CN101271407A - Gridding scheduling method based on energy optimization - Google Patents

Abstract

The invention relates to a grid scheduling method based on energy optimization, which includes: 1. energy resources are the research focus of scheduling and factors such as initial energy value, network bandwidth, etc. are taken into consideration; 2. the energy optimization is achieved to ensure minimum energy consumption in the resource scheduling and the energy consumption in the grid resource scheduling is divided into calculation consumption and network communication consumption; 3. the optimization of time span is taken into consideration so as to ensure the load balance of the resource in the process of energy optimization; 4. a grid scheduling method based on energy optimization with a comprehensive consideration of energy constraint and time constraint is proposed. The method has the advantages that: 1. the limitation of singleness of resource type in the traditional resource scheduling calculation is changed; 2. the energy optimization is taken as the research focus of the scheduling model; 3. the problem of load imbalance caused in the process of energy optimization is solved; 4. an energy cost function is defined on the basis of time span optimization and energy optimization and the grid resource scheduling method based on energy optimization is proposed on the basis of the cost function.

Description

A kind of based on energy-optimised gridding scheduling method

Technical field

The invention belongs to a kind of gridding scheduling method, particularly a kind of based on energy-optimised gridding scheduling method.

Background technology

Grid computing is a kind of Internet computation schema that rises gradually in recent years, its objective is for the dynamic Virtual Organization of structure on the Internet resources environment of distribution, isomery, autonomy, and portion realizes striding the resource sharing and the resource collaboration in autonomous territory within it, effectively satisfies towards the complicated applications of the internet demand to large-scale calculations ability and mass data processing.Therefore, gridding resource has distributivity and these characteristics of isomerism, and common gridding resource has: computational resource, storage resources, Internet resources, energy resource etc.Along with fast development and its application in grid of Ad Hoc network and wireless sensor network (WSN), energy resource is more and more general in grid heterogeneous environment in recent years.Such as, in Ad Hoc grid, the energy reserves of each resource node all are limited, in realizing the scheduling of resource process, energy constraint is a very important factor.In addition, data-intensive in recent years grid application is on the increase, and such as high-performance physical study, celestial body research, weather forecast etc., these application all are based upon on the high-performance data grid.Characteristics such as it is long that the high-performance data grid has computing time, and energy consumption is big, capacity factor is must not an irrespective problem in carrying out high-performance calculation.Current, along with the continuous increase of people to environmental problem concern dynamics, energy-optimised problem has become a problem being badly in need of solution in the various fields.Therefore, in gridding scheduling research, consider energy constraint, realize that the energetic optimum of scheduling also will become an interior focus of grid computing research field.

The scheduling of resource of grid computing is a np complete problem.Because np problem also can not find effective solution at present, people have proposed some heuristics and have sought its suboptimal solution, as genetic algorithm, and ant group algorithm, Min-Min, tabu search, neural network, simulated annealing scheduling algorithm.Present stage, the research of heuristic dispatching algorithm mainly is divided into two aspects: static scheduling algorithm and dynamic dispatching algorithm.State algorithm is meant that all tasks-mapping resources strategy is definite before scheduling, and dynamic dispatching algorithm is meant that part task-mapping resources strategy is to determine according to actual conditions between schedule periods.Therefore the static scheduling algorithm is relatively simple, and the operation expense is little, and is little to the data dependence, but the static scheduling algorithm is not enough for distribution of resource in the grid environment and isomerism supporting dynamics.Load balance problem, the effect that the isomerism distributivity brings measured problem, task is migrated problem and dynamic dispatching algorithm well solves.Dynamic dispatching algorithm can be divided into online mode (onlinemodel) and batch mode (batch model).These two kinds of methods respectively have relative merits, for online line model, owing to when task arrives, just consider to distribute, as far as possible in time task is dispatched, therefore reaction is fast, the task delay time is short, but may cause the distribution of resource to be optimized inadequately, because the characteristics of task before and after not considering, may cause the low task of requirement to take the strong node of processing power, and demanding Task Distribution is to handling weak node or being in waiting status.Batch processing mode then can be considered more request and resource situation, can obtain more effective gridding resource utilization factor potentially, but for individual task, time delay may be longer, has no idea to implement for some service quality.

The appearance of continuous development, the especially service grid environment of grid application promotes and has quickened research for mesh services quality (QoS).Various dispatching algorithms based on QoS engender that also these dispatching algorithms all are to add various QoS constraint conditions evolutions to form on the basis of original classical dispatching algorithm.The QoS constraint of general Study mainly concentrates on: aspects such as the network bandwidth, network delay, cost, time, survivability, degree of belief.Corresponding improvement algorithm also can solve the key problem in a lot of scheduling processes: deadline optimization, dispatching efficiency optimization, economic expense optimization etc.But, seldom have for the energy-optimised problem in the scheduling process to relate to because the gridding resource type that relates to is more single and the limitation of QoS constraint condition itself.

Summary of the invention

The purpose of this invention is to provide and a kind of energy resource is incorporated in the grid resource scheduling, take all factors into consideration energy constraint and time-constrain based on energy-optimised gridding scheduling method.

To achieve these goals, technical scheme of the present invention is as follows:

1, with the research emphasis of energy resource, introduces the energy initial value, factors such as the network bandwidth as scheduling.

2, realization is energy-optimised, makes the energy consumption values minimum in the scheduling of resource, and the energy consumption in the grid resource scheduling mainly is divided into calculation consumption and network service consumption.

3, consider time span optimization (Makespan), reach the balancing resource load in the energy-optimised process.

4, propose one take all factors into consideration energy constraint and time-constrain based on energy-optimised gridding scheduling method.

Consider in the grid resource scheduling process that gridding task and gridding resource all have distributivity and isomerism, gridding scheduling model of the present invention can not be simulated fully to real grid environment, therefore makes following setting:

(1) each gridding task all is self-existent, and no datat relies on or communication between the task.

(2) energy resource in scheduling model can be realized task computation, functions such as task execution, with sense stricto provide the energy resource of energy supply to have any different.

(3) each resource can only be carried out a gridding task simultaneously.

(4) energy consumption in the gridding scheduling is only limited to that task carry out to consume and network service consumption, and network service consumes the energy consumption of main finger task when communicating by letter with resource data, and the energy consumption of communicating by letter between resource is ignored.

(5) the network service time is the inverse of the network bandwidth, ignores other network factors such as network delay.

(6) energy reserve of each resource is limited, and the energy initial value of different resource is different.

When (7) resource is in idle condition, noenergy consumption.

Below be the definition of some basic parameters among the present invention:

(1) set R={r ₁, r ₂..., r _nThe grid environment of energy resource composition of n isomery of expression.

(2) set T={t ₁, t ₂..., t _mM independently gridding task set of expression.

(3) matrix

ET wherein _IjRefer to energy resource r _jT executes the task _iThe needed execution time, the ETC in this patent obtains by NWS.

(4) G _iExpression task t _iCarry out the primary data value size that the forward power resource is submitted in task.

(5) B _jExpression energy resource r _jThe zero energy value.

(6) E _jRepresentation unit time self-energy resource r _jThe energy consumption values of executing the task and being consumed.

(7) C _jRepresentation unit time self-energy resource r _jThe energy consumption values of communication unit incremental data.

(8) BW _jExpression energy resource r _jThe network bandwidth.

Concrete steps of the present invention are as follows:

First step: for each task among the task-set T it is mapped to each machine among the resource set R, obtain each corresponding cost value Cost (i, j);

Second step: with all resource mark in the resource set is unmarked;

Third step: choose any one the task ti in the task-set, the mapping task is to cost value Cost (i, j) Zui Xiao that Taiwan investment source rj, and (this value is for minimum cost for i, j) value to calculate Cost;

The 4th step: calculate the sufferage value of this mapping, if this sufferage value representation will be paid more cost with duty mapping during to other resources except that rj, the sufferage value equals the difference of minimum cost and time minimum cost;

The 5th step: judge that whether resource rj is for not indicating:

If resource is then deleted task ti for not indicating from task-set T, simultaneously resource rj is denoted as mark;

If be mark, then be mapped to task tk on the resource rj and the sufferage value size of task ti more, if the sufferage value of tk is littler, then tk is put back among the task-set T again, and ti is mapped to rj, simultaneously ti is deleted from task-set T;

The 6th step: repeat third step--the 5th step, till no new task can dispense, finish iterative process one time;

The 7th step: the ready time D that is updated in the resource that has been assigned with new task in this iterative process _iWith the resources left energy value;

The 8th step: repeat second step--the 7th step is finished iteration for several times, and all tasks in task-set are all finished, calculate parameter EC _AvgValue.

Present stage mainly concentrates on optimum span (Makespan) to Study of Grid Resource Scheduling, service quality (Qos), load balancing, aspects such as economic principle.The dispatching algorithm or the scheduling model that propose at these research directions also can both well solve its corresponding problem, but these researchs seldom relate to the problem of energy-optimised aspect.Along with the continuous development of gridding technique, the limited importance of mobile device in grid application of various energy reserves increases gradually, and the grid application of various high energy consumptions also increases gradually.This shows that energy-optimised problem has become a problem that can not be ignored in the grid resource scheduling, also will be a primary study direction of grid resource scheduling to its research.The present invention is energy-optimised from grid resource scheduling, takes all factors into consideration QoS constraint conditions such as energy and time, proposed one a kind of based on energy-optimised gridding scheduling model.This model as the primary study object, has also been taken all factors into consideration the optimal time span in the general dispatching algorithm to the energy-optimised problem in the scheduling of resource simultaneously, problems such as load balance.Energy-optimised scheduling model mainly is divided into energy-optimised module and time-optimized module.These two submodules have proposed prioritization scheme separately respectively from energy and time-constrain.Scheduling model has provided an energy cost function based on these two prioritization schemes, and this cost function can well be weighed the consumption situation of energy in dispatching algorithm.Based on this cost function, the present invention proposes an energy-optimised dispatching method.

Compare with traditional grid resource scheduling, the present invention has the following advantages: 1, have these characteristics of diversity at resource under the isomerous environment, in this scheduling model, add this resource type of energy resource, changed single this limitation of resource type in the traditional resource dispatching algorithm; 2, energy constraint is incorporated in the scheduling of resource, the energy consumption problem in execution of emphasis consideration task and the network service is with energy-optimised research emphasis as this scheduling model; 3, this scheduling model has also been considered time span (Makespan) optimum in the traditional scheduler method, and this has well solved the unbalanced problem of load that is caused in the energy-optimised process; 4, on time span optimization and energy-optimised basis, defined energy cost function,, proposed a kind of based on energy-optimised resource regulating method based on this cost function.

Description of drawings

Fig. 1 is a dispatching method process flow diagram of the present invention.

Embodiment

The present invention is described in further detail below in conjunction with accompanying drawing.

Scheduling model of the present invention is to set up in order to realize the energy-optimised of grid resource scheduling in the grid heterogeneous environment, promptly under the prerequisite that all tasks all will be finished in guaranteeing task-set T in the task implementation, realizes the energy consumption values minimum.Simultaneously, for fear of undue use of low energy consumption resource and the idle load that is caused of high energy consumption resource long-term is unbalanced, time span Makespan optimization problem also is the emphasis of this scheduling model research.Therefore, this scheduling model can be divided into energy constraint and two submodules of time-constrain.

The energy constraint module: the energy consumption in this scheduling model comprises the execute the task energy consumption and the energy consumption of communicating by letter.By the above parameter that the provides t that to go out on missions _iAt resource r _jLast execution energy consumption values EEC _{I, j}:

EEC _i，j＝ET _i，j*E _j.......................................(1)

Task t _iBe assigned to resource r _jLast communication energy consumption figures CEC _{I, j}:

CEC _i，j＝G _i*(1/BW _j)*C _j.................................(2)

Task t _iAt resource r _jOn total power consumption values be EEC _{I, j}With CEC _{I, j}Summation:

EC _i，j＝EEC _i，j+CEC _i，j................................(3)

Resource r _jEnergy consumption values in whole scheduling process is:

EC _j＝∑(EC _i，j)＜＝B _j..................................(4)

Wherein i is at resource r _jThe mission number of the task of last execution.

In this model, have only the gridding resource that has satisfied formula (4) to be only available resources.

The EC that comes out by the above derivation of equation _jJust drawn individual machine (r _j) go up the situation of energy consumption, and therefore the energy-optimised problem in the model is necessary to introduce one and weighs parameter EC towards whole scheduling process _Avg:

${\mathrm{EC}}_{\mathrm{avg}}={\mathrm{\Σ}}_{j=0}^{n-1}({\mathrm{EC}}_j/B_j)/n,\mathrm{rj}\∈R---\left(5\right)$

In formula (5), EC _AvgWhat represent is the average energy consumption ratio, is the criterion of dispatching algorithm energy consumption.EC _AvgMore little, be illustrated in consuming little energy in the whole scheduling process, EC _AvgBig more, the energy consumption in the expression scheduling process is many.Therefore, the energy-optimised problem in this model has become and has asked EC _AvgThe problem of minimum value.

The time-constrain module: time-optimized the same in the time optimal problem in this scheduling model and the most of Grid Resource Schedule Algorithms, mainly realize optimum span Makespan.In this model, Makespan will not be the optimization objects that an emphasis is considered, its introducing also is in order to realize the load balance in the energy resource assigning process.

Setting task t _iAt resource r _jOn the beginning execution time be D _j, t then can go out on missions _iAt machine r _jOn expection deadline C _IjFor:

C _ij＝D _j+ET _ij..........................................(6)

Thus, can draw expection deadline matrix PEC:

$\mathrm{PEC}=\left(\begin{array}{ccc}{C}_{11}& ...& C_{1n}\\ .& & .\\ .& ...& .\\ .& & .\\ C_{m1}& ...& C_{\mathrm{mn}}\end{array}\right)$

Setting task ti has distributed to resource rj and has carried out, and by matrix PEC, can draw actual execution time Ci:

C _i＝C _ij...............................................(7)

Maximum execution time Makespan just equals the difference between maximum task deadline and the scheduling zero-time, promptly begins to carry out a last operation from first operation and is performed the institute's time spent that finishes, and is zero and has if establish zero-time:

Makespan＝Max(Ci)，ti∈T..............................(8)

Makespan is a criterion of heterogeneous computing system, and the fundamental purpose of grid resource scheduling will reduce Makespan exactly.The Makespan value is more little, represents that then the time of dispatching office cost is more little.Therefore the time optimal problem in this model has become the problem of asking the Makespan minimum value.

By analysis to energy constraint and two submodules of time-constrain, can draw the cost function of this model, this Function Synthesis has been considered energy cost and the time cost in the scheduling of resource process:

Obj＝w*EC _avg+(1-w)*Makespan...........................(9)

Wherein, parameter w is used to regulate EC _AvgTwo factors of factor and Makespan factor shared ratio in cost function.Especially, when w equaled 1, the energy consumption cost was only considered in the expression scheduling, and when w equaled 0, time cost was only considered in the expression scheduling.

Grid resource scheduling energy optimizing method based on energy constraint is on the basis of heuristic dispatching algorithm sufferage, adds energy constraint and develops and come.In method, will use cost function quickly:

Cost(i，j)＝w*(EC _i，j)+(1-w)*(Di+ET _ij)..............(10)

This filial generation valency function is used for representing being assigned to the last situation that influences to whole cost function 0bj value of rj as task ti.

The dispatching method execution in step specifically describes as follows:

(1) for each task among the task-set T it is mapped to each machine among the resource set R, obtain each corresponding cost value Cost (i, j).

(2) be unmarked with all resource mark in the resource set.

(3) choose any one task ti in the task-set, the mapping task is to cost value Cost (i, j) Zui Xiao that Taiwan investment source rj, and (this value is minimum cost for i, j) value to calculate Cost.

(4) calculate the sufferage value of this mapping, if this sufferage value representation will be paid more cost with duty mapping during to other resources except that rj, the sufferage value equals the difference of minimum cost and time minimum cost.

(5) judge that whether resource rj is for not indicating:

If resource is then deleted task ti for not indicating from task-set T, simultaneously resource rj is denoted as mark;

If be mark, then be mapped to task tk on the resource rj and the sufferage value size of task ti more, if the sufferage value of tk is littler, then tk is put back among the task-set T again, and ti is mapped to rj, simultaneously ti is deleted from task-set T.

(6) repeat (3)-(5), till no new task can dispense, finish iterative process one time.

(7) be updated in the ready time D of the resource that has been assigned with new task in this iterative process _iWith the resources left energy value.

(8) repeat (2)-(7), finish iteration for several times, all tasks in task-set are all finished, calculate parameter EC _AvgValue.

In iterative process each time, i.e. step (3)-(6), the task number that at every turn can dispatch is not away fixed, and it mainly is subjected to the influence of three factors:

1) total number of machine in the gridding scheduling;

2) the machine number of being competed by a plurality of different tasks;

3) total number of task in the gridding scheduling;

In the worst case, each iterative process can only distribute a task, that is to say and will just can finish whole scheduling of resource process through m iteration.Therefore the time complexity of dispatching method is O (m ²N).

The content that is not described in detail in this instructions belongs to this area professional and technical personnel's known prior art.

Claims (1)

1, a kind of based on energy-optimised gridding scheduling method, its concrete steps are:

First step: for each task among the task-set T it is mapped to each machine among the resource set R, obtain each corresponding cost value Cost (i, j);

Second step: with all resource mark in the resource set is unmarked;

Third step: choose any one the task ti in the task-set, the mapping task is to cost value Cost (i, j) Zui Xiao that Taiwan investment source rj, and (this value is for minimum cost for i, j) value to calculate Cost;

The 4th step: calculate the sufferage value of this mapping, if this sufferage value representation will be paid more cost with duty mapping during to other resources except that rj, the sufferage value equals the difference of minimum cost and time minimum cost;

The 5th step: judge that whether resource rj is for not indicating:

If resource is then deleted task ti for not indicating from task-set T, simultaneously resource rj is denoted as mark;

If be mark, then be mapped to task tk on the resource rj and the sufferage value size of task ti more, if the sufferage value of tk is littler, then tk is put back among the task-set T again, and ti is mapped to rj, simultaneously ti is deleted from task-set T;

The 6th step: repeat third step to the five steps, till no new task can dispense, finish iterative process one time;

The 7th step: the ready time D that is updated in the resource that has been assigned with new task in this iterative process _iWith the resources left energy value;

The 8th step: repeat second step to the, seven steps, finish iteration for several times, all tasks in task-set are all finished, calculate parameter EC _AvgValue.

Images