https://www.dagstuhl.de/19101
March 3 – 8 , 2019, Dagstuhl Seminar 19101
Analysis, Design, and Control of Predictable Interconnected Systems
Organizers
Kunal Agrawal (Washington University – St. Louis, US)
Enrico Bini (University of Turin, IT)
Jens Schmitt (TU Kaiserslautern, DE)
Giovanni Stea (University of Pisa, IT)
For support, please contact
Susanne Bach-Bernhard for administrative matters
Michael Gerke for scientific matters
Documents
Dagstuhl Seminar Schedule (Upload here)
(Use seminar number and access code to log in)
Motivation
The number and type of applications requiring time-predictable behavior has vastly increased in the recent past and is likely to increase even more in the near future, when Industry 4.0, cyber-physical systems, smart grids, and intelligent transportation systems see massive deployment. These systems have a common characteristic: they combine computation and communication. Time predictability in both domains was studied in the past, e.g., using Real-Time Scheduling (RTS) and Network Calculus (NC), respectively. What is required to enable a safe deployment of the above, and has been missing so far, are methods to ensure time predictability in distributed systems where both computation and communication are involved, such that predictability at the application level is achieved via a holistic view of the two. Virtualization (of both computation resources and network functions) adds another challenge for the design of predictable interconnected systems.
This Dagstuhl Seminar aims to bring together researchers to address the following question: how can we design interconnected, geographically distributed systems whose performance is predictable while subject to uncertain inputs? Despite having similar high-level goals, the analytical techniques developed within different research areas are quite different, and so is the terminology used. Hence, a first goal of this seminar is to establish a common ground and lexicon, so as to foster long-lasting collaborations towards solving the new interdisciplinary problems. Researchers will exchange ideas, open problems, analytical and experimental techniques, etc.
A second, equally important goal is to discuss open problems requiring a joint effort, leveraging the presence of scientists from both pure and applied research fields. The interaction with neighboring research areas (such as dataflow, distributed computing, cloud/edge computing, control theory) will be sought.
Following is a (non-exhaustive) sample of research problems that we would like to discuss at the seminar. Invitees are welcome (and will be asked) to integrate this list with their own suggestions.
- Can control methods be applied to NC to improve the bounds in presence of disturbances?
- Can NC be applied to RTS problems? In particular, if the underlying hardware is a parallel or distributed machine, can NC techniques be used to provide performance bounds to such interconnected systems?
- Multipath forwarding becomes more common in networks. What is the implication on performance guarantees? What is the optimal partitioning of traffic flows given several concurrent paths? Can we establish a synergy between the methods for multipath forwarding over networks (being addressed by NC) and multiprocessor scheduling problems (being addressed by RTS)? Can we establish a synergy between the methods for partitioning of a flow among many paths and the partitioning of a DAG application over a multiprocessor?
- Data flow applications are parallel applications that use the abstraction of a graph to describe computational and communication demands. Is NC applicable to analyze dataflow applications on a parallel or distributed machine?
- Can NC be extended to analyze communication delays due to distributed memory or cache effects on parallel and distributed machines?
- Can traversal-time guarantees be given for virtualized network elements? If so, how do they relate to real-time guarantees of the underlying operating system?
- How can we approach system design problems (e.g., compute the minimum communication bandwidth or computational power) having worst-case timing guarantees as a constraint? What types of problems arise in this case, and what are the best tools to solve them?
License
Creative Commons BY 3.0 DE
Kunal Agrawal, Enrico Bini, Jens Schmitt, and Giovanni Stea
Classification
- Networks
- Optimization / Scheduling
Keywords
- Real-time systems
- Network calculus
- Distributed resource management