Timing Analysis Seminar 2011-05-12
T-106.5840, organized by Vesa Hirvisalo, Embedded Systems Group,
Department of Computer Science and Engineering, Aalto University,
Challenges of analysis
Niklas Holsti, Tidorum Ltd
Why WCET analysis does not work now
and will not work in the future
I will talk about the difficult or currently unsolved problems in WCET analysis,
both for the static-analysis tools and the measurement-based tools (which of
course have to some extent different problems), why we can expect these problems
to remain hard, and also can expect that new hard problems will appear. Of
course this talk is intended to be provocative rather than blackly pessimistic,
so I hope to give some pointers to possible solutions or work-arounds, as future
research directions. But I do not promise to be optimistic...
WCET analysis is practical now only for relatively simple
programs on relatively simple microcontrollers.
However, "simple" does not imply "small".
Current applications are in highly critical systems:
aerospace, automotive, nuclear.
Static analysis of worst-case processor behaviour seems
hopeless for high-end, general processors. Open question:
are predictable but powerful processors possible?
Measurement-based analysis is unreliable for the same reasons,
but is more reliable than end-to-end measurements.
Control-flow analysis has promising problems to work on.
Increased use of static analysis for bug-finding and the
like may push programs to be more analysable.
Existence of WCET tools pushes the state of the art,
which may make WCET analysis required for critical SW.