Florian Fittkau

Search-based genetic optimization for deployment and reconfiguration of software in the cloud

Migrating existing enterprise software to cloud platforms involves the comparison of competing cloud deployment options (CDOs). A CDO comprises a combination of a specific cloud environment, deployment architecture, and runtime reconfiguration rules for dynamic resource scaling. Our simulator CDOSim can evaluate CDOs, e.g., regarding response times and costs. However, the design space to be searched for well-suited solutions is extremely huge. In this paper, we approach this optimization problem with the novel genetic algorithm CDOXplorer. It uses techniques of the search-based software engineering field and CDOSim to assess the fitness of CDOs. An experimental evaluation that employs, among others, the cloud environments Amazon EC2 and Microsoft Windows Azure, shows that CDOXplorer can find solutions that surpass those of other state-of-the-art techniques by up to 60%. Our experiment code and data and an implementation of CDOXplorer are available as open source software.

CDOSim: Simulating cloud deployment options for software migration support

The evaluation of competing cloud deployment options (CDOs) forms a major challenge when migrating software systems to cloud environments. For example, there exists a plethora of potential cloud provider candidates, components must be mapped to suitable virtual machine instances, and, to exploit elasticity, appropriate runtime adaptation strategies for specific usage profiles have to be defined. But analyzing potential CDOs manually is intractable, costly, and time-consuming due to the heterogeneity of the cloud environments and the overall combinatorial design space complexity. We present the simulation tool CDOSim that can simulate cost and performance properties of those CDOs. It builds upon and significantly extends the cloud simulator CloudSim and integrates into our cloud migration framework CloudMIG. Additionally, we created a cloud benchmark to augment CloudMIGs cloud environment models with provider-specific performance characteristics. Along with this simulation input, CDOSim utilizes reverse-engineered architectural models and can employ actual monitored workload. We report on extensive experiments incorporating Eucalyptus and Amazon EC2 which show that CDOSim can sufficiently accurate predict the cost and performance properties of CDOs.

Live trace visualization for comprehending large software landscapes: The ExplorViz approach

The increasing code complexity in modern enterprise software systems exceeds the capabilities of most software engineers to understand the systems behavior by just looking at its program code. Large software landscapes, e.g., applications in a cloud infrastructure, further increase this complexity. A solution to these problems is visualizing the applications of the software landscape to ease program comprehension and to understand the respective communication. An established visualization concept is the 3D city metaphor. It utilizes the familiarity with navigating a city to improve program comprehension. Dynamic analysis, e.g., monitoring, can provide the required program traces of the communication. In this paper, we present our live visualization approach of monitoring traces for large software landscapes. It combines a landscape and a system level perspective. The landscape level perspective provides an overview of the software landscape utilizing the viewers familiarity with UML. The system level perspective provides a visualization utilizing the city metaphor for each software system.

Exploring software cities in virtual reality

Software visualizations, such as the software city metaphor, are usually displayed on 2D screens and controlled by means of a mouse and thus often do not take advantage of more natural interaction techniques. Virtual reality (VR) approaches aim to improve the user experience. Emerging new technologies, like the Oculus Rift, dramatically enhance the VR experience at an affordable price. Therefore, new technologies have the potential to provide even higher immersion - and thus benefits - than previous VR approaches.

Cloud user-centric enhancements of the simulator cloudsim to improve cloud deployment option analysis

Cloud environments can be simulated using the toolkit CloudSim. By employing concepts such as physical servers in datacenters, virtual machine allocation policies, or coarse-grained models of deployed software, it focuses on a cloud provider perspective. In contrast, a cloud user who wants to migrate complex systems to the cloud typically strives to find a cloud deployment option that is best suited for its sophisticated system architecture, is interested in determining the best trade-off between costs and performance, or wants to compare runtime reconfiguration plans, for instance. We present significant enhancements of CloudSim that allow to follow this cloud user perspective and enable the frictionless integration of fine-grained application models that, to a great extent, can be derived automatically from software systems. Our quantitative evaluation demonstrates the applicability and accuracy of our approach by comparing its simulation results with actual deployments that utilize the cloud environment Amazon EC2.

EXPLORVIZ: VISUAL RUNTIME BEHAVIOR ANALYSIS OF ENTERPRISE APPLICATION LANDSCAPES

Enterprise application landscapes are complex and hard to manage systems. Enterprise models abstract from this complexity and seek to capture relevant information to cope with business requirements, i.e., increase flexibility while reducing costs of IT. Recent automated approaches focus on utilizing existing information sources. We identified two issues in current approaches. First, in line with these approaches we observe that enterprise models often get outdated and, thus, are an unreliable basis for decision making. Second, these approaches lack detail and after larger transformations commonly the entire application landscape exhibits an altered load profile. In this paper, we present how ExplorViz can be utilized to ensure consistency between an enterprise model and the actual information systems. We exemplify our approach via modeling the application landscape of the Kiel Data Management Infrastructure at the Helmholtz Centre for Ocean Research Kiel (GEOMAR). In our scenario, we depict an application landscape and exemplary drill-down to our EPrints productive installation to the source-code level. We explain the importance of underlying concepts and features, which help to optimize the responsiveness of an entire application landscape based on runtime information.

Software landscape and application visualization for system comprehension with ExplorViz

Hierarchical and multi-layer visualization of large software landscapes with ExplorViz, generated from monitoring traces.Reusable design and execution of a controlled experiment comparing a flat landscape visualization to our hierarchical landscape visualization in system comprehension tasks.Controlled experiments for comparing ExplorViz with the Extravis trace visualization approach, for employing physical 3D-printed ExplorViz models, and for exploring 3D ExplorViz models in virtual reality. Display Omitted Context: The number of software applications deployed in organizations is constantly increasing. Those applications often several hundreds form large software landscapes.Objective: The comprehension of such landscapes and their applications is often impeded by, for instance, architectural erosion, personnel turnover, or changing requirements. Therefore, an efficient and effective way to comprehend such software landscapes is required.Method: In our ExplorViz visualization, we introduce hierarchical abstractions aiming at solving system comprehension tasks fast and accurately for large software landscapes. Besides hierarchical visualization on the landscape level, ExplorViz provides multi-level visualization from the landscape to the level of individual applications. The 3D application-level visualization is empirically evaluated with a comparison to the Extravis approach, with physical models and in virtual reality. To evaluate ExplorViz, we conducted four controlled experiments. We provide packages containing all our experimental data to facilitate the verifiability, reproducibility, and further extensibility of our results.Results: We observed a statistical significant increase in task correctness of the hierarchical visualization compared to the flat visualization. The time spent did not show any significant differences. For the comparison with Extravis, we observed that solving program comprehension tasks using ExplorViz leads to a significant increase in correctness and in less or similar time spent. The physical models improved the team-based program comprehension process for specific tasks by initiating gesture-based interaction, but not for all tasks. The participants of our virtual reality experiment with ExplorViz rated the realized gestures for translation, rotation, and selection as highly usable. However, our zooming gesture was less favored.Conclusion: The results backup our claim that our hierarchical and multi-level approach enhances the current state of the art in landscape and application visualization for better software system comprehension, including new forms of interaction with physical models and virtual reality.

Comparing trace visualizations for program comprehension through controlled experiments

For efficient and effective program comprehension, it is essential to provide software engineers with appropriate visualizations of the programs execution traces. Empirical studies, such as controlled experiments, are required to assess the effectiveness and efficiency of proposed visualization techniques. We present controlled experiments to compare the trace visualization tools E xtravis and ExplorViz in typical program comprehension tasks. We replicate the first controlled experiment with a second one targeting a differently sized software system. In addition to a thorough analysis of the strategies chosen by the participants, we report on common challenges comparing trace visualization techniques. Besides our own replication of the first experiment, we provide a package containing all our experimental data to facilitate the verifiability, reproducibility and further extensibility of our presented results. Although subjects spent similar time on program comprehension tasks with both tools for a small-sized system, analyzing a larger software system resulted in a significant efficiency advantage of 28 percent less time spent by using ExplorViz. Concerning the effectiveness (correct solutions for program comprehension tasks), we observed a significant improvement of correctness for both object system sizes of 39 and 61 percent with ExplorViz.

Synchrovis: 3D visualization of monitoring traces in the city metaphor for analyzing concurrency

The increasing code complexity in modern software systems exceeds the capabilities of most software engineers to understand the systems behavior by just looking at its program code. The addition of concurrency issues through the advent of multi-core processors in the consumer market further escalates this complexity. A solution to these problems is visualizing a model of the system to ease program comprehension. Especially for the comprehension of concurrency issues, static information is often not sufficient. For this purpose, profiling and monitoring can provide additional information on the actual behavior of a system. An established visualization approach is the 3D city metaphor. It utilizes the familiarity with navigating a city to improve program comprehension. In this paper, we present our trace-based SynchroVis 3D visualization approach for concurrency. It employs the city metaphor to visualize both static and dynamic properties of software systems with a focus on illustrating the concurrent behavior. To evaluate our approach, we provide an open source implementation of our concepts and present an exemplary dining philosophers scenario showing its feasibility.

Hierarchical software landscape visualization for system comprehension: A controlled experiment

In many enterprises the number of deployed applications is constantly increasing. Those applications - often several hundreds - form large software landscapes. The comprehension of such landscapes is frequently impeded due to, for instance, architectural erosion, personnel turnover, or changing requirements. Therefore, an efficient and effective way to comprehend such software landscapes is required. The current state of the art often visualizes software landscapes via flat graph-based representations of nodes, applications, and their communication. In our ExplorViz visualization, we introduce hierarchical abstractions aiming at solving typical system comprehension tasks fast and accurately for large software landscapes. To evaluate our hierarchical approach, we conduct a controlled experiment comparing our hierarchical landscape visualization to a flat, state-of-the-art visualization. In addition, we thoroughly analyze the strategies employed by the participants and provide a package containing all our experimental data to facilitate the verifiability, reproducibility, and further extensibility of our results. We observed a statistically significant increase of 14 % in task correctness of the hierarchical visualization group compared to the flat visualization group in our experiment. The time spent on the system comprehension tasks did not show any significant differences. The results backup our claim that our hierarchical concept enhances the current state of the art in landscape visualization.