epoc embedded performance analysis for organic computing
play

EPOC Embedded Performance Analysis for Organic Computing Steffen - PowerPoint PPT Presentation

Platzhalter fr Bild, Bild auf Titelfolie hinter das Logo einsetzen EPOC Embedded Performance Analysis for Organic Computing Steffen Stein , Moritz Neukirchner, Rolf Ernst, 08.10.2010 Outline Project Reminder Selected Topics of


  1. Platzhalter für Bild, Bild auf Titelfolie hinter das Logo einsetzen EPOC – Embedded Performance Analysis for Organic Computing Steffen Stein , Moritz Neukirchner, Rolf Ernst, 08.10.2010

  2. Outline • Project Reminder • Selected Topics of this year‘s work • Dissemination • Conclusion October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 2

  3. Reminder: Demonstrator October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 3

  4. Reminder - Contracting application 1 application n  Software has to declare its behavior , request assert contract requirements and constraints  The system verifies itself based on contracting interface software description and platform capabilities  The system supervises adherence of the RTE software to its descriptions contracting interface ⇒ The system protects itself from infeasible configuration specification contract changes CPU CPU October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 4

  5. Architectural Approach Application code & model Contract interface Update contract Controller RTE code model Controller Task Task Controller Model-based Contract Task Task Optimization model Negotiation contract Contract configuration metrics Enforcement metrics contract model execution violation metrics Contract Model-based Repository Analysis Observer Contract contract Supervision Model domain Execution domain Observer October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 5

  6. Open Problems Testcase Generation Mode Application code & model Change Protocol Contract interface Update contract Controller RTE code Suitable Heuristics model Task Task Model-based Contract Task Task Optimization model Negotiation contract Contract configuration metrics Enforcement metrics contract model Bounded Time execution violation Algorithm metrics Contract Model-based Isolation Repository Analysis Contract contract Supervision Model domain Execution domain Convergence Detection October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 6

  7. 1 2 Platzhalter für Bild, Bild auf Titelfolie hinter das Logo einsetzen Analysis in Bounded Time ECRTS'2010: "A polynomial time algorithm for computing response time bounds in static priority scheduling with convex event models"

  8. Performance Analysis in Bounded Time • Based on Proof of Convergence • Analysis States form Discrete Chain • Bound on Reachable States  Number of Analysis Steps Bounded • Bound Computational Time needed for each step • Last Year: Properties of Busy-Window Algorithms • This Year: Novel Polynomial Time Algorithm October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 8

  9. Schedulability in Polynomial Time Workload Bound 1 1 Inaccuracy    C C     2 t C 1   d d Bound 2 C JC C      t t C ( 1 ) T T T Δ t d T October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 9

  10. Novel Algorithm 1. Aggregate Workload Bounds + + = 3. Invert – Time needed for Service 2. Spare Capacity/Time - = 4. Response-Time = Time Needed for Service – Minimum Interarrival Time Maximum through Candidates - = October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 10

  11. Results Prior Art This Algorithm October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 11

  12. Platzhalter für Bild, Bild auf Titelfolie hinter das Logo einsetzen Mode Change Protocols SORT'2010: "Consistency Challenges in Self-Organizing Distributed Hard Real-Time Systems"

  13. The Problem of Dependant Tasks (Echo Effect) Echo Effect described by Henia et al. (RTSS 2007) CPU 2 Consider Removal of Red Task 1. WCRT of lower priority tasks decrease 2. Output Jitter of T2 decreases 3. Output Jittter of T1 decreases 4. WCRT of T0 decreases J W T0 T1 J Potential Cyclic Dependencies W T2 Settling time Unknown CPU 3 W CPU 1 October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 13

  14. Mode Changes – Identified Open Challenges Before During After Global Asynchronous/ Locally Asynchronous (GALA)  Tasks may be started at any time  Easy to Implement  Load Peaks Globally Synchronous/ Locally Synchronous (GSLS) Before During After  Wait for Effects on System Timing to cease before starting tasks  No Load Peaks  Requires state Synchronization Globally Asynchronous/ Locally Synchronous (GALS)  Wait for Effects on Local Timing to cease before Before During After starting tasks  Tradeoff between GALA/GALS?

  15. Protocol Analysis: GALA WCRT CPU 2 Settle? Protocol WCRT trans WCRT steady,2 CPU 3 WCRT steady,1 CPU 1 t sw,start t sw,end t t steady October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 15

  16. Solution: ModeWaves See DFG Project "Szenarien in vernetzten Multicore-Systemen – Analyse und Optimierung" (ER 168/23-1)

  17. 1 0,8 DCS DCS DCS Platzhalter für Bild, Bild auf Titelfolie hinter das Logo einsetzen 0,6 0,4 0,2 DPA DPA DPA 0 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 1 Heuristic Distributed Self-Configuration

  18. Heuristic Distributed Self-Configuration Objective: Find a priority assignment on all processors and busses, such that all end-to-end path latency constraints are satisfied. Constraint: No resource has a complete view of the system model. τ 4 τ 3 τ 4 τ 3 Comm. τ 2 τ 1 τ 2 τ 1 Proc. 2 Proc. 1 Proc. 2 Proc. 1 Priorities? Priorities? October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 18

  19. Heuristic Distributed Self-Configuration Approach: Extend each instance of the Distributed Performance Analysis Algorithm with an instance of a heuristic distributed constraint solver Distributed Constraint Solver (DCS) DCS DCS DCS instance instance instance Worst-case New Priority Timing Assignment Information DPA DPA DPA instance instance instance Distributed Performance Analysis (DPA) October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 19

  20. Heuristic Distributed Self-Configuration • Heuristic based on gradient search • Metric based on path constraint violation and task worst-case response time ⇒ No explicit communication required between DCS instances • Oscillations in exploration process can occur in distributed concurrent execution τ 4 τ 3 τ 2 τ 1 Proc. 2 Proc. 1 October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 20

  21. Heuristic Distributed Self-Configuration • Analysis shows that oscillation is indeed an issue Search Space Coverage histogram 1 0,9 0,8 0,7 Frequency 0,6 0,5 0,4 0,3 0,2 0,1 0 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 1 Unique Configurations/Analyzed Configurations • Introduction of a lazy threshold that prevents execution of single DCS instances to break loop of oscillations October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 21

  22. Heuristic Distributed Self-Configuration • Comparison to a design-time genetic algorithm as benchmark Number of solvable testcases solved 0% 25% 50% 75% 100% Small Systems Large Systems Search Space Coverage Histogram Runtime relative to genetic algorithm 0,5 1 0,9 avg. 7.35 0,4 0,8 0,7 Frequency Frequency 0,3 0,6 0,5 0,4 0,2 0,3 0,2 0,1 0,1 0 0 1 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 -1000 -100 -10 - equal 10 100 1000 1 1 Unique Configurations/Analyzed Configurations Improvement w.r.t. # of analysis runs vs. GA October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 22

  23. Platzhalter für Bild, Bild auf Titelfolie hinter das Logo einsetzen System Models For Free: SMFF Available soon on sourceforge.net

  24. System Models For Free: SMFF • When evaluating heuristic optimization algorithms a large number of testcases is required. • Testcase system models are not available for distributed realtime systems • SMFF: Tool for pseudo-random generation of testcase system models Structural Definition Real-Time Property Definition Platform Model Generation Application Model Generation Constraint Definition Assignment of Timing Application Mapping Constraints Assignment of Scheduling Assignment of Timing Parameters Properties October 8, 2010 | Steffen Stein | Runtime Verification and Management of Real-time Applications | Folie 24

Recommend


More recommend