1. Lab Meeting Performance Analysis of Distributed Embedded Systems Lothar Thiele and Ernesto Wandeler Presented by Alex Cameron 17 th August, 2012

2. One cause for end-to-end timing constraints is the fact that embedded systems are frequently connected to a physical environment through sensors and actuators. Typically, embedded systems are reactive systems that are in continuous interaction with their environment and they must execute at a pace determined by that environment. A real-time constraint is called hard, if not meeting that constraint could result in a catastrophic failure of the system, and it is called soft otherwise. As a consequence, time-predictability in the strong sense can not be guaranteed using statistical arguments. Performance Analysis using Network Calculus presents an elegant methodology for offering performance guarantees in deterministic queuing systems. “It is the purpose of performance analysis to determine the timing and memory properties of such systems.” !! Failures of embedded systems very often relate to timing anomalies that happen infrequently and therefore, are almost impossible to discover by simulation….

3. Consider an Embedded Real-Time System Comprising Two Applications 3 Threat Sensor CPUDSP Actuator Detection Data Launch BCET WCET t A1 A2 P1, P2 P3 P5,P6 P4 Interference between bus and Apps means competing BCET and WCET

4. Consider the System when Network Enabled 4 Threat Sensor CPUDSP Actuator Detection Data Launch A1 A2 Service 1Service 2 BCET WCET t

5. Example Arrival Patterns Basic arrival functions for a set of arrival patterns that can be derived using Patterns (e.g. sensor), Trace (measure) or Specifications (Data Sheets)

6. Effect of Deadline Variance for given Event Arrival Rate Figure 12: Graph shows the normalised rate of missed deadlines for the LLF scheduling algorithm plotted against the ratio of the deadline to the Poisson arrival mean for a range of variances (jitter) in the deadline. The graphs have been smoothed but were based on a sample of 200 arrivals for each measured point on the curve. These results are for a complex workflow comprising five services running on a single CPU.

7. One Approach is to manage the Event Arrival pattern EMIF = Event Model Interfaces perform type conversions between arrival patterns. EAF = Event Adaption Functions: Making the systems analysable, e.g. adding buffers etc. when EMIF is not present Bound to the implementation

8. The Concept The guarantee – either regulator or leaky bucket

9. Performance Network Approach Resource Modelling: In comparison to functional validation, we need to model the resource capabilities and how they are changed by the workload of tasks or communication. Therefore, contrary to the approaches described before, we will model the resources explicitly as ‘first class citizens'. Abstractions Arrival Curves Abstractions Resource Modelling Service Functions

10. Primary Difficulty - Modelling the Workload WCET and BCET: The simplest possibility is to assume that each event of an event stream triggers the same task and that this task has a given worst case and best case execution time determined by other methods. Application Modelling: Take into account the characteristics of the application, e.g. (a) distinguishing between different event types each one triggering a different task and modelling various WCET (or BCET). This way, one can model correlations in event streams. Each incoming event, a subtask generates the associated workload and the program branches to one of its successors. Trace: As in the case of arrival curves, we can use a given trace and re- port the workloads associated to each event, e,g, by simulation. Based on this information, we can easily compute the upper and lower envelope.

11. The Outcome Figure 12: Representation of the delay and accumulated buffer space computation in a performance network.

12. DISCUSSIONS

13. BACKUP SLIDES

14. 14 Distributed Real-Time Event Driven Service Oriented Architectural Implementation Deterministic Measures Performance Measures Threat Sensor Actuator Detection Launch VMSA Event Synchronised Petri Nets (CSPN) Model Architecture to Petri Net Mapping Random Events Decision Variables Random Variate Generator Predictive Measures Discrete Event Simulator Response Applicability

15. And What about the Event Arrival Patterns? periodic T Periodic with Jitter TJ Admissible occurrence of event