Analysis, Design, and Control of Predictable Interconnected Systems (Dagstuhl Seminar 19101)

Abstract

We call "Interconnected Systems" any collection of systems
distributed over a metric space whose behavior is influenced by its
neighborhood. Examples of interconnected systems exist at very
different scales: different cores over the same silicon, different
sub-systems in vehicles, communicating nodes over either a physical
(e.g., optical) network, or - more recently - virtualized
network. Examples also exist in contexts which are not related to
computing or communication. Smart Grids (of energy production,
distribution, and consumption) and Intelligent Transportation Systems
are just two notable examples.  The common characteristic among all
these examples is the presence of a spatially distributed demand of
resources (energy, computing, communication bandwidth, etc.)  which
needs to be matched with a spatially distributed supply.  Often times
demands and availability of resources of different types (e.g.,
computing and link bandwidth in virtualized network environments) need
to be matched simultaneously.

Time predictability is a key requirement for above systems.
Despite this, the strong market pressure has often led to ``quick and
dirty'' best-effort solutions, which make it extremely challenging to
predict the behavior of such systems.

Research communities have developed formal theories for predictability
which are specialized to each application domain or type of resource
(e.g., schedulability analysis for real-time systems or network
calculus for communication systems).  However, the emerging
application domains (virtualized networks, cyber-physical systems,
etc.)  clearly require a unified, holistic approach.  By leveraging
the expertise, vision and interactions of scientists that have
addressed predictability in different areas, the proposed seminar aims
at constructing a common ground for the theory supporting the
analysis, the design, and the control of predictable interconnected
systems.