The steadily increasing performance of embedded systems leads to
a tighter integration of differing software models. The differences
become manifest in especially three distinct issues: Timeliness (hard
real-time, Quality of Service, soft real-time), activation models
(periodic, aperiodic and sporadic) and variation of execution
times. The last item becomes an issues because of data dependend
activation of complex software models and throughput optimized hardware architectures.
Existing run-time systems only focus on partial requirements of their
domain in which they are applied. Thus they are hardly suited to cover
the evolving requirements according to timeliness, activation models
and execution time variation of hybrid systems. In this thesis a robust and
transaction-based run-time system is presented, that fulfills the
stated requirements.
- The transaction-based run-time semantic is based on the optimal
dynamic earliest deadline first (EDF) scheduling and leads to a
realistic real-time analysis.
- The robust extension RTEDF checks the actual computation demand of the
system at run-time. During underload conditions the remaining
slack (through variation of execution times, non occurring blocking
times and variation of event times) is used to enhance
(maximizing) the Quality of Service of QoS/Soft transactions. The
scheduler operates as an EDF scheduler. In overload conditions
however, an a priori defined pattern for quality reduction is
applied. The pattern is verified by the real-time analysis
pre run-time in order to guarantee a robust behavior.
- The specification of the QoS Pattern is effected as metadata in the
software model. The consideration of this metadata leads to a
parallel Meta design flow.
- The partitioning of the overload detection algorithm is done by mapping
the algorithm onto a loosely coupled coprocessor, which leads to a
hardware based separation of concerns. This approach enables the separation of the meta design flow from the functional design flow.
The TEDF run-time system and the RTEDF algorithm have been implemented
on a prototype platform based on a SOPC (System On a Programmable
Chip) with real-time operating system and soft-core coprocessor. The
obtained results are presented as well as the timing, performance and
data demand measurements of the real implementation.
«
The steadily increasing performance of embedded systems leads to
a tighter integration of differing software models. The differences
become manifest in especially three distinct issues: Timeliness (hard
real-time, Quality of Service, soft real-time), activation models
(periodic, aperiodic and sporadic) and variation of execution
times. The last item becomes an issues because of data dependend
activation of complex software models and throughput optimized hardware architectures.
Existi...
»