Design Patterns for Self-Adaptive RTE Systems Specification
Table 1
Adaptation concepts correspondence between the proposed patterns and the object tracking application.
Patterns feature
Case study instance
Context elements
Three magnitudes:
(i) application QoS (the tracking error);
(ii) execution time;
(iii) power consumption
Sensors
Four observers:
(i) a task (T10);
(ii) SW timer of RTOS;
(iii) battery gauge component;
(iv) observer estimator
Adaptation period
Configuration period ; that is, a new configuration is evaluated after each application iteration
Monitor + Analyzer
Local configurationmanager (LCM)
Monitoring Thresholds, irregular status
QoS reference (the tracking maximum error) is set to 10% and reduced to 2% within the critical area to guarantee good reactivity. Task T10 provides the LCM with the application QoS metric (error between prediction and object position):
(i) a value close to 0 but lower than the reference (10%) means a very high tracking quality →it can be relaxed by reducing the application speed;
(ii) a value higher than the reference → the application rate must be increased with a faster configuration.
Adaptation request
User requirements (e.g., QoS, power, and performance references) and magnitudes’ values
Decisionmaker
(i) LCM for application specific algorithmic configuration decisions
(ii) Global ConfigurationManager (GCM) for global architectural configuration decisions
Decision strategy
(i) LCM: algorithmic configuration selection using LCM rules; lists of configurations and transition rules based on tasks metrics.
(ii) GCM: selection of the closest solution below the user reference (QoS reference). A Borda vote is used in the case of multiple solutions.
Adaptation Plan
Architectural configuration CID
Actor (refinement)
A mask-based CID analysis
AdaptationAction
Four cases result from the CID analysis:
(i) nothing to do;
(ii) requests to HAL to get new I/O information (physical addresses);
(iii) a reconfiguration message sent by a HW task to its legal representative (LR) to activate its SW implementation;
(iv) the LR directly intercepts the new CID and activates the SW task
Effector
(i) HAL
(ii) HW task LR
Changeable elements
Tasks
Assessor
System stability and avoidance of reconfiguration
Performance Metrics
(i) System stability: use of proportional integrator
(ii) Requests to HAL to get new I/O information (PI) regulator (coefficients: ) and a least mean square (LMS) observer (coefficient: )
(ii) Reconfiguration avoidance: use of a minimum delay Tk required to accept the reconfiguration overhead compared to expected benefits
