Energy Efficient Scheduling Techniques For Real-Time Embedded Systems

Transcription

1 Energy Efficient Scheduling Techniques For Real-Time Embedded Systems Rabi Mahapatra & Wei Zhao This work was done by Rajesh Prathipati as part of his MS Thesis here. The work has been update by Subrata Acharya & Nitesh Goyal Copyright: A&M 3/9/2004 1

2 Outline Introduction Motivation Related Work Single Processor Systems Distributed Multiprocessor Systems Experiments & Results Summary 3/9/2004 2

3 Sample Embedded Systems Introduction PDA audio/video entertainment devices robots Handheld computer Mobile Phone Network Camera Wireless presentation Gateway Cerfcube 3/9/2004 3

4 Application Specification for Embedded Systems Periodic Task graphs Each task characterized by: Period Execution time Deadline Sporadic Tasks Invoked at any time Hard deadline Soft Aperiodic Invoked at any time No deadline t2 t1 t4 t3 Period =90, Deadline =90 t5 Sporadic Task, Deadline =30 Typical Input Specification of Embedded Systems 3/9/2004 4

5 Why Low Power? High Power dissipation causes Chip failures Expensive Cooling & Packaging overheads High Manufacturing Costs Portable Systems, User convenience limited by: Battery Size Recharging Interval 3/9/2004 5

6 Power Management Processor power dissipation is a function of α. C l. V 2 dd. f Various Low-Power Techniques System-Level Architecture-Level Circuit-Level System-Level power reduction techniques: Dynamic Voltage Scaling Dynamic Power Management 3/9/2004 6

7 System Level Power Management Taxonomy SLPM DPM LPS (DVS) Fixed Tasks Variable Tasks Fixed Task set Variable Task set Single Processor Single processor Single Processor Single Processor Multiprocessors Multiprocessors D P (contd..) Multiprocessors No Restrictions Multiprocessors SLPM System Level Power Management DPM Dynamic Power Management LPS Low power Scheduling Tolerance DL Hard Realtime 3/9/2004 7

8 System Level Power Management Taxonomy (contd( ) D P Tolerance DL Hard Real time No Precedence With Precedence Periodic Periodic Periodic + Sporadic Periodic + Sporadic 3/9/2004 8

9 Our Objective Given Embedded system and its application task graphs with library functions (i.e. period, execution time, Deadline etc.), our goal is to Reduce the system wide power consumption while guaranteeing the deadlines 3/9/2004 9

10 Related Work Multi-Processor J.Luo and N.K.Jha,, 2001 Battery-Aware static scheduling Global shifting scheme & local schedule transformations More suitable to small scale systems R.Mishra, N.Rastogi,, and D.Zhu,, 2003 Energy aware scheduling for distributed Greedy and gap-filling dynamic power management techniques Limited to task graphs with equal deadline D. Zhu, R. Melhem,, and B. Childers, 2003 Scheduling with Dynamic Voltage/Speed Slack sharing among processors, global queue Limited homogenous systems with shared memory Single Processor: G.Quan,, and X. HU, 2001 Minimum constant voltage for each interval Assumes deadline less than or equal to period. V.Swaminathan,, and K.Chakrabarty,, 2000 Low-energy earliest deadline first heuristic No guarantee on required maximum processor speed 3/9/

11 Contributions Provides a framework for single processor that consider tasks Whose response time is greater than the period. With Precedence constraints Introduced chain of task set based execution approach to model low-power in distributed embedded systems. 3/9/

12 Energy Efficient Scheduling Techniques for Single Processor 3/9/

13 Proposed Approach Proposal: A 3-step approach to reduce power in single processor embedded systems with arbitrary response times and precedence constraints. Step1: Task priority assignment that guarantees precedence constraints. Step 2: Determination of task speed that guarantees deadlines. reduces power consumption. Step 3: Dynamic power management Idle Intervals. Run-time variations in task execution time. 3/9/

14 Task Modeling Periodic task graphs Scheduled according to their priorities Sporadic task Invoked at any time Hard deadline Execution slot is needed Let µ be the worst-case execution time and d be the deadline Execution Slots are defined with Period : d -µ Deadline: d -µ 3/9/

15 STEP 1 : Priority Assignment Arrange the task graphs & EX. Slots in increasing order Of their period Remove the task graph with smallest period no no Remove the node with no Predecessor and least slack time If all nodes in the Graph are Assigned priorities Assign the node next highest priority yes List is empty yes END 3/9/

16 STEP 2 : Task Speed Determination Arrange tasks in decreasing order of priority Find the task with largest speed, s. For each task in the list, determine the speed at which the task and all high priority tasks in the list can be run Mark the speed for this task and all other high priority tasks as s Remove all these tasks from the list no List is empty yes END 3/9/

17 Task Schedulability Let I = {T 1,T 2,,T N } be the task set arranged in decreasing order of priorities. Characteristics of T i : {P i, e i, D i }. A task set is feasible if the deadline of all tasks are always met. Critical Instant Theorem (Liu( and Layland, 1973) Scheduling algorithms for multiprogramming if a task meets its deadline whenever the task is requested simultaneously with all the high priority tasks, then the deadline will always be met for all task phasing. 3/9/

18 Task Schedulability (Contd ) In other words, the task set I = {T 1,T 2,,T N } is schedulable if and only t i D i i =1,..n, where i 1 k = 1 t i e Pk k + e i t i if P i D i.. (1) otherwise t i,j D i,j i =1,..n, and j instances of t i, where t i,j = R(t i,j + (j-1)p i ) (j-1)p i, where i 1 k = 1 t i R(t i,j ) = ek + j*e i (2) P k 3/9/

19 STEP 2 : Task Speed Determination Arrange tasks in decreasing order of priority Find the task with largest speed, s. For each task in the list, determine the speed at which the task and all high priority tasks in the list can be run Mark the speed for this task and all other high priority tasks as s Remove all these tasks from the list no List is empty yes END 3/9/

20 Step 3: Dynamic Power Management During System operation, idle intervals arise when: Actual task execution time is less than the worst-case execution time. (that is assumed at the time of fixed priority scheduling). Since these Idle intervals can not be exploited by off-line methods. An on-line method that adapts the clock speed to take advantage of idle intervals is needed. 3/9/

21 DPM (Contd(..) Schedule the tasks according to their pre- determined speeds in a preemptive manner. If the current task has finished and the queue of ready tasks is empty, then: Determine the length of idle interval If feasible, put the processor in the power down mode. 3/9/

22 Experimental Setup Event driven simulator Intel Strong Arm SA-1100 Embedded Processor Specifications Real-world test cases (CNC controller, INS, avionics, ) 3/9/

23 Benchmarks Test cases # Periodic # sporadic # tasks Utilization task graphs tasks with D > P Synthetic I Synthetic II Synthetic III CNC [1] INS [2] Avionics [3] Characteristics of various test cases 3/9/

24 % Energy savings VLPS [5] proposed technique Various low power techniques Comparison of % Energy savings with various Low power techniques CNC INS 3/9/

25 100.00% % Energy Savings 90.00% 80.00% 70.00% 60.00% 50.00% 40.00% 30.00% 20.00% % %Energy Savings 0.00% Syntheti c I Syntheti c II Syntheti c III CNC INS Avi oni cs various test cases % Energy Savings with the proposed technique on various test cases 3/9/

26 Energy Efficient Scheduling Techniques for Multi-Processor Embedded Systems 3/9/

27 Overview Preliminaries System model Slack distribution heuristic Periodical determination of service rate Experiments & Results 3/9/

28 Preliminaries Command and control systems that comprise of hard real-time applications in a distributed environment. An application comprises of: Chain(s) ) of tasks Hard deadlines Exchange of messages during execution Admitting task set (Connection establishment) : Key Issues Traffic descriptor [6] Worst-case delay analysis Power Reduction approaches slack distribution Clock speed adaptation during system run-time 3/9/

29 System Model A task set is described by a vector triplet ( P, C, i i ) where C,.. i D P i i ( C i 1 C i n ) ({ D i 1 / },... D in) P i1,,... D i PE1 PE2 PE3 M 1 M 2 A distributed system with 3 nodes & 2 task sets 3/9/

30 Connection Establishment Task set admission: Key Phases Setting up task set Reply task set Setting up task set : Key Issues local worst-case delay < local deadline end-to to-end worst-case delay < end-to to-end deadline Reply task set : Key Issues Slack distribution Service rate < 1 (periodic service rate determination) 3/9/

31 Observations Processing of messages at a node can be extended up to their delay bounds. This slack can be utilized to increase the worst-case delay tolerable at the computational nodes involved in processing the task set. The actual processing time demanded by the messages of a task set during the run-time varies and is less than the worst-case specification. A technique to adapt the clock speed periodically is introduced to take advantage of run-time variations 3/9/

32 Slack Distribution The slack in a task set is the difference between the end- to-end deadline and the sum of the worst-case delays suffered at each node. This slack can be distributed among the nodes serving task set to reduce the system energy consumption. The slack is distributed among the nodes according to the service rate of the nodes. 3/9/

33 Service Rate Determination Key Issues: Monitoring the traffic pattern Feedback incorporation while determining service rate. Periodical service rate determination guarantees processing of messages of outstanding intervals by their delay bounds guarantees processing of messages of upcoming interval by their delay bounds Scheduling policies considered: FCFS & WRR 3/9/

34 FCFS Scheduling Policy The new service rate at the beginning of every interval is determined according to S and the corresponding queue is determined according to Q t t = = k j= 1 k j = 1 j Q t S j t + Γ d ( ) t FCFS where k = { t max( s, t ( n 1) ) }/ t The service rate S t j should be such that it must process the outstanding messages that arrived during the interval (t-j,t-(j-1) ) by their remaining delay bound. i.e., (d fcfs -j ).. ( j ) j FCFS j S t d Q t 3/9/

35 WRR Scheduling Policy The new service rate at the beginning of every interval is determined according to S i t k = j = 1 S i, t j + Γ Ξ i t i ( ) ( ) d and the corresponding queue is determined according to k i Q t = Q j = 1 i, j t S i t, j i The service rate and the corresponding processing time demanded by the outstanding messages that arrived during the interval (t-j,t-(j-1) ) are given by S. ( - j ) Q i, j t i d i, j t 3/9/

36 Experimental Setup Event driven simulator Socket interface for communication Intel PXA250 XScale Embedded Processor Real-life life test cases (DSP, Multimedia,..) 3/9/

37 Benchmarks Test Cases Number Of Nodes Number Of Connections Number Of Modes Synthetic I Synthetic II Synthetic III Multimedia DSP [4] Characteristics of various test cases 3/9/

38 Benchmarks (Contd ) Test Cases Mode 1 (nodes, connections) Mode 2 (nodes,connecti ons) Mode 3 (nodes,connectio ns) Synthetic (10,30) (9,20) (9,25) (10,30) Multimedia( 4,4) (3,2) (3,3) (4,4) Mode configurations for Multimedia and Synthetic test cases 3/9/

39 Energy Saving versus Slack distributation System Energy Savings % FCFS (3,10) (5,20) (10,30) MM(4,4) DSP(16,31) Srate Equal Wcet Greedy Slack Distribution Schemes System Energy Savings % WRR Srate Equal Wcet Greedy Slack Distribution Schemes 3/9/

40 Energy Saving at different Modes System Energy Savings% Synthetic 1 Synthetic 0.8 Multimedia 1 Multimedia 0.8 Normalised Peak Power Mode 1 Mode 2 Mode 3 3/9/

41 Service rate at intervals Norm alised Service Rate (10,30) at one node Intervals /9/

42 Service rate vs MI (3,10) at one node Normalised Service Rate MI (Monitoring Interval) 3/9/

43 Overhead due to number of task sets on service Overhead(usecs) (10,30) at one node Number of connections 3/9/

44 Summary Energy Efficient Scheduling technique for Single Processor that: handles Sporadic and periodic task graphs with precedence constraints takes into account tasks with arbitrary response times determines minimum speed for each task adapts clock speed to take advantage of idle intervals. A connection based task execution approach for distributed embedded systems that: effectively distributes the slack available in the connection to reduce system wide power consumption. periodically adjusts the clock speed to take advantage of run-time variations. Experimental results indicate that the proposed techniques yield significant energy savings. 3/9/

45 References 1. N. Kim, M. Ryu, S. Hong, M. Saksena, C. Choi, and H. Shin, Visual assessment of a real time system design: A case study on a CNC controller, in Proc. IEEE Real-Time Systems Symposium, December A. Burns, K. Tindell, and A. Wellings, Effective analysis for engineering real-time fixed priority schedulers, IEEE Trans. on Software Eng., vol. 21, no. 5, pp , May C. Locke, D. Vogel, and T. Mesler, Building a predictable avionics platform in Ada: A casestudy, in Proc. IEEE Real-Time Systems Symposium, December C. M. Woodside and G. G. Monforton, Fast allocation of processes in distributed and parallel systems, Proc. IEEE Trans. Parallel & Distr. Systems., vol. 4, no. 2, pp , Feb /9/

46 References (Contd(..) 5. G.Quan,, and X.Hu,, Energy efficient fixed priority scheduling for real-time systems on variable voltage processors, In Proc. Design Automation Conference, June A.Raha, N.Malcom,, and W.Zhao,, Guaranteeing end-to to-end deadlines in ATM networks, In Proc. International conference on Distributed Computing Systems, May /9/

47 THANK YOU 3/9/