Continuation proposal: Integrated Control and Scheduling
2000-08-11
Karl-Erik Årzén
Klas Nilsson
Dept of Automatic Control Dept of Computer
Science
Lund University
Lund University
The focus of the ARTES project "Integrated Control and Scheduling" is practical
management of hard real-time demands in embedded software for real-time
control. The project is based on two key ideas. The first idea is to combine
control theory and scheduling theory in such a way that the nominal requirements
on hard deadlines for control systems can be relieved. The approach taken
is based on using dynamic feedback from scheduler to the controllers and
from the controllers to the scheduler. The second, complementary, idea
is to use attribute grammars and incremental semantic analysis to carry
out on-line interactive analysis of worst-case timing of software and generation
of exception handling code for coping with unexpected delays. Combining
these ideas forms a more complete and practical methodology than available
today.
The first half of the project started 980501 and will terminate 001031
(80% activity level). The first half of the project has been very successful,
see e.g., the recent ARTES evaluation by Andy Wellings. A number of results
have been obtained in the two parts of the project:
Feedback Scheduling
and Interactive Execution Time Analysis.
For more details see the
Second Year Project Report. The two involved PhD students, Patrik Persson
and Anton Cervin, have both successfully presented their licentiate theses
during Spring 2000.
The second half of the project will follow the general outline defined
in the original project proposal (enclosed) together with the more detailed
specification below. The collaboration with Prof. Lui Sha will continue,
although he is now affiliated with University of Illinois at Urbana-Champaign
rather than with SEI, CMU. The industrial connection will also remain unchanged,
i.e., with Sigma Exallon Systems and with DDA Consulting.
Budget
We apply for two ARTES PhD-student slots of 600 kSEK/year for 2.5 years
at 80% activity level for Patrik Persson and Anton Cervin starting 001101.
Project plan
The project plan corresponds to the second half of the project plan in
the original proposal. The main activities during 2001 and 2002 will be
refinements of the developed theory and techniques and implementation of
software tools. The main deliverables will be two PhD theses together with
the software tools. During this second half of the full project more emphasis
will be put on integrating the two subprojects and the software tools.
The integration will be done in two ways. First, the Skånerost interactive
timing analysis environment will be interfaced with the co-simulation toolbox.
Second, a Java-based real-time kernel with support for feedback scheduling
will be developed.
The planned research topics are the following:
Feedback Scheduling:
-
Real-time co-simulation of distributed systems. The real-time control
simulator will be further extended. Special emphasis will be put on support
for simulation of distributed real-time control systems.
-
Control performance with feedback scheduling. In a feedback scheduling
based control system the scheduler dynamically adjusts the control task
attributes, e.g., the task periods. We will study how this influences the
control performance and stability.
-
Dynamic tradeoffs between execution time and control performance.
Certain common control technologies are based on iterative on-line
optimization in every sample. One example of this is model-predictive control.
This makes it possible for a feedback scheduler to assign varying execution
time quotas to the different controller tasks. We will study this and how
it influences the control performance.
-
Feedback scheduling implementation. Implementation of feedback scheduling-based
control systems will be performed. The main environment will be a Java-based
kernel.
A study visit for Anton Cervin at University of California, Berkeley is
planned for early Spring 2001.
Interactive Execution
Time Analysis:
-
Practical evaluation. Our analysis techniques and the Skånerost
environment should be evaluated for realistic code examples. This requires
support for a fuller Java subset, and also collaboration with the IVM (Infinitesimal
Virtual Machine) project in Lund, where a Java virtual machine with small
memory footprint is being developed. It further requires the development
of a timing model of the hardware platform in question.
-
Further development of live memory analysis. Our present techniques
for live memory analysis support a number of types of data structures.
To make these techniques more generally applicable, the algorithms should
be extended to handle more general data structures.
-
Using information from an object-oriented design in scheduling analysis.
The scheduling analysis requires some information about the structure of
the code, such as which parts are to be executed as sequential or periodic
processes. Such information is typically expressed in an object-oriented
design. We plan to investigate how this design information can be automatically
exploited by scheduling analysis, and how an API (Application Programmer's
Interface) can support the expression of this information.
-
Expressing timing information in the byte code. By generating platform
independent WCET information along with the generated byte code, we can
facilitate an on-line platform-specific WCET analysis in the virtual machine.
This also allows quality-of-service (QoS) applications to monitor their
execution time requirements. Steps have been taken for cooperation with
the Computer Science department at the University of York on this subject.
Karl-Erik Arzen and Klas
Nilsson
Last modified: 2000-08-09
|
