Real Time Scheduling for CPU and Hard Disk Requirements-Based Periodic Task with the Aim of Minimizing Energy Consumption

Published Online September 2015 in MECS DOI: 10.5815/ijitcs.2015.10.07

• II.    Related Work

• A.    Techniques for Saving Energy in CPU

In recent years, a significant number of studies have been done in the field of special-purpose systems for scheduling real time tasks, in which DVFS technique was used in order to improve energy or power consumption. Kamga [1] proposed a mechanism for emulating a precise CPU frequency by using the DVFS management in virtualized environments. In [10] the focus is on a scheduling approach towards the sporadic tasks set in the uni-processor system according to DVFS technique. In [5] they presented a DVFS-based algorithm to reduce energy consumption of processors through efficient use of the generated tasks’ slack times by an independent scheduler. Laszewski et al. [6] proposed a scheduling algorithm for DVFS- enabled clusters for executing multiple virtual machines in order to reduce power consumption. In [8] they introduced a mechanism for scheduling of real-time tasks on a non-ideal DVS processor with shared resources in order to obtain better energy efficiency.

• B.    Techniques for Saving Energy in Hard Disk

Many case studies have been done on the reduction of energy consumption of hard disks. In [12], [13] and [14] they focused on presenting mechanisms for reduction of power and energy consumption of hard disks regardless of them being scheduled for a set of real time tasks. Swaminathan et al. [15] presented the scheduling for a set of hard disk real time tasks with I/O requests based on DPM technique. In this paper we proposed a new algorithm for scheduling a real time periodic task with CPU and hard disk requests, in order to lower energy consumption as much as possible.

Also in Kong et al. [11] they addressed the problem of multi-resource energy minimization for frame-based realtime tasks running on a DVS-capable processor of a discrete set of operational modes with multiple off-chip DPM-capable devices.

• III.    Optimizing Energy Consumption for a Periodic Real Time Task With Cpu and Hard Disk Requests

• A. Phase1: Computing Energy Consumption of CPU and Hard Disk and Comparison of their Respective Values

In this section we will present our new algorithm for computing and comparing hard disk and CPU energy consumption .With the aim of reducing energy consumption, the execution time of the CPU and/or the R/W time of the hard disk up until the deadline will be increased. Based on which of the hard disk or CPU consumes more energy, the execution time of the task changes with one the items below:

• •    The R/W time of the hard disk increases.

• •    The execution time of the CPU increases.

• •    Both R/W time of the hard disk and CPU execution time increase.

The major parameters of the task are described in Table1.

In Figs.1, 2 and 3 CPU Energy Consumption, HDD Energy Consumption and Decision algorithm are shown, respectively. In the first step of algorithm 1, we have calculated CPU power consumption that is denoted in Eq. (1).

• 1:    Input : F_cpu,V, λ

• 2:    Output : E_cpu

• 3:    Use Eq. (1) to calculate the power consumption of CPU

• 4:    U se Eq. (2) to calculate the CPU energy consumption

• Fig.1.    CPUEC algorithm

• 1:    Input : RPM

• 2:    Output : E_HDD

• 3:    Use Eq. (3) to calculate the power consumption of HDD

• 4:    U se Eq. (4) to calculate the hard disk energy consumption

• Fig.2.    HDDEC algorithm

Table 1. The major parameters of task.

Parameter

Definition

F cpu P E HDD R t D t RE t EX t F cpu λ V min-cpu max-cpu E t cpu RWt HDD T t AS t RPM Et task

CPU Energy consumption(J) Power consumption Hard Disk Drive Energy consumption(J) Ready time(ms) Deadline time(ms) Remaining time up until the deadline(ms) Extended period of time from the deadline(ms) Frequency of CPU(GHz) Effective capacitance Voltage of CPU(V) Minimum frequency of CPU(GHz) Maximum frequency of CPU(GHz) CPU Execution time(ms) Read and Write time for hard disk drive(ms) Transfer time(ms) Average Seek time in hard disk(ms) Revolutions Per Minute in hard disk The execution time of modified task(ms)

P = λ×F cpu ×V²                (1)

Where F, λ and V represent processor’s working frequency, the effective capacitance, and processor’s working voltage, respectively. In this paper for Intel Pentium M Processor model, the maximum frequency is 1.6 GHz and the minimum frequency is 0.6GHz, so CPU energy consumption is defined as:

E cpu = P ×E t_cpu                    (2)

For calculating hard disk energy consumption, at first, we must compute power consumption that is defined as:

P=Number of platters ×(RPM)2.8×(Diameter)4.6  (3)

For Eq. (3)[17] we used HP 300GB-6G-SAS 15k rpm-SFF(2.5inch) model of hard disk drive with 2 platters, four levels for RPM and 2.5inch diameter, so hard disk energy consumption that is denoted in Eq. (4).

E HDD = P × T t                   (4)

In Fig.3 at first we compared energy consumption of the hard disk and that of the CPU. If energy consumption of the CPU is greater than that of the hard disk, algorithm 4(First Frequency-Last RPM) will be executed, otherwise algorithm 5(First RPM-Last Frequency) will be executed.

• 1:    E_cpu = CPUEC algorithm

• 2:    E_HDD= HDDEC algorithm

• 3:    If E_HDDcpu then

• 4:   FF-LR algorithm

• 5:    else If E_HDD>E_cpu then

• 6:   FR-LF algorithm

• 7:    end if

• Fig.3.    Decision algorithm

• B.    Phase2: Computing the Optimal Execution Time of the Task Based on Obtained Ideal Frequency and RPM

In this section, the second phase of our proposed algorithm will be described. In this phase after determining which of the two (CPU or hard disk) consumes more energy, based on the above algorithms, either FF-LR algorithm or FR-LF algorithm in Figs. 4 and 5 will be selected.

• B.I.    Energy Consumption of CPU is More than that of the Hard Disk

In Fig.4 first, the frequency of the CPU is set to minimum and the RPM of the hard disk is set to maximum, in other words, the CPU performs with minimum frequency, which causes the execution time of CPU to increase and the hard disk performs with maximum level of RPM, which causes an decrease of the R/W time of the hard disk. Then the execution time of the CPU and the R/W time of the hard disk are calculated with these new amounts according to Eq. (6) and Eq. (9)[16]. If the sum of the execution time of the CPU and the R/W time of the hard disk is less than Dt- Rt (task is not missed), it means in addition to prolonging the execution time of the CPU, maybe we can also prolong the R/W time of the hard disk, so the remaining time will be calculated from Eq. (5).

RE t = D t - (E t_cpu +RW t_HDD )            (5)

Based on the sum of the REt from Eq. (5), and the

RWt , the new R/W time of the hard disk is calculated. According to the Eq. (6) and the new R/W time of the hard disk, the rotational delay time will be calculated.

RW t_HDD =RD t + AS t +T t (6)

• 1:    Input : R t , D t , E t cpu ^,RW t HDD , F cpu , F max-cpu , F min-cpu AS t , T t , RPM

• 2:    Output : E t_task

• 3:    RPM = RPM_max max

• 4:    F cpu = F min-cpu

• 5:    Use Eq. (6) to calculate RW_t

• 6:    Use Eq. (9) to calculate E_t

• 7:    If E t_cpu +RW t_HDD < D t - R t then

• 8:   Use Eq. (5) to calculate RE_t

• 9:   RD t = (RW t HDD + RE t ) - (AS t +T t )

• 10:   Use Eq. (7) to calculate RPM

• 11:  Compare calculated RPM with 4 levels RPM

• 12:   Assign the first RPM level that is greater than

calculated RPM

• 13:   Use Eq. (6) with ideal RPM level to calculate

^RW t HDD

• 14:    else if E t_cpu +RW t_HDD > D t - R t then

• 15:   Use Eq. (8) to calculate EX_t

• 16:   CPU frequency changes to F ideal (from Eq. (10))

• 17:     If F ideal >  F max-cpu then

• 18:        F ideal = F max-cpu

• 19:    end if

• 20:   Use Eq. (9) with ideal F_cpu to calculate E_t

• 21:    end if

• ²²:    ^E t task = ^E t cpu ^{+ RW} t HDD

• Fig.4.    FF-LR algorithm

The desired RPM will be obtained from Eq. (7).

RPM = ( ¹ × 60)/ RD t            (7)

Then this new RPM will be compared with the levels of RPM and the first RPM level that is greater than the calculated RPM will be considered as the desired RPM value. Finally the new R/W time of the hard disk will be calculated with this RPM.

If the sum of the execution time of the CPU and the R/W time of the hard disk is more than Dt- Rt (missed task), the extended period of time will be calculated from the deadline that is defined as:

EX t = (E t_cpu +RW t_HDD )- D t            (8)

The execution time of the CPU when frequency is minimum will be calculated from Eq. (9).

E t cpu

_ sN=i CPi i

F cpu

Where N and CPI represent the total number of instructions and Clock cycle Per Instruction, respectively. In addition, the execution time of the CPU can be calculated by the current execution time of the CPU, the minimum frequency and the current frequency. According to calculated Et and EXt , the ideal CPU frequency will be obtained (Higher frequency) which is defined as follows:

р , ,=

1 ideal

F - xR min-cpu ^tcpu

E t _cpu-EX t

If this obtained frequency is more than the maximum CPU frequency, the maximum frequency will be selected as ideal. Consequently the new execution time of the CPU is calculated from Eq. (9) according to the new frequency. Finally according to these obtained execution time of the CPU and R/W time of the hard disk, the new execution time of the task will be calculated.

B.II. Energy Consumption of Hard Disk is more than that of the CPU

• 1:    Input : R t , D t , E t cpu ,RW t HDD , F cpu , F max- _C p _u, F min-cpu , AS t , T t , RPM

• 2:    Output : E t_task

• 3:    RPM = RPM_mln

• 4:    F cpu = F max-cpu

• 5:    Use Eq. (6) to calculate RW_t

• 6:    Use Eq. (9) to calculate E_t

• 7:    If E t_cpu +RW t_HDD < D t - R t then

• 8:   Use Eq. (5) to calculate RE_t

• 9:   CPU frequency changes to F_ideal (from Eq. (11))

• 10:    If F ideai <  F min-cpu then

• ^11:        F ldeal = F mln-cpu

• 12:    end if

• 13:   Use Eq. (9) with ideal F_cpu to calculate E_t

• 14:    else if E t_cpu +RW t_HDD > D t - R t then

• 15:   Use Eq. (8) to calculate EX_t

• 16:    Rd. = (RW t HDD - EX t ) - (AS t +T t )

• 17:   Use Eq. (7) to calculate RPM

• 18:  Compare calculated RPM with 4 levels RPM

• 19:  Assign the first RPM level that is greater than

calculated RPM

• 20:   Use Eq. (6) with ideal RPM level to calculate

  RW t

  
  

  HDD

  
  

• 21:   Use Eq. (5) to calculate RE_t

• 22:    If RE t > 0 then

• 23:      CPU frequency changes to F_ideal(from Eq.

(11))

• 24:      Use Eq. (9) with ideal F_cpu to calculate E_t

• 25:    end if

• 26:    end if

• ²⁷:    ^E t task = ^E t cpu + ^RW t HDD

Fig.5. FR-LF algorithm

In Fig.5 first the RPM of the hard disk is set to minimum and the frequency of the CPU is set to maximum. Then the execution time of the CPU and the R/W time of the hard disk are calculated with these new amounts according to Eq. (6) and Eq. (9). If the sum of the execution time of the CPU and the R/W time of the hard disk is less than the Dt- Rt(task does not miss), it means in addition to the prolongation of the R/W time of the hard disk, we can also prolong the execution time of the CPU, so this remaining time will be calculated form Eq. (5).

According to calculated Et and REt, the ideal CPU frequency will be obtained (lower frequency) which is defined as follows:

p 1 , — ideal

F max-cpu ×E ^tcpu

E t _cpu+RE t

If this obtained frequency is less than minimum CPU frequency, minimum frequency will be selected. Finally the new execution time of the CPU will be calculated according to the new frequency.

If the sum of the execution time of the CPU and the R/W time of the hard disk is more than Dt- Rt (missed task), the extended period of time will be calculated from the deadline according to Eq. (8). Then based on subtraction of the RWt from the EXt, the new R/W time of the hard disk is calculated. According to the Eq. (6) and the new R/W time of the hard disk, the rotational delay time will be calculated. The desired RPM will be obtained from Eq. (7).

As a result, the new R/W time of the hard disk will be calculated according to the ideal RPM level. Finally, according to obtained execution time of the CPU and R/W time of the hard disk, the new execution time of the task will be calculated.

If the task is not initially missed, after applying our algorithms, we will have a task with an execution time which approaches the deadline, therefore the energy consumption of CPU and/or hard disk is reduced as much as possible, otherwise the execution time of the task is decreased until the deadline.

• IV.    Performance Evaluation

This section presents the results of our proposed algorithm. Table2 summarizes the specifications of CPU and hard disk that are used in our simulation. We used Matlab software for evaluating our proposed algorithm. 50 tasks were randomly generated of which 10% were initially missed. Fig.6 shows the average execution time of the tasks prior and subsequent applying our algorithm. After applying the algorithm, the average execution time of the tasks increases (because most of them were not initially missed). This algorithm could extend the execution time of the CPU and/or the R/W time of the hard disk up until the deadline. Therefore the energy consumption is reduced by about 25%, which is very significant (Fig.7). Also the execution time of each separate task can be seen in Fig.8 prior and subsequent applying the algorithm.

Real Time Scheduling for CPU and Hard Disk Requirements-Based Periodic

Task with the Aim of Minimizing Energy Consumption

Fig.6. Average execution time of task

Fig.7. Average energy consumption

Number of tasks

Fig.8. Execution time of tasks

Fig.9 shows the number of tasks that are missed prior and subsequent to the algorithm execution in which the number of the missing tasks is reduced by 80%. The amount of RPMs of the hard disk and frequencies of the CPU are shown in Fig.10. As shown in Fig.10. (a) and Fig.10. (b) the frequency of the CPU and the RPM of the hard disk in most of the tasks reduced to a minimum, respectively. Their respective average values before and after applying our algorithm are also shown in Fig.11. In this figure a reduction in the values of the CPU frequency and the RPM of the hard disk can be observed after running the algorithm.

Table 2. Specification of CPU and hard disk drive

HP hard drive
Name	Value
Capacity	300,000MB
Height	0.62 in (15.6 mm)
Width	2.98 in (75.7 mm)
Depth	4.67 in (118.7 mm)
Interface	SAS
Transfer            Rate Synchronous(Maximum)	6 Gb/sec
Physical Configuration	Bytes/Sector 512 Logical Blocks 585,937,500 Rotational   Speed:   5400(Min), 7200, 10,000 and 15,000(Max)rpm
Operating Temperature(System Inlet Air Temperature)	50° to 95° F (10° to 35° C)
Number of Platters	2
Intel Pentium M Processor
Name	Value
High         Frequency Mode(HFM)	Frequency 1.6 GHz Voltage 1.484V TDP(Thermal Design Power) 24.5w
Low        Frequency Mode(LFM)	Frequency 600 MHz Voltage 0.956V TDP(Thermal Design Power) 6w

Fig.9. Number of missed tasks

(a) CPU frequency

(b) RPM of hard-disk

Fig.10. (a) The values of CPU frequencies and (b) RPM of hard-disk, before and after applying algorithm

z

У 1.1

■ Before running presented algorithm ■ After running presented algorithm

(a) Average frequency

(b) Average RPM

Fig.11. (a) Average CPU frequency and (b) Average RPM of hard disk drive before and after applying algorithm

• V.    Conclusion And Future Work

In this paper we have presented a novel scheduling algorithm in order to reduce energy consumption in a real time periodic task with CPU and hard disk requests. To achieve this goal, the execution time of the CPU and/or the R/W time of the hard disk must be changed according to the remaining time and consumed energy. In the future study we intend to propose a new algorithm with the aim of minimizing energy consumption with a minimal performance reduction in multitask-multicore processes.