Towards a Platform for Widespread Embedded Intelligence - ERCIM

Formally Bounded Embedded Systems
by Kevin Hammond
The EU-funded EmBounded project aims to develop new techniques for providing
formal guarantees on resource usage bounds. It is a collaborative project between
AbsInt GmbH, a high-tech company specialising in high integrity real-time systems,
and academic partners in the UK, France and Germany.
A major difficulty with constructing
embedded systems is that the available
system resources (processor, computer
memory, power) are necessarily
restricted due to cost and other considerations.
If it were possible to determine
strong bounds on the use of such
resources, then potentially there would
be significant benefits in terms of manufacturing
cost, reliability and performance.
However, determining such
bounds automatically is a difficult
problem and doing so manually is
becoming impracticable as embedded
software increases in complexity.
The EmBounded project, led by Dr
Kevin Hammond at the University of St
Andrews, aims to research this problem.
Our vision is one where certificates of
the bounds on resource usage can be
obtained from a source program, through
automatic program analysis, independently
of the usual software compilation
process. These certificates may then be
verified using formal proof techniques
based on a program logic that captures
the meaning and behaviour of programs.
The EmBounded project builds on
European strengths in programming language
design, program analysis and
embedded applications. Our work is
embound, v.
poet. arch.
trans. To set bounds to; to confine, contain, hem in.
Hence embounded ppl. a.
1595 SHAKESPEARE: The Life and Death of King John IV. iii 137
That sweete breath which was embounded in this beauteous clay.
The 'EmBounded' vision.
SPECIAL THEME: Embedded Intelligence
based around a new domain-specific
programming language notation, Hume,
that aims to find an optimum point
between good programming abstraction
and the ability to derive good cost information.
Hume embeds a rule-based
notation for sequential computations
within a finite-state-machine-based notation
for concurrency control. The process
notation is designed to be easily
tractable to standard analyses such as
model-checking for deadlock detection;
while the rule-based notation is designed
so that we can easily expose information
about cost. Hume is designed to be practical
as well as having interesting language
properties. Prototype implementations
of Hume produced at Heriot-Watt
University can run in less than 30KB on
a standalone Renesas M32C development
board, with minimal dynamic
memory requirements,
depending on the end-application.
To date, project researchers at
Ludwig-Maximilians-
Universität, München, and St
Andrews University have developed
new theoretical models of
bounded resource usage for
dynamic memory and time metrics
based on Hume program
source. They have related these
formally to the underlying
machine operations using an
abstract-machine approach. We
are now constructing automatic
analyses based on these models
that will expose resource constraints
and other program properties
in our program logic. This
will give a good model of program
cost based on the program
structure.
In order to obtain accurate information
about time usage on a
specific computer architecture,
we are using the aiT tool developed
by AbsInt GmbH to provide
precise and guaranteed timing
costs for fragments of machine
code programs. This low-level
ERCIM News No. 67, October 2006 45