Inhalt des Dokuments
MA: Marvis - Hybrid Co-Simulation Testbed for Distributed Applications
- A hybrid co-simulation testbed integrates IT resources and domain simulators.
- © Jossekin Beilharz
Today, there is an increasing number of distributed systems that play a role on our daily lives. Some of them are mostly entertainment, but others are absolutely critical as they are used for the control of public infrastructures like energy grids, transportation systems, and water networks. For such distributed software systems, there are usually specific dependability requirements and testing them under a wide variety of possible real-world conditions is a requirement for using them in practice.
While testing centralized and distributed cloud systems is fairly well understood, there are many specific challenges when implementing critical distributed systems for the use cases of the IoT and edge/fog computing environments: errors that depend on the particular hardware used, the need for realistic power consumption predictions, as well as the large variances in network latencies and available bandwidths are some examples of important challenges in this domain. A virtual testbed for widely geo-distributed nodes can support the development of dependable distributed applications. Such a testbed should be hybrid, so it should integrate virtual and physical nodes, to show the effects of specific hardware and power consumption in the real world, while using virtual nodes to allow the evaluation of distributed applications on a larger scale than otherwise feasible. Furthermore, a testbed for infrastructure control use cases should integrate infrastructure simulations (like a traffic or water network simulator) to make the communication patterns of the applications more realistic. Furthermore, experiments should be reproducible, frameworks high-level and user-friendly, and testing resource-efficient.
Project Marvis is a joint effort of the TU Berlin and the Hasso Plattner Institute, University of Potsdam, to efficiently test the behavior of distributed applications across a variety of diverse computing infrastructure scenarios. Previous work exists in the form of two frameworks. The framework Hector, developed at the TU Berlin, allows testing distributed processing jobs in hybrid edge/cloud environments. The framework cohydra, developed at the HPI, provides higher level abstractions and also uses a co-simulation approach to virtually place a distributed application into different kinds of environments. To combine these approaches researchers from both universities work together in the project Marvis to create a comprehensive framework for efficiently testing the dependability of distributed systems in IoT and edge/fog environments.
We offer a variety of master theses topics in this area. Possible research areas are:
- Distributed and scaled testing and simulation
- Model development for and evaluation of network technologies
- Scheduling of testing tasks across computing resources
- Automated deployment of distributed applications
- Co-simulation technologies and standards
- Monitoring of distributed applications
- Techniques for fault injection
All theses will entail designing a general method, implementing a prototype in the context of the existing open source systems, and empirically evaluating the prototype with adequate experiments.
This is a joint proposition together with the Operating Systems and Middleware (OSM)  group at Hasso-Plattner-Institute, where the contact for this is Jossekin Beilharz .
If this sounds interesting to you, please send us an email with a bit of background information about you, so we can quickly identify a fitting topic and scope together.