JP4385263B2 - Job scheduling apparatus and job scheduling method - Google Patents

Description

  The present invention relates to a job scheduling apparatus and a job scheduling method.

Conventionally, many scheduling apparatuses that perform job scheduling for a system composed of a plurality of machines have been disclosed (for example, see Patent Documents 1 and 2).
By the way, the conventional job scheduling apparatus is based on the premise that the system is composed of the same machine. For this reason, job scheduling is performed on the assumption that the same execution time is used regardless of which machine executes the job, and that the same system resource is used when any machine executes the job.

Therefore, when scheduling a parametric job with the same job but different parameters, simple scheduling is performed in which the same number of parametric jobs are assigned to machines that can execute the parametric job without distinguishing the machines. Just do it.
JP 2002-358293 A International Publication No. 03/083693 Pamphlet

  However, at present, there are many systems composed of different machines. In such a system, when the conventional job scheduling method is performed, a machine that finishes running quickly and a machine that does not finish running come out.

  Such a situation is unsatisfactory for both the user who wants to obtain the execution result at the fastest speed and the administrator who wants to improve the machine utilization rate.

  SUMMARY OF THE INVENTION An object of the present invention is to provide a job scheduling apparatus and a job scheduling method that can perform optimal job scheduling of parametric jobs even for systems composed of different machines.

In order to achieve the above object, the present invention provides:
In a job scheduling apparatus that performs scheduling for assigning a plurality of parametric jobs to a system including one or more types of machines,
Trial execution means for causing each of the plurality of machines to execute one of the plurality of parametric jobs;
Information collecting means for collecting trial execution results of parametric jobs of each of the plurality of types of machines;
Scheduling processing means for scheduling the plurality of parametric jobs based on the result of trial execution of each parametric job of the plurality of types of machines.

The information collecting means acquires the trial execution time of each of the plurality of types of machines,
The scheduling processing means may perform scheduling for executing the plurality of parametric jobs in a minimum time based on trial execution times of the plurality of types of machines.

In addition, the information collection unit acquires power consumption of each of the plurality of types of machines,
The scheduling processing unit may perform scheduling for executing the plurality of parametric jobs with minimum power consumption based on the power consumption of each of the plurality of types of machines.

Further, the information collecting means acquires the trial execution time and power consumption of each of the plurality of types of machines,
The scheduling processing unit is configured to execute the plurality of parametric jobs with the minimum power consumption under a setting for minimizing a machine usage rate based on the trial execution time and the power consumption of each of the plurality of types of machines. Perform scheduling.

  As described above, according to the present invention, the trial execution unit causes each of a plurality of types of machines to execute any one of a plurality of parametric jobs, and the information collection unit includes each of the plurality of types of machines. The trial execution results are collected, and the scheduling processing means schedules a plurality of parametric jobs based on the trial execution results of each of the plurality of types of machines.

  Therefore, even when the system is composed of a plurality of types of machines, an effect that a parametric job can be appropriately scheduled can be obtained.

  The best mode for carrying out the present invention will be described below with reference to the drawings.

  Referring to FIG. 1, the job scheduling apparatus 101 of this embodiment performs optimal scheduling of the parametric job group 102 for the system 103.

  Each of the parametric jobs in the parametric job group 102 has the same job, but only the parameters. That is, each parametric job has the same execution time when executed on the same machine.

System 103 includes each or one of the N kinds of machines, M _1, M ₂ N kinds of machines, respectively, ..., and M _{N. Further, M 1, M 2, ···} , a collection of machine M _N, respectively group of machines (M ₁₎ 104 _1, machine group _{(M 2) 104 2, ···} , machine group (M _N) 104 _N. Also, machine group M _1, M _2, · · ·, a viable number of jobs currently in a machine M _N, respectively E _1, E _2, · · ·, and E _N.

Job scheduling apparatus 101, machine M ₁ of N type, M _2, · · ·, for each of the M _N, a trial execution unit 111 to execute attempt any one parametric job parametric job group 102, N species machine M _1, M _2, ..., an information collecting unit 112 to collect information trial run results of the M _N, machine M _1, M ₂ of the N type, ..., the trial run results of the M _N And a schedule processing unit 113 that schedules the parametric job group 102 based on the schedule processing unit 113.

  Hereinafter, a scheduling method performed by the scheduling apparatus 101 illustrated in FIG. 1 will be described.

The scheduling apparatus 101 performs scheduling for executing any of the following (1) to (3) as an optimal scheduling method.
(1) Execute parametric job group 102 with minimum power consumption (2) Execute parametric job group 102 with minimum power consumption (3) Execute parametric job group 102 with minimum power consumption under minimum machine utilization rate setting Each of the scheduling methods 1) to (3) will be described.
(1) Executing the parametric job group 102 in the shortest time First, the trial execution unit 111 performs trial execution. Specifically, trial execution unit 111, N type machine group M _1, M ₂ as shown in FIG. 1,..., For one machine M _N, of any one of a parametric job group 102 Run a parametric job on a trial basis.

Next, the information collecting unit 112, machine group M ₁ of N type, M _2, · · ·, to measure the execution time each machine has executed the M _N. Here, machine group M _1, M _2, · · ·, the execution time of each machine M _N, respectively T _1, T _2, · · ·, and was T _N.

Here, the scheduling processing unit 113 defines the scheduling index S _x of the machine group M _x (x = 1, 2,..., N) as in Expression 1.

In this case, assuming that the total number of jobs in the parametric job group 102 is J, the schedule processing unit 113 obtains the number of jobs A _x to be assigned to one machine in each machine group M _x as shown in Equation 2.

  Actually, since the value obtained by Expression 2 may be a decimal, the schedule processing unit 113 performs the fraction processing below the decimal point as follows.

  First, only the numbers after the decimal point of the number of jobs of each machine are compared, and the machines are sorted in descending order of the value after the decimal point.

  Next, the number of temporary jobs is determined by rounding down the number of jobs of each machine.

  Next, the number of temporary jobs for each machine is summed, and the difference from the number of jobs actually executed is calculated. Let this difference be the number of remaining jobs.

Thereafter, jobs are assigned one by one from the top machine in the sorted order until the number of remaining jobs is the same.
(2) Executing Parametric Job Group 102 with Minimum Power Consumption First, the trial execution unit 111 performs trial execution. Specifically, trial execution unit 111, N type machine group M _1, M ₂ as shown in FIG. 1,..., For one machine M _N, of any one of a parametric job group 102 Run a parametric job on a trial basis.

Next, the information collecting unit 112, machine group M _1, M _2, · · ·, during parametric job execution by the machine M _N, for capable machine power measurement of hardware monitoring, hardware Measure power consumption by monitoring. On the other hand, for machines where power consumption cannot be measured by hardware monitoring, the CPU usage rate is measured, and a predicted power consumption value is derived from the value and a predicted power function set in advance from the machine specifications. . Here, machine group M _1, M _2, · · ·, the power consumption of each machine M _N, respectively F _1, F _2, · · ·, and was F _N.

Next, the scheduling processing unit 113 sorts F _1, F _2,..., The F _N in the ascending order, F _'1, F' ₂ from the beginning, ..., and F _'N.

  Assume that the above replacement is performed as shown in Equation 3.

In this case, the scheduling processing unit 113, similarly M _1, M _2, · · ·, performs replacement such as Equation 4 also _{M N, M '1, M} ' 2, ···, M 'N and To do.

Further, the scheduling processing unit 113, E ₁ similar sort, E _2, · · ·, performed also _{E N, E '1, E} ' 2, ···, and E _'N.

In this case, executing the entire job parametric job group 102 the machine groups M _'1 of the machine, a parametric job group 102 can be performed with minimal power consumption.

Since the machine groups M 'number of ₁ machine E' is _one, the scheduling processing unit 113, the job number A ₁ to be allocated to per machine machine groups M _'1, calculated as Equation 5.

Also in this case, since the value obtained by Equation 5 may be a decimal, the scheduling processing unit 113 performs the same fraction processing as in (1).
(3) Executing Parametric Job Group 102 with Minimum Power Consumption under Minimum Machine Utilization Setting First, the trial execution unit 111 performs trial execution. Specifically, trial execution unit 111, N type machine group M _1, M ₂ as shown in FIG. 1,..., For one machine M _N, of any one of a parametric job group 102 Run a parametric job on a trial basis.

For the scheduling process, the information collecting unit 112, machine group M _1, M _2, · · ·, to collect the power consumption F and execution time T in which each machine has executed the M _N. Here, machine group M _1, M _2, · · ·, the execution time of each machine M _N, respectively T _1, T _2, · · ·, a T _N, machine group M _1, M _2, · Suppose that the power consumption of each machine of _MN is F ₁ , F ₂ ,..., F _N , respectively.

  Next, if the minimum machine utilization rate is R, the trial execution unit 111 obtains the number K of machines to be used for executing the parametric job group 102 as shown in Equation 6.

Next, the scheduling processing unit 113 sorts F _1, F _2,..., The F _N in the ascending order, F _'1, F' ₂ from the beginning, ..., and F _'N.

  Assume that the above replacement is performed as shown in Equation 7.

In this case, the scheduling processing unit 113, similarly M _1, M _2, · · ·, performs replacement, such as Equation 8 also _{M N, M '1, M} ' 2, ···, M 'N and To do.

Further, the scheduling processing unit 113, E ₁ similar sort, E _2, · · ·, E _N and T _1, T _2, · · ·, performed also T _N, respectively E _'1, E' ₂ , ···, E _'N and _{T' 1, T '2,} ···, T' and _N.

Next, the scheduling processing unit 113 sequentially adds E ′ ₁ , E ′ ₂ ,... Until the number of machines becomes K or more. Here, it is assumed that K is equal to or greater than K when E ′ _M (1 ≦ M ≦ N) is added.

In this case, the scheduling processing unit 113 defines E ′ _M as Equation 9 so that the number of used units is exactly K.

Here, the scheduling processing unit 113 defines the scheduling index S ′ _x of the machine group M ′ _x (x = 1, 2,..., M) as in Expression 10.

In this case, assuming that the total number of jobs in the parametric job group 102 is J, the scheduling processing unit 113 obtains the number of jobs A _x to be assigned to one machine in each machine group M ′ _x as shown in Equation 11.

  Also in this case, since the value obtained by Expression 10 may be a decimal, the scheduling processing unit 113 performs the same fraction processing as in (1).

As described above, in this embodiment, the trial execution unit 111 causes each of the machines M _{1 to} M _N to execute one of the parametric job groups 12 and the information collection unit 112 performs the machine group M 1. The trial execution results of each of the machines _{1 to MN} are collected, and the scheduling processing unit 113 schedules the parametric job group 12 based on the trial execution results of the machines of the machine groups M _{1 to MN} .

Therefore, the parametric job group 12 can be appropriately scheduled even when the system 103 includes a plurality of types of machine groups M _{1 to} M _N.

  Also, the user or system administrator can execute (1) the parametric job group 102 in the shortest time, (2) execute the parametric job group 102 with the minimum power consumption, and (3) the parametric job group 102 under the minimum machine utilization rate setting. Can be selected from three types of scheduling methods.

  Further, the job scheduling method of the present embodiment is very simple and can be realized only by adding a small program.

  In this embodiment, a simple machine configuration and a single job have been described. However, in the present invention, when one machine has a plurality of CPUs, a job is a parallel job. It is also applicable to such cases.

It is a figure which shows the structure of the scheduling apparatus by one Embodiment of this invention.

Explanation of symbols

101 scheduling apparatus 102 parametric job group 103 system 104 ₁ -104 _N machine group 111 trial execution unit 112 the information collection unit 113 schedule processing unit

Claims (2)

In a job scheduling apparatus that performs scheduling for assigning a plurality of parametric jobs to a system including one or more types of machines,
Trial execution means for causing each of the plurality of machines to execute one of the plurality of parametric jobs;
Information collecting means for collecting trial execution results of parametric jobs of each of the plurality of types of machines;
Based on the trial run results of each parametric job of the plurality of types of machines, have a, a scheduling processing unit configured to perform scheduling of the plurality of parametric jobs,
The machine M ₁ of N type in the system, M _2, ···, if it contains M _N, machine M _1, M _2, ···, the setting under which minimizes the utilization of the M _N , When executing J parametric jobs with minimum power consumption,
It said information collecting means, said N kinds of machine M _1, M _2, · · ·, obtains the power consumption of each of the trial running time of M _N,
The scheduling processing means includes
Obtain the number of machines K to be used to execute J parametric jobs,
Said N kinds of machine M _1, M _2, · · ·, and sorted in the order power consumption of each of M _N is small, defined from the beginning of the machine _{M '1, M' 2,} ···, and M _'N as well as, the machine _{M '1, M' 2,} ···, M 1 'F the power consumption of _{N', F '2, ···} , F' N, the number of _{E '1, E' 2,} · ..., defined E _'N, a trial execution time _{T' 1, T '2,} ···, T' is _N,
E ′ ₁ , E ′ ₂ ,..., E ′ _N are sequentially added to obtain E ′ _M (1 ≦ M ≦ N) when K becomes equal to or greater than K ,

When defined by the machine M'x (x = 1,2, ···, M) the number of jobs A _x to be assigned to each,

A job scheduling apparatus obtained by In a job scheduling method for performing scheduling for assigning a plurality of parametric jobs to a system including one or more types of machines,
A trial execution step of causing each of the plurality of types of machines to execute any one of the plurality of parametric jobs;
An information collecting step of collecting trial execution results of each parametric job of the plurality of types of machines;
Based on the trial run results of each parametric job of the plurality of types of machines, have a, a scheduling processing step of performing scheduling of the plurality of parametric jobs,
The machine M ₁ of N type in the system, M _2, ···, if it contains M _N, machine M _1, M _2, ···, the setting under which minimizes the utilization of the M _N , When executing J parametric jobs with minimum power consumption,
Wherein in the information collection step, the N kinds of machine M _1, M _2, · · ·, acquires the respective trial execution time and power consumption of the M _N,
In the scheduling process step,
Obtain the number of machines K to be used to execute J parametric jobs,
Said N kinds of machine M _1, M _2, · · ·, and sorted in the order power consumption of each of M _N is small, defined from the beginning of the machine _{M '1, M' 2,} ···, and M _'N as well as, the machine _{M '1, M' 2,} ···, M 1 'F the power consumption of _{N', F '2, ···} , F' N, the number of _{E '1, E' 2,} · ..., defined E _'N, a trial execution time _{T' 1, T '2,} ···, T' is _N,
E ′ ₁ , E ′ ₂ ,..., E ′ _N are sequentially added to obtain E ′ _M (1 ≦ M ≦ N) when K becomes equal to or greater than K ,

When defined by the machine M'x (x = 1,2, ···, M) the number of jobs A _x to be assigned to each,

Job scheduling method determined by

Images