a survey of energy-aware real time scheduling...

A Survey of Energy-Aware Real Time Scheduling Tools

Idawaty Ahmad

University Putra Malaysia

Faculty of Computer Science and Information Technology, University Putra Malaysia UPM Serdang

43400 Selangor

[email protected]

ABSTRACT

This paper gives a brief survey on the existing

simulation tools that are available to further

investigate the performances of energy-aware real

time scheduling algorithms. The previous work

from Y. Chandarli et.al, suggested four properties

for a real time scheduling tools to become a

reference tool by the real time scheduling

community. This paper compares the properties of

the most recent real time scheduling tools

considering to the above mentioned properties from

simulation software perspective. Programming

language is one of the important properties that

influence the popularity of these tools. This criteria

is compared between these tools. Java language is

observed to dominate the development of these

simulation tools.

KEYWORDS

Real time system, energy-aware real time

scheduling, simulation tools, DVFS, DPM.

1 INTRODUCTION

Energy consumption is a critical design issue in

real-time systems in which the system needs to

continue meeting task deadline while staying

within energy constraints especially in battery

operated systems. The Dynamic Voltage and

Frequency Scaling (DVFS) and Dynamic

Power Management (DPM) are some of the

basic technique that optimizes power

consumption at the operating system level [2-

6] in single core and multicore platforms.

2 ENERGY-AWARE REAL TIME

SCHEDULING

In general, real time scheduling problems deal

with the ordering of tasks to be executed in a

way that the deadline constraints are satisfied.

Typically, a task is characterized by its

execution time, ready time, deadline and many

other resource requirements. Some basic and

popular algorithms for single core are Earliest

Deadline First (EDF), Rate-Monotonic (RM),

Preemptive Utility Accrual Scheduling (PUAS)

[7, 8]. These algorithms are integrated with

basic DVFS and DPM mechanism to achieve

meeting the deadline constraint and minimizes

energy consumed in real-time system.

2.1 DVFS

DVFS is an efficient technique for reducing

CPU energy. DVFS adjusting supply voltage

and frequency used by the CPU as shown in

Figure 1. The DVFS framework aims at

stretching out task executions through speed

and voltage reduction. A higher speed reduces

the execution time but increase the power

consumption.

Figure 1. Optimization of consumption with DVFS [2]

Proceedings of Third International Conference on Green Computing, Technology and Innovation (ICGCTI2015), Serdang, Malaysia, 2015

ISBN: 978-1-941968-15-4 ©2015 SDIWC 19

Table 1 simplifies the notation used in Figure 1

and 2.

Table 1. Properties of a real time task.

Notation Parameters

Ai

Power

Ci Execution

time

Di Deadline

Ti Period

Instead of lowering processor voltage and

frequency as much as possible, energy-efficient

real-time scheduling adjusts voltage and

frequency according to some optimization

criteria, such as low energy consumption and

still meeting the deadline of real-time tasks.

However, reduction of processor voltage and

frequency causing slowdown in the task

execution of programs, making a trade-off

between energy saving and meeting deadline

performance. Hence it is required to schedule

the tasks properly, deciding where to change

the processor voltage and frequency to have the

best energy-efficient performance and yet does

not miss any deadline. For real-time systems,

the integrated DVFS schemes and real time

scheduling algorithm such as EDF focus on

minimizing energy consumption in the system,

while still meeting the deadlines of real time

tasks.

2.2 DPM

DPM is an effective technique for reducing

power dissipation in which it has the selective

shutdown of system components that are idle or

underutilized as shown in Figure 2. The basic

mechanism of DPM is to stop the processor

when it is not required and to wake up when

there are existing tasks [2].

Figure 2. Optimization of consumption with DPM [2]

3 EXISTING SIMULATION TOOLS

Many of the researchers in real time scheduling

community evaluate the performances of new

energy-aware real time scheduling algorithms

by using simulation. Therefore, many

simulation tools have been developed in last

few years. The effort in comparing real time

scheduling tools are done in [9, 10]. In [9], a

comprehensive comparisons are made to 18

simulation tools based on eight criteria such as

simulation, visualization, live-simulation,

shared resources, multiple cores, sporadic tasks,

open-source and programming languages. [10]

summarizes over 20 simulation tools in the

literature under four properties such as

language, design, performance analysis

scheduler profiling. However, none of them

focuses specifically on a real time scheduling

tools and energy awareness environment. This

paper summarizes simulation tools in real time

scheduling domain that support energy

optimization as shown in Table 2.

Table 2. General properties of the existing energy aware

real time scheduling tools.

Tools Year

(doc)

Language Built-in Energy-

aware algorithm /

energy profile

Yartiss

[1]

2013-

2015

Java

(open-

source)

-Energy harvesting

algorithms:

- PFP-ASAP

- PDP-ST

- PFPSlacktime

- etc

- Energy Profile

SimSo

[11]

2014-

2015

Python

(open-

- Static-EDF

- CC-EDF


ISBN: 978-1-941968-15-4 ©2015 SDIWC 20

source)

Storms

[2]

2014 Java

(open-

source)

- DPM-Processors

- DVFS Processors

Sparts

[12]

2011 Java

(open-

source)

-Leakage-Aware

Energy

Management

Algorithm

- LC-EDF

- ERTH

RTSIM

[13,14]

2001 C++ and

Java

(open-

source)

-RT-DVS

Algorithm:

- Static-EDF

- CC-EDF

- LookAhead-

EDF

- Power model

Stream

[10]

2015 Java

(not open-

source)

CC-EDF with

Total Bandwidth

Server (TBS)

algorithms

SimDVS

[15]

2003 -

(not open-

source)

InterDVS and

IntraDVS family

algorithms

It is observed that most of the tools for energy

aware real time scheduling are developed in

Java.

3.1 Yartiss

Yartiss is a real-time multiprocessor scheduling

simulator developed by researchers in Software,

Network and Real-Time Team (LRT) research

group in France [1]. Yartiss is an open source

simulator and it is built from Java. The main

interface of Yartiss is shown in Figure 4.

Yartiss allows researcher to model the energy

harvesting technique from renewable energy

sources such such as battery, solar and other

sources.

Figure 4. Snapshot of Yartiss main interface

It consists of three part i.e., the harvester,

storage unit and computing system as shown in

Figure 5[16]. The harvester converts the energy

from ambient surroundings into usable

electrical power. The storage unit is a device to

store the electrical power such as rechargeable

battery or capacitor. The computing system is a

real-time system that utilizes the energy in the

battery to run the software [16].


ISBN: 978-1-941968-15-4 ©2015 SDIWC 21

Figure 5. Energy harvesting embedded system [16]

Yartiss is the only simulator that has an energy

profile specifically for energy harvesting

purposes.

The energy profile consists of the energy source

model and energy consumption model. Some of

the energy harvesting scheduling algorithms

built in Yartiss is shown in Table 3. Two type

of scheduling algorithm i.e., fixed priority and

dynamic priority such as EDF. The scheduling

parameters of these algorithms are processor

and the energy load, the amount of incoming

energy and the energy level in the storage unit.

Table 3. Some of the energy-harvesting algorithm in

Yartiss [16-18].

Algorithms Description

EHPDP5

(Preemptive

Dynamic

Priority)

Two thresholds Emin and Emax are

configured. When energy fails under

Emin, the system is paused for a

maximal duration equal to the slack

time. If energy state reaches Emax, the

schedule is resumed.

PFP-

Slacktime

(Preemptive

Fixed

Priority)

If there is not enough energy, the

system is paused for duration equal to

the slack time. If no slack time, the

system is paused 1 time unit. If the

energy state reaches Emax, the

schedule is resumed.

PFP-ASAP

(Preemptive

Fixed

Priority)

If there is not enough energy, the

system is paused for some time unit.

Figure 6 shows an example of source codes of

PFP_ASAP in Java that is built in Yartiss.

Figure 6. Source code of PFP-ASAP energy aware

scheduling algorithm in Yartiss


ISBN: 978-1-941968-15-4 ©2015 SDIWC 22

The developer has addresses some future works

of Yartiss [1]

i. Implement the resource sharing

protocols such as memory, cache and

processor synchronization issues.

ii. Use the XML format for inputs and

outputs.

iii. Develop the processors with DVFS

capability.

3.2 SimSo

SimSo stands for Simulation of Multiprocessor

Scheduling with Overheads is a scheduling

simulator for real-time multiprocessor

architectures that takes into account some

scheduling overheads such as scheduling

decisions, context switches and the impact of

caches in real-time systems [11].

SimSo is developed by the researchers in

Laboratory for Analysis and Architecture of

System, (LAAR) research group in France [11].

SimSo is an open-source simulator developed

based on a discrete event simulation using

Python programming language. Currently, more

than 25 popular schedulers are available

including the basic energy-aware scheduling

algorithm that is Static-EDF and CC-EDF.

These are RT-DVFS family algorithms

developed at early development of energy-

aware real time scheduling. Figure 7. Snapshot of SimSo main interface

Some of the scheduling algorithms built in

SimSo is simplified in Table 4.

Table 4. RT-DVS scheduling algorithm built in SimSo


Static-EDF An offline algorithm that uses the

static slack time technique to scale the

CPU frequency. The execution speed

is set equal to the lowest available one

which guarantees the task set

feasibility.


ISBN: 978-1-941968-15-4 ©2015 SDIWC 23

Figure 8. Static-EDF[20]

CC-EDF Cycle conserving EDF uses the

dynamic slack to scale the CPU

frequency. Initially, tasks execute up to

its worst case execution time, and the

frequency is set accordingly. Upon

completion, if the actual execution

time is lesser, then the extra unused

cycles are transferred to the remaining

tasks. The remaining tasks get more

cycles and the frequency can be scaled

down.

Figure 9. CC-EDF[20]


CC-EDF in Python programming language that

is built in SimSo.

Figure 10. Source code of CC-EDF built in SimSo

A recent survey from the Association for

Computing Machinery (ACM) shows that

Python surpassed Java as the top language used

to introduce students to programming and

computer science in the US [21]. This scenario

will affect the used of Python as the most

popular language in future.

3.3 Storm

Storm stands for Simulation TOol for Real time

Multiprocessor scheduling. It is an open source

tool developed in Java by the researchers in

Parallel Heterogeneous Energy efficient Real-

time Multiprocessor Architecture, (Pherma)

research group in France [2]. It is a tool that

allows multiprocessor simulation and analyzes

energy consumption based on estimations.

Storms can analyzed the system behavior and

performances considering many features of


ISBN: 978-1-941968-15-4 ©2015 SDIWC 24

hardware architecture such as multicore design,

multiprocessor architecture with shared

memory, distributed architecture with

communication network, memory architecture

[2]. Figure 11 shows the main interface of

Storm.

Storm provides support for DPM and DVFS

techniques and can analyze the behavior and to

evaluate the performances of the policies of

scheduling while taking into account the

algorithms of energy management. The

processor type that can be used in Storm by the

energy-aware scheduler is PXA270. It supports

the dynamic adjustment of the power and the

performance of the processor based on CPU

demand.

Figure 11. Snapshot of the console and commands in

Storm

3.4 Sparts

Sparts is an open source simulator implemented

in Java developed by researchers from

Polytechnic Institute of Porto in Portugal [12].

Sparts has the capability to simulate schedulers

with various overheads such as preemption,

energy consumption and migration for multi-

core environment. The power model in Sparts

provides the information about the energy

consumption in the current state of CPU

execution such as MaxSpeed, SleepMode and

IdleMode as shown in Figure 12. It calculates

the overhead of these transitions including

switching between different sleep states, by

taking into account power properties of each

state.

Figure 12. Source codes of the power model in Sparts


ISBN: 978-1-941968-15-4 ©2015 SDIWC 25

Sparts currently contains 5 uniprocessor

scheduling algorithm i.e., EDF, RM, LLF

including the leakage aware algorithms such as

LC-EDF and ERTH. Leakage Control EDF

(LC-EDF) is the first scheduling algorithm to

minimize the leakage energy consumption in

real-time systems [22-24]. Table 5 shows the

leakage aware algorithms built in Sparts.

Table 5. Leakage aware algorithms built in Sparts


LC-EDF

[22]

Leakage Control-EDF algorithm

computes the time interval and delays

the execution of the task when CPU is

idle to extend the idle intervals and

reduce the number of power

transitions. As long as the total

utilization is less than or equal to 1, the

schedulability of the task set is

guaranteed [22].

ERTH

[25]

Enhanced Race-To-Halt algorithm

tends to run the system at top speed

with an aim to create long idle

intervals, using slack management

approach which are used to deploy a

sleep state.


LC-EDF in Java that is built in Sparts.

Although Sparts provides the Java source file,

the execution file is not provided thus not be

able to be used directly by the researchers. The

recent work that used Sparts is in [25] to

measure the performance of a power aware

thermal scheduling algorithm known as ERTH

algorithm.

Figure 13. Snapshot of LC-EDF in Sparts

3.5 RTSIM

RTSIM stands for Real-Time system

SIMulator is an open source framework to

perform discrete event simulations of real-time

control systems in distributed environment.

RTSIM is written in C++ and the GUI is

implemented in Java. The interface of RTSIM

is shown in Figure 14.


ISBN: 978-1-941968-15-4 ©2015 SDIWC 26

Figure 14. Snapshot of RTSIM taken from[26]

The scheduling algorithms that are built in

RTSIM is shown in Table 6. Most of the

scheduling algorithms in RTSIM exploit the

slack time for reducing energy consumption

when tasks have a variable execution time.

Table 6. Energy-aware algorithms build in RTSIM


RT-DVS Refer to Table 4.

DVSST

[27]

Dynamic Voltage Scaling

Algorithm for Sporadic tasks.

DVSST schedules sporadic hard

real-time tasks, reclaiming the

unused bandwidth to lower the

processor frequency. It keeps track

on the total bandwidth used by all

active sporadic tasks and CPU

frequency is changed depending on

this value. DRA

[28] Dynamic Reclaiming Algorithm

DRA is based on detecting early

completions and adjusting

(reducing) the speed of other tasks

on-the-fly in order to provide

additional power savings while still

meeting the deadlines [28].


RT-DVS in C++ that is built in RTSIM.

Figure 15. Source codes of RT-DVS in RTSIM

Although RTSIM is a bit old, it is used by the

recent work in [14] that proposed energy-aware

scheduling algorithm known as PAS-ES (Power

Aware Scheduler under EDF Scheduling). In

[26], the Java GUI is no longer available but

instead a C++ implementation of the GUI [9].


ISBN: 978-1-941968-15-4 ©2015 SDIWC 27

3.6 Stream

Stream stands for Simulation Tool for Real

time Energy-efficient scheduling and Analysis

for Multi-core processors. It is developed based

on Java language but not an open source

simulator. The software architecture of Stream

is shown in Figure 16 [10]. For simulating

energy-aware real time scheduling, it consists

of:

energy profile with power and

frequency voltage mapping

energy controller with DVFS and DPM

together with the existing algorithms

EDF and RM integrated with aperiodic

Total Bandwidth Server (TBS).

DVFS trace

Energy consumption analyzer

Figure 16. Software architecture of Stream [10]

Although Stream is published in year 2015 that

is the most recent energy aware real-time

scheduling tools, it is not an open source.

Therefore, the documentation is very hard to

find to further investigate the properties and

scheduling algorithms available in Stream.

3.7 SimDVS

SimDVS is a unified simulation environment

that provides a framework for objective

performance evaluations of InterDVS and

IntraDVS integrated with the EDF and RM real

time scheduling policy [15]. Figure 17 shows

the SimDVS simulation environment. Similar

to Stream, This tool is not an open source tools.

Therefore, the documentation is very hard to

find.

Figure 17. SimDVS simulation environment [15]


ISBN: 978-1-941968-15-4 ©2015 SDIWC 28

4 COMPARISONS

According to [1], there is no standard

simulation tool approved by the real time

community. [1] suggested four properties for a

simulator to become a reference tool by the

researchers:

a) The software must be under open source

license.

b) API of the software must be well-

documented.

c) Each part of the simulator such as task

generator, scheduler, result analyzer must

be independent from the other parts.

d) Easy to use in a way that a non-developer

researchers can be able to use it easily.

This paper summarizes the properties of the

most recent energy-aware real-time scheduling

tools according to the criteria suggested in [1].

Table 7 shows the comparisons. Column (a) to

(d) in summarizes the properties of these tools

according to the criteria suggested in [1]. In

addition, one more property that influence the

selection criteria of energy aware real-time

scheduling tools is the popularity of the

programming language being used.

Table 7. Comparison of energy aware real time

scheduling tools from simulation software perspective.

Properties (a) (b) (c) (d) (e)

Tools

SimSo

Yartiss

Storm

Sparts

RTSIM

Stream

SimDVS

It is observed that Java dominates the language

being used by the tools. However, as mentioned

earlier, in the recent survey from ACM, it

shows that Python surpassed Java as the top

language used to introduce students to

programming and computer science in the US

[21]. This scenario will affect used of Python as

the most popular language in future.

6 CONCLUSIONS

This paper gives a brief survey on the existing

energy-aware real-time scheduling tools in the

literature. This paper compares the properties of

recent real time scheduling tools considering to

the criteria proposed in [1] from simulation

software perspective. Java language is observed

to dominate the development of these

simulation tools. Overall, Yartiss is easy to use

and gives many properties to model an energy

aware scheduling including its power/energy

profile.

REFERENCES [1] Y. Chandarli, M. Qamhieh, F. Fauberteau and D.

Masson, “Yartiss: A general, modular and energy-aware scheduling simulator for real-time multiprocessor systems,” Research Report UPE LIGM ESIEE, October 2014.

[2] R. Cheour, R. Urunela, Y. Trinquet, and M. Abid, “Simulation of efficient real-time scheduling and power optimisation,” International Journal of Computer Science Issues, vol. 10(2), pp. 338-346, March 2013.

[3] M. Bambagini, M.Bertogna, and G. Buttazzo, “On the effectiveness of energy-aware real-time scheduling algorithms on single-core platforms,” Proceedings of 19th IEEE International Conference of Emerging Technologies and Factory Automation (IETA14), pp.1-8, September 2014.

[4] C.S. Stangaciu, M.V. Micea, and V.I. Cretu, “Energy efficiency in real-time systems: A brief overview,” Proceedings of IEEE 8th International Symposium on Applied Computational Intelligence and Informatics (SACI), pp.275-280, May 2013.

[5] D. Li and J. Wu, “Energy-aware scheduling on multiprocessor Platforms,” Springer Briefs in Computer Science: Springer, 2013, pp. 1-12.

[6] L. Benini, A. Bigliolo, and G.D. Micheli, “A survey of design techniques for system-level dynamic power management,” IEEE Transactions On Very Lage Scale Integration (VLSI) Systems. vol.(8)3, pp. 299-316, June 2000.

[7] G.T. Buttazzo, “Rate monotonic vs. EDF: Judgment

day,” Lecture notes in Computer Science, Springer-Verlag-Berlin Heidelberg, pp.67-83, 2003.


ISBN: 978-1-941968-15-4 ©2015 SDIWC 29

[8] I. Ahmad, S. Subramaniam, Z. Zulkarnain, and M. Othman, “A preemptive utility accrual scheduling algorithm for adaptive real-time system,” International Journal of Computer Science and Network Security, vol. 8(5), pp. 57-61, May 2008.

[9] P. Munk, Visualization of scheduling in real-time

embedded systems. Master Thesis: University of Stuttgart Germany, 2013.

[10] M. Digalwar, P.Gahukar, S. Mohan, and B.K. Raveendran, “Stream: A simulation tool for energy efficient real-time scheduling and analysis,” Proceedings of 6th International Workshop on Analysis Tools and Methodologies for Embedded and Real Time Systems (WATERS 2015), July 2015. DOI: 10.13140/RG.2.1.3148.5925

[11] M. Cheramy, P.E. Hladik, and A. M. Deplanche, “Simso: A simulation tool to evaluate real-time multiprocessor scheduling algorithms,” Proceedings of 5th International Workshop on Analysis Tools and Methodologies for Embedded and Real Time Systems (WATERS 2014), pp. 37-42, July 2014.

[12] B. Nikolic, A. Muhammad Ali, and S. Petters,

“Sparts: Simulator for power aware and real-time systems,” Proceedings of IEEE 10th International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom), pp.999-1004, November 2011.

[13] A. Manacero, M.B. Miola, and V.A.Nabuco, “Teaching real-time with a scheduler simulator,” Proceedings of Frontiers in Education Conference, vol(2) pp.15-19, October 2001.

[14] Y. Seol, J. Kim, and Y.K.Kim, “A power-aware

scheduler exploiting all slacks under EDF scheduling,” Lecture notes in Electrical Engineering, Springer Berlin Heidelberg, pp.51-57, 2014.

[15] D. Shin, W. Kim, J. Jeon, J.Kim, and S.L.Min,

“SimDVS: An integrated simulation environmnet for performance evaluation of dynamic voltage scaling algorithms,” Lecture notes in Computer Science, Springer-Verlag-Berlin Heidelberg, pp.141-156, 2003.

[16] Y. Abdeddaim, Y. Chandarli, and D. Masson, “The optimality of PFPASAP algorithm for fixed-priority energy-harvesting real-time systems,” Proceedings of IEEE 25th Euromicro Conference on Real-Time Systems (ECRTS), pp.47-56, July 2013.

[17] M. Kumari, and A. Kumar, “Real-time scheduling with DVS and harvesting energy constraints,” International Journal of Computer Applications, vol(56) 8, pp.16-22, October 2012.

[18] M., Chetto, “EDF option for scheduling real-time jobs in energy harvesting systems,” Journal Information Technology Software Engineering (JITSE). vol(3)3, pp.1-2, 2013.

[19] M., Chetto, “Task scheduling in energy harvesting real-time embedded system,” Journal Information Technology Software Engineering (JITSE). vol(2)3, pp.1-2, 2012.

[20] S. Saha, An Experimental Evaluation of Real-Time DVFS Scheduling Algorithms. Master Thesis: Virginia Polytechnic Institute and State University, 2011.

[21] P. Guo, “Python is Now the Most Popular Introductory Teaching Language at Top U.S. Universities,” URL: http://cacm.acm.org/blogs/blog-cacm/176450-python-is-now-the-most-popular-introductory-teaching-language-at-top-us-universities/fulltext. Last accessed on 22 November 2015.

[22] Y.H. Lee, K.P. Reddy, and M.K. Krishna,

“Scheduling techniques for reducing leakage power in hard real-time systems,” Proceedings of the 5th Euromicro Conference on Real-Time Systems, pp.105-112, July 2003.

[23] R. Jejurikar, C. Pereira, and R. Gupta, “Leakage

aware dynamic voltage scaling for real-time embedded systems,” Proceedings of the 41st annual Design Automation Conference, pp.275-280, 2004.

[24] J.J. Chen, and C.F. Kuo, “Energy-efficient

scheduling for real-time systems on dynamic voltage scaling (DVS) Platforms,” Proceedings of the 13th IEEE International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA 2007), pp.28-38, August 2007.

[25] M.A. Awan, Energy and temperature aware real-

time systems. Phd Thesis: Polytechnic Institute of Porto Portugal, 2014.

[26] C.Bartolini, and G. Lipari, “RTSIM,” URL: http://rtsim.sssup.it/. Last accessed on 22 November 2015.

[27] A. Qadi, S. Goddard. and S. Farritor, “DVSST: A

dynamic voltage scaling algorithm for sporadic tasks,” Technical Report TR-CSE-UNL-2003-2, May 2003.

[28] H. Aydin, R. Melhem, R. Mosse, and P. Mejia-Alvarez, “Dynamic and aggressive scheduling techniques for power-aware real-time system,” Proceedings of IEEE 22nd Real-Time Systems Symposium. (RTSS 2001), pp.95-105, December 2001.


ISBN: 978-1-941968-15-4 ©2015 SDIWC 30

a survey of energy-aware real time scheduling...

Documents