1 / 21

Presenter: Lin Huang Lin Huang and Qiang Xu CU hk RE liable computing laboratory (CURE)

Energy-Efficient Task Allocation and Scheduling for Multi-Mode MPSoCs under Lifetime Reliability Constraint. Presenter: Lin Huang Lin Huang and Qiang Xu CU hk RE liable computing laboratory (CURE) The Chinese University of Hong Kong. Lifetime Reliability Becomes A Serious Concern. Infant

Download Presentation

Presenter: Lin Huang Lin Huang and Qiang Xu CU hk RE liable computing laboratory (CURE)

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Energy-Efficient Task Allocation and Scheduling for Multi-Mode MPSoCs under Lifetime Reliability Constraint Presenter: Lin Huang Lin Huang and Qiang Xu CUhk REliable computing laboratory (CURE) The Chinese University of Hong Kong

  2. Lifetime Reliability Becomes A Serious Concern Infant mortality Useful life Wearout 90nm 130nm 180nm Failure mechanisms Electromigration NBTI TDDB Failure rate Time [T. M. Mak] < 7 year ~ 7 year ~ 10 year

  3. Task Allocation and Scheduling • Multiprocessor system-on-chip (MPSoC) platform • Energy-efficient task allocation and scheduling • Multi-mode MPSoC • For instance, a modern smart phone can serve as • MP3 player • Game console • Digital camera • Video decoder • GPS navigation • … MPSoC platform It is essential to explicitly consider lifetime reliability issue in energy-efficient embedded system designs

  4. Problem Formulation • Given • and the joint probability density function • Determine a task schedule for each execution mode such that • The expected energy consumption is minimized • The performance and reliability constraints are met MPSoC platform Task graphs

  5. Prior Work • [Huang&Xu DATE’09] explicitly takes the lifetime reliability into account during task allocation and scheduling • Energy consumption issues are not considered • Focus on single execution mode only • Maximize the expected service life under performance constraint

  6. Agenda • Introduction and motivation • Problem formulation • Proposed algorithm for multi-mode embedded systems • Task schedule generation for each execution mode • Multi-mode combination • Experimental results • Conclusion

  7. Feasible Solution Set Systemwide reliability threshold G F D E A Energy Consumption B C O Reliability

  8. X Y Feasible Solution Set w X – all the task schedules Y – feasible solution set u v Internal stability Given two solutions u,v ∈ Y, if u consumes more energy than v, it must have higher lifetime reliability at the target service life, and vice versa External stability For any solution w ∈ X \ Y, there exists at least one solution u ∈ Y such that u consumes less energy and have higher lifetime reliability than w

  9. Feasible Solution Set Identification • Static strategy Systemwide reliability threshold G F Domain IV D E A Energy Consumption Domain I B C O Domain II Domain III Reliability Pareto optimal solution set Feasible solution set = {O,D,E} = {O} = {O,D} The reached schedule is a feasible solution iff it is in the first or third domain of all elements in feasible solution set

  10. Feasible Solution Set Identification • Dynamic strategy • Avoid heavy memory overhead • Every newfound solution is processed according to … • Rule 1 If the new solution is in domain I or III of ALL elements in set , it should be included into • Rule 2 If the new solution is in domain II of ANY solution X in , we include the new solution into and eliminate X from • Rule 3 If the new solution is in domain IV of ANY solution in , we ignore the new solution Systemwide Reliability Threshold new solution original updated G F {} C {C} {C} O {O} D E {O,E} E {O} A Energy Consumption {O,E} {O,E,D} D {O,E,D} {O,E,D} F {O,E,D} B {O,E,D} B C O {O,E,D} A {O,E,D} {O,E,D} G {O,E,D} Reliability

  11. Searching Procedure for a Single Mode • Modified simulated annealing • Classic SA keeps the current solution and the best one so far • Modified SA keeps a possible solution set • Static strategy • Dynamic strategy • Solution representation • (schedule order sequence; resource binding sequence) • Example: (0, 2, 1; P1, P1, P2) • Cost function

  12. Searching Procedure for a Single Mode • Solution representation • (schedule order sequence; resource binding sequence) • Example: (0, 2, 1; P1, P1, P2) • Cost function (0,2,1;P1,P1,P2;.6Vdd,.8Vdd,Vdd) Resource binding Solution space (0,2,1;P1,P1,P2) DVS Schedule order

  13. Searching Procedure for a Single Mode • Solution representation • (schedule order sequence; resource binding sequence) • Example: (0, 2, 1; P1, P1, P2) • Cost function Task schedule Deadline P1 0 1 P2 2

  14. min st. or Multi-Mode Combination • Optimization problem • Joint probability density function

  15. Experimental Setup • Task graphs are generated by TGFF • The power consumption values are randomly generated, while the range is set according to state-of-the-art technology • Well-studied electromigration failure model • The proposed model is applicable for the combination of multiple failure mechanisms • Baseline solution • We first build a schedule to shorten schedule length and reduce energy consumption with list scheduling • We then attempt to meet the reliability constraint in a greedy manner • Single mode method

  16. Case Study • Task graphs • Occurrence probability • (a) 0.3 (b) 0.3 (c) 0.4 • Reliability constraint • The system reliability at 10 years is no less than 36.8%

  17. Case Study

  18. Sensitivity Analysis 32% 39% 49% 17.27 12% 42% 27%

  19. Extensive Results 29% 40% 49% 26-28% energy reduction

  20. Conclusion • Lifetime reliability has become a serious concern nowadays • Today’s complex embedded system typically have multiple execution modes • We propose novel task allocation and scheduling algorithm • Objective: to minimize the expected energy consumption under performance and reliability constraints • We first identify a set of “good” schedules for each execution mode • We then introduce novel techniques to obtain an optimal combination • The effectiveness has been demonstrated by experiments

  21. Energy-Efficient Task Allocation and Scheduling for Multi-Mode MPSoCs under Lifetime Reliability Constraint Thank you for your attention !

More Related