Doctoral College "Adaptive Distributed Systems" -
Program

The program is designed for three years and consists of courses and research work to be conducted by the PhD candidates.

Jump to

PhD topics

The research context of the Doctoral College Adaptive Distributed Systems is divided into four thematic fields

  • (A) Foundation
  • (B) Middleware
  • (C) Modeling
  • (D) Applications

into which the individual PhD topics are embedded.


Doctoral College Research Scope with PhD topics

In the following, we give a short description of the individual PhD topics we are planning to offer to our students. For each topic, we explain its relationship to the research context. In particular, we show the assignment of each topic to its primary thematic field and secondary thematic field. Each PhD student will be closely supervised by two faculty members (first supervisor and second supervisor). The proposed topics may be adjusted upon commencement of employment according to recent research progress.

PhD Topic 1: White-Box Testing for Concurrent and Distributed Systems

The PhD thesis will develop white-box testing methods for concurrent and distributed systems. Based on our FQL testing framework for single-threaded code, we will investigate suitable notions of coverage, and define a specification language for coverage of concurrent and distributed systems. We will provide tools for multi-threaded C to generate test cases with a specified coverage; we will also write tools to assess the coverage achieved by other tools. The thesis will be co-funded (50%) by the National Research Network RiSE (see www.arise.or.at).
More information on topic #1

PhD Topic 2: Algorithms for Virtual Network Mapping in the Future Federated Internet

The Future Internet (FI) will consist mainly of a large number of rather isolated overlays for different applications, especially also including applications for social computing. Efficiency requirements for the FI prohibit the operation of a plurality of physically separated networks or overlays. Hence, the FI has to consolidate them into a single physical system. The use of Network Virtualization (NV) techniques can achieve this aim and has appealing advantages for distributed applications, e.g. the instant creation of slices with application-specific topology, routing, and resource management mechanisms. Therefore, NV is considered as the major paradigm in many international FI initiatives, e.g. GENI (USA), AKARI (Japan), and G-Lab (Germany).
More information on topic #2

PhD Topic 3: Engineering of Self-Adaptive Systems

Heterogeneous, large-scale, and dynamic software systems that typically run continuously often tend to become inert, brittle, and vulnerable after a while. As a result, these systems experience permanent dependability degradation throughout their life-time. This in turn requires continuous and highly responsive human maintenance intervention and repetitive software development processes. While this need for intervention is costly, error-prone, and hence further impairs dependability, it may, in some cases, even become prohibitively slow compared to the system's pace in normal operation.
Therefore, self-adaptation has been suggested as means against dependability degradation. Self-adaptive systems reason at run-time about their state, environment, and goals. This reasoning typically involves feedback control loops with four key activities Monitor, Analyze, Plan, and Execute, MAPE for short. While the idea is not new, there are no general solutions available so far: Current success stories of self-adaptive systems either focus on very specific application requirements or target particular application domains such as robotics.
In order to achieve more general solutions, several research questions have to be answered.
More information on topic #3

PhD Topic 4: Self-managed Compositions in Adaptive Distributed Systems

Context-dependent interactions ensure that all actors whose state and information view is impacted by a particular context are synchronized. This requires mechanisms to dynamically establish such a context, and to associate and share it with the mixed services implying a joint outcome. At runtime, the state of related mixed services must be tracked and in case agreement on the outcome is not achieved proper exception-handling actions must be initiated. Additionally, compositions evolve based on dynamically discovered interaction patterns implying the development of novel models for service compositions. Similarly, the definition of the contribution of each of these services to the exception-handling actions will require new mechanisms.
More information on topic 4

PhD Topic 5: Self-Optimizing Business Processes by Marrying Diagrammatical and Mathematical Models

Business processes started as simple step-by-step descriptions for specifying the solution for a business problem.
Nowadays, a plethora of diagrammatical modeling languages is available for describing complex process structures by using either general purpose modeling languages, e.g., flowcharts, Petri nets, and UML activity diagrams, or domain-specific languages such as BPMN, EPC, and YAWL. The advantage of these diagrammatical modeling languages is twofold. First, they are simple to understand, and thus, they are usable by business analysts forming the basis for communication between people. Second, these languages are directly executable by current IT infrastructures or at least may be transformed by code generators into executable artefacts. On the downside, the optimization of processes defined in diagrammatical languages is limited to the designer's intuition.
More information on topic 5

PhD Topic 6: A Visual Programming Language for Adaptive Real-World Robotic Applications

Upcoming robotic systems are highly complex, resource constraint dynamic systems that operate in a nondeterministic environment, the real world. In consequence, these systems have to handle uncertainty, unforeseen events, and even partial failures, in order to fulfill mission and safety critical tasks. One way to succeed under these conditions is that of adaptability. An autonomous mobile robotic system has to adapt to changing environmental conditions as much as to its internal state in presence of partial degradation.

It is obvious that the development of application software for these devices is difficult and error prone. In addition, programming a specific behavior on the one hand requires high expertise within the targeted field of application, and on the other hand (at least at the
moment) also within the field of distributed, real-time embedded systems programming.
More information on topic 6

PhD Topic 7: Adaptive Load Balancing for Distributed Computation

Computing the aerodynamic behavior of an aircraft or a racing car is nowadays still a computational challenge. The flow field is described by a partial differential equation, which is treated numerically by some discretization method such as the finite volume method, or, more recently, by a high order discontinuous Galerkin method . These methods are based on a computational mesh, which consists typically of a huge set of cells (100 millions). These cells are distributed onto a parallel computer. Since the flow fuild shows shocks, boundary layers, and singularities, a solution-driven adaptive mesh refinement for capturing these features is needed. In particular for time-dependent models a flexible load balancing technique is required. Efficient equation solving algorithms such as multi-level methods and domain decomposition methods (with sub-domain preconditioners) take advantage of the hierarchical structure of the refined mesh.
More information on topic 7

PhD Topic 8: Understanding Massively Distributed and Heterogeneous Systems

Whereas the other topics point at engineering and modeling perspectives of massively distributed and heterogeneous IT systems, the proposed PhD topic follows a more analytical (empirical) view on these infrastructures. It looks at the Web (as the most prominent representative of this class of systems) and its new problem solving capabilities. In this context the Web is seen as a combination of machines and humans, both interacting in order to master a huge set of different tasks. And it understands the Web in its role as a "data provider" describing human (and human - computer) interaction and communication. It sees the Web as a "mirror of the physical world" which opens a wide range of research possibilities.
More information on topic 8

PhD Topic 9: On Service Composition and Product Line Engineering

The Service-oriented Architecture (SOA) paradigm represents an enabling pattern for designing distributed systems. In information systems designed and implemented following this paradigm, functionality is typically encapsulated through so-called services. As a consequence and amongst other advantages, service-oriented systems support the strategic reuse of functionality by consuming different services. Another aspect of the SOA paradigm addresses the integration of existing, often heterogeneous, information systems which is typically achieved through boxing existing functionality in services. Having such services at hand eases the task of integrating heterogeneous, distributed, information systems.
More information on topic 9

PhD Topic 10: Adaptation for Heterogeneous Multi-Core Processors

This topic addresses the need for adaptation from the underlying systems perspective. Current and emerging processor architectures, also as parts or larger, distributed infrastructures, are both parallel (multi-core) and apply various types of accelerators to increase performance, lower power consumption or to be fast enough for special tasks that can arise in the distributed context (e.g., GPUs, routers, network cards). Heterogeneity and parallelism pose major challenges for (legacy) software to adapt statically to different system configurations (number of cores, memory system, types of accelerators) as well as dynamically to system load and availability. This kind of adaptation within a very large space of possibilities must be handled by automatic means to provide reasonable efficiency guarantees. A somewhat vague keyword is performance portability. To what extent this can be achieved is largely an open problem.
More information on topic 10

Curriculum

Following the general requirements for PhD studies, the doctoral students will have to select courses in the amount of 18 ECTS points from a list of offered courses. In addition to the 18 mandatory ECTS points, students are required to successfully pass another 6 ECTS. Thus, graduates of the Doctoral College Adaptive Distributed Systems have to successfully pass courses to the extent of 24 ECTS.

The 24 ECTS are divided into (i) mandatory courses and (ii) courses to be selected from a predefined list.

(i) Mandatory courses to be attended by all students of the Doctoral College


The following three courses must be attended by all students of the Doctoral College. These mandatory courses represent an equivalent of 9 ECTS.

(ii) Courses to be selected from a predefined list


The remaining 15 ECTS have to be selected from the following list of courses. This list of proposed courses will be updated at the beginning of each semester to reflect changes in the teaching offer. Each student will choose an appropriate subset from this list after consulting with his/her supervisor and the Dean of Studies.

This selection will be based on several criteria. First of all, some courses will be defined based on the requirements of the individual PhD topic to provide the necessary technical/scientific background (hard skills). From a more general perspective, particular attention will be paid to a thorough acquaintance with scientific working methods already at an early stage of the doctoral program. Besides that, students will be encouraged to participate in special lectures and seminars if improvement of presentation and/or writing skills is required. Depending on individual deficiencies, the supervisors will recommend particular courses from the TU soft skills list.

Soft Skills and Research Seminars

Foundation

Middleware

Robotics

Computational Science

Web Science and Social Computing

Interorganizational Systems and Modeling

Qualifying Exam

Students' progress on their research activities is evaluated through the qualifying exam. The qualifying exam is essentially the PhD proposal for an innovative research contribution with a clear and detailed description of the research directions and objectives as well as the methodological approach (approx. 10 pages). It includes:


  • Problem description
  • Expected result
  • Methodological approach
  • State of the art (incl. references)
  • Work plan
  • List of student publications (if already available)

This proposal is first assessed by the advisor, then by key faculty members of the Doctoral College. Students will then be invited to discuss their proposals.