PhD opportunities are open in any of the above areas. Please direct any enquires to bernat@cs.york.ac.uk.
My research is centered on all aspects of Probabilistic
Real-Time Systems. Next generation of real-time systems need to cope with
increased levels of uncertainty and complexity and therefore a probabilistic
view is a suitable view..
The high-level analysis includes the investigation of scheduling algorithms
that provide in addition to a minimum guaranteed level of service an effective
use of available resources in a context in which the scheduling information
is not precisely known (exact arrival times and execution time of tasks).
It also includes the investigation of the scheduling of weakly-hard
real-time systems, systems that can tolerate a bounded degree of missed
deadlines and of Quality of Service issues in uniprocessor and distributed
systems.
The low-level analysis includes the investigation of statistical methodsfor
the determination of accurate execution profiles of programs by means
of measurement and new statistical methods to reason about such nformation
in order to be used effectively into a flexible scheduling context.
Probabilistic Hard Real-Time Systems
Current timing analysis techniques of real-time systems assume absolute knowledge
of the arrival time of tasks and their execution time, as well as assuming
that execution times of programs do not vary. This assumption no longer
holds for the new generation of complex real-time systems where there
is usually a big difference between worst-case and average-case behaviour.
In this systems capturing absolute worst-case information through modelling
is becoming increasingly difficult due to advanced features in modern processors
and where environments are increasingly dynamic.
This research addresses the notion of probabilistic
analysis and guarantees on a system at all levels of abstraction, from low-level
timing analysis issues to flexible schedulability analysis. The target of
the analysis techniques is to provide extremely high levels of confidence
Initial work includes probabilistic Worst-Case Execution
Time analysis pWCET (in collaboration with Antoin Colin and Stefan Petters) under
the NextTTA EU project. See the publications page for papers on pWCET
You can also check the source code of the examples used in the evaluation o pWCET.
Weakly-Hard Real-Time Systems
The strong classification between hard systems where it is compulsory to
meet all deadlines and soft/firm where an unbounded number of deadlines
missed are accepted is too severe. A weakly-hard real-time systems is a
system in which it is possible to specify that a system can tolerate a
bounded number of missed deadlines in a particular window of time. For
example (3 10) means that a system has to meet at lease 3 deadlines
in every 10 consecutive invocations.
Some guaranteed on-line and off-line scheduling approaches for weakly-hard
real-time systems have already been proposed. This research project involves
a detailed and deeper analysis of such scheduling algorithms and its applicability
in wider contexts from QoS servers to network scheduling. Probabilistic issues
and high-levels of guarantees will be also analysed.
Flexible scheduling
The success of the new generation of real-time systems relies on the ability
to provide guaranteed levels of service while at the same time being able
to adapt to dynamic and changing environments. Current approaches do not
scale up to the size and computational requirements of next generation of
real-time systems where hundreds or even thousands of computational tasks
can be active in a system at the same time.
This project addresses the issue of providing software architectures that
support flexible real-time systems based on flexible schedulers. Some of
the issues to address include the identification of probabilistic models
of the computation which are scalable, low cost scheduling algorithms
and acceptance tests and how to provide robust and predictable behaviour
in an increasingly unpredictable environment.
Worst-case execution time analysis
The analysis of the timing behaviour of real-time sytems can be basically
done in two ways, either by means of measurement and by analysis techniques
(schedulability and worst-case execution time analysis WCET). The problem
of measurement is that it is unsafe and it is possible that the real worst
case scenario is not covered. On the other hand schedulability and WCET techniques
provide a safe upper bound on the maximum execution time of a program, however
they tend to provide quite pessimistic results.
This research addresses the issue of how can these two set of techniques
be used together to reduce the pessimism of the analytical techniques by
combining evidence from measurement and viceversa, and the usage of new
statistical analysis techniques.
York Robocup
Robocup is an international competition of small autonomous robots that play
football. I am currently setting up the York robocup project to compete in
the small size league, I am specially interested in the application of flexible
scheduling and probabilistic models to the particular requirements of the
Robocup scenario.
This research aims to adapt a set of AI techniques in this real-time scenario.
In particular, the research investigates how machine learning approaches
need to be updated in order to be able to integrate them into a flexible
real-time scheduling framework. A strong emphasis will be put on the applicability
of this research into the York RoboCup Team.
An experimental simulator is available Robosim1.0
|
