1 / 45

Energy Aware Real-Time Systems

Energy Aware Real-Time Systems. G. Sudha Anil Kumar Real Time Computing and Networking Laboratory Department of Electrical and Computer Engineering Iowa State University CprE 545 class presentation. Real-Time System: Characteristics. Real-Time Guarantees Meeting deadlines Fault Tolerance

rian
Télécharger la présentation

Energy Aware Real-Time Systems

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 Aware Real-Time Systems G. Sudha Anil Kumar Real Time Computing and Networking Laboratory Department of Electrical and Computer Engineering Iowa State University CprE 545 class presentation

  2. Real-Time System: Characteristics • Real-Time Guarantees • Meeting deadlines • Fault Tolerance • Tolerating faults • Quality of Service • Acceptable quality of service • Energy Consumption • Minimize overall energy consumption anil@iastate.edu

  3. Real Time system Energy Quality Fault tolerance anil@iastate.edu

  4. Fault Tolerance vs. Quality • Imprecise Computation technique • Trading off quality for fault tolerance • (m, k)-firm deadline task model • Trading off quality for scheduling flexibility anil@iastate.edu

  5. Imprecise Computation (IC) Normal Task Ci Mandatory Optional Mi Oi Imprecise Computation task anil@iastate.edu

  6. IC: Relevant Applications • Image Processing: Fuzzy image in time are better than too late perfect image • Tracking: Rough estimate of target location in time is better than too late accurate location data. anil@iastate.edu

  7. (m, k)-firm deadline tasks Task (T): C = 1; P = 2; 0 2 4 6 8 10 Time M = 2; K = 3; 0 2 4 6 8 10 Time anil@iastate.edu

  8. (m, k): Relevant Applications • Radar tracking: A few well spaced deadlines can be tolerated • Automobile control, multi-media streaming, etc.. anil@iastate.edu

  9. Real Time system Energy Quality Fault tolerance anil@iastate.edu

  10. Energy vs. Quality • Conflicting Design Objectives • Energy savings • Quality of Service anil@iastate.edu

  11. Organization of the presentation • Energy issues in RT-Embedded systems • Dynamic Voltage Scaling (DVS) • RT-DVS schemes • Energy aware RT-DVS for IC and (m, k) tasks anil@iastate.edu

  12. Energy consumption in RT-ES • Energy consumption is an important issue in RT-embedded systems like: • Laptops, PDAs. • Digital camcorders, cellular phones • portable medical devices. anil@iastate.edu

  13. Important Facts (1) • The peak computing rate needed is much higher than the average throughput that must be sustained • High performance is needed only for a small fraction of time, while for the rest of time, a low-performance, a low-power processor would suffice anil@iastate.edu

  14. Workload Profile Work load Peak Computing Rate is needed Average rate would suffice Time anil@iastate.edu

  15. Important Facts (2) CMOS based processors Varying voltage and frequency we can reduce the energy consumption Power (P) αV2 .f V αf Energy (Ei) αcci .f2 anil@iastate.edu

  16. Variable Voltage Processors • Modern processors operate at multiple frequency (and voltage) levels. • Crusoe Processor: Transmeta Corporation • PowerNow! Technology: AMD • Intel XScale: Intel • Higher the frequency level higher the energy consumption anil@iastate.edu

  17. Dynamic Voltage Scaling (DVS) • DVS scales the operating voltage of the processor along with the frequency. • Since the energy consumption is proportional to V2 , DVS can potentially provide a very large energy savings. anil@iastate.edu

  18. DVS-example • Consider a task with a computation time 20 units. • Energy of Ti without DVS • E1 = K * 20 * F2. • Energy of Ti with DVS • E2 = K * 20 * (F/2)2. • Clearly, E2 = (E1)/4. anil@iastate.edu

  19. Energy-Time Tradeoffs 60 40 Energy Savings 20 10 Time anil@iastate.edu

  20. Energy aware RT-scheduling: objectives • Minimizing energy consumption • Maximize the quality • Meeting the deadlines anil@iastate.edu

  21. Energy aware RTS Techniques • OS Level Energy Management • Inter-task DVS • Compiler Level Energy Management • Intra-task DVS anil@iastate.edu

  22. Intra vs. Inter-task DVS • Inter-task DVS scheme: Voltage scheduling is done on a task by task basis. T3 T1 T2 • Intra-task DVS scheme: Voltage scheduling is done within a task boundary. • Each task is modeled as a control flow graph. Voltage scheduling points T3 T1… …T1 T2… …T2 anil@iastate.edu

  23. Quality Oi Energy Aware RT-Scheduling IC tasks System Model • OS level DVS • Inter-task DVS Each periodic task is specified by : Ci, Pi, Mi, Oi Energy budget per hyper-period: Eb Mi anil@iastate.edu

  24. Energy aware RT-Scheduling of IC tasks [8] • Goal: • To schedule a set of Imprecise Computation tasks • Objective: • maximize the quality • Constraints: • without exceeding the deadlines • Without exceeding the total energy available anil@iastate.edu

  25. Optimal Solution [8] Find the Minimum energy frequencies settings of each task Find the Maximum quality solution With the above frequency settings anil@iastate.edu

  26. Minimum energy frequency settings • Theorem: All tasks will execute at the same frequency in the minimum-energy solution • Due to the concave nature of the energy function • The above theorem is proved using rigorous mathematical tools. • The intuition follows…… anil@iastate.edu

  27. Energy function characteristics anil@iastate.edu

  28. Example • Consider two tasks: • T1 = (3, 12) and T2 = (3,12) +ΔE2 Energy T2 @ f = 0.7 -ΔE1 f = 0.5 T1 @ f = 0.4 Time 0 7.5 12 ΔE1 < ΔE2 anil@iastate.edu

  29. Example: Minimum energy frequency settings • Consider two tasks: • T1 = (3, 12) and T2 = (3,12) Energy f = 0.5 T1 @ f = 0.5 T2 @ f = 0.5 6.0 0 12 Time anil@iastate.edu

  30. Calculating the minimum energy frequency • Given: energy budget, Eb per LCM • We know: power, P = k * f3 • Solve for fop: k * fop3 = Eb / LCM anil@iastate.edu

  31. Reduced Problem [8] • Goal: • To schedule a set of Imprecise Computation tasks • Objective: • maximize the quality • Constraints: • without exceeding the deadline • Without exceeding the total energy available anil@iastate.edu

  32. Reducing the problem Ti = (Ci, Pi, Mi, Oi) Ti = (Ci/fop, Pi, Mi/fop, Oi/fop) Ti = (C’i, Pi, M’i, O’i) anil@iastate.edu

  33. Optimal Solution to the reduced problem • Theorem: There exists an optimal solution to the reduced problem where the optional parts of a task Ti receive the same service time at every instance • The above theorem is proved using rigorous mathematical tools. • The intuition follows…. anil@iastate.edu

  34. Optimal solution with equal optional service times M11 O11 M21 O21 Both satisfy constraints Oi1 = (O11 + O21)/2 M11 O11 M21 O21 anil@iastate.edu

  35. Algorithm for linear quality functions • Step1: Sort all the tasks in the order of (ki/bi), where bi is the number of instances of Ti in LCM. • Step 2: Allocate maximum possible slack to the task with largest (ki/bi) Mi Quality Qi = ki * ti Oi anil@iastate.edu

  36. The entire procedure Find the optimal frequency which isthe same forall tasks Find the Maximum quality solution by determining the optional service times anil@iastate.edu

  37. The (m,k) firm guarantee Task Model • Energy aware (m,k) Problem: • (m, k)-firm deadlines • Minimize energy consumption anil@iastate.edu

  38. Static algorithm by Gang et al. [1] • Assumptions: • Each task is specified by: (Pi,Di,Ci,mi,ki). • Processor provides two voltage/frequency modes (high and low). anil@iastate.edu

  39. Static algorithm (contd..) • Algorithm: • Sort all the tasks as per their utilizations. • Test the task set for the schedulability at the High Frequency Mode. • Considers the next highest utilization task, and checks (with respect to schedulability) if it can be slowed down. anil@iastate.edu

  40. Static Algorithm: drawbacks • Algorithms’ run time increases exponentially with the number of voltage levels. • Does not capture the energy-value tradeoffs. anil@iastate.edu

  41. Conclusions • Energy-Quality-Time tradeoff is an important issue in Embedded RTS. • There is a lot of scope to work in this area (e.g. better energy aware (m, k)-firm deadline task scheduling) anil@iastate.edu

  42. References • [1] Quan, G., L. Niu and J. P. Davis, "Power Aware Scheduling for Real-Time Systems with (m,k)-Guarantee", Proceedings CNDS-04: Communication Networks and Distributed Systems Modeling and Simulation, The Society for Modeling and Simulation International, 2004. • [2] http://www.transmeta.com/crusoe/faq.html#8 • [3] Real-Time Dynamic voltage scaling for Low-Power Embedded Operating Systems, P. Pillai and K. G. Shin, in ACM SOSP, pages 89-201, 2001. anil@iastate.edu

  43. References • [4] Intra-task Voltage Scheduling on DVS-Enabled Hard Real-Time Systems, D. Shin and J. kim, IEEE Design and Test of Computers, March 2001. • [5] Maximizing the System Value while Satisfying Time and Energy Constraints, Cosmin Rusu, Rami Melhem, Daniel Mossé; ,IBM Journal of R&D, vol 47, no 5/6, 2003 • [6] Hard Real-Time scheduling for Low-energy using stochastic data and DVS Processors, Flavius Gruian, symposium on low power electronics and design, 2001. anil@iastate.edu

  44. References • [7] Scheduling with Dynamic Voltage/Speed Adjustment Using Slack Reclamation in Multi-Processor Real-Time Systems, D. Zhu, R. Melhem, and B. Childers, IEEE Trans. on Parallel & Distributed Systems, vol. 14, no. 7, pp. 686 - 700, 2003. • [8] C. Rusu, R. Melhem and D. Mossé, "Maximizing Rewards for Real-Time Applications with Energy Constraints", Accepted for publication in ACM Transactions on Embedded Computer Systems. • [9] R. Mishra, N. Rastogi, D. Zhu, D. Mosse, R. Melhem, "Energy Aware Scheduling for Distributed Real-Time Systems", Proc. of the International Parallel and Distributed Processing Symposium (IPDPS'03), Nice, France (April 2003). anil@iastate.edu

  45. Thank You!! anil@iastate.edu

More Related