficient specialized frequent tree component mining algorithm with novel heuristics to prune search space. These frequent components are used to determine the possible set of candidate queries for creation of materialized views. Experimentation on standard, real and synthetic data sets, and also the theoretical basis, proved that the proposed method is able to optimize a large number of queries with less number of materialized views and showed a significant improvement in performance compared to traditional methods.

This paper presents an approach to handling frequent variations of modeling languages and models. The approach is based on Domain-Specific Modeling and linking of modeling tools with adaptive Run-Time Systems. The applicability of our solution is illustrated on an example of domain-specific languages for robot control. Special attention was given to the following problems: 1) model-level debugging; 2) performing fast transformation of models to native code for various hardware platforms and operating systems; and 3) specification of views and view-based generation of applications for validation of meta-models, models, and generated code. The feedback for automated refinement of models and meta-models is provided by a custom adaptive Run-Time System. For the purpose of synchronizing models, meta-models, and the target Run-Time System, we introduce action reports, which allow model-level debugging. In order to simplify handling of frequent model variations, we have introduced the linguistic concept of a modifier.