CN104102532A - Low-energy-consumption-based scientific workflow scheduling method in heterogeneous cluster - Google Patents

Abstract

The invention relates to a low-energy-consumption-based scientific workflow scheduling method in a heterogeneous cluster. The method comprises the following steps: 1) reading configuration file information required by a program; 2) acquiring a task scheduling sequence and calculating task parameters; 3) grouping tasks in the task scheduling sequence and acquiring a task grouping result; 4) mapping the tasks in each group to processors; 5) searching a task with idle time in each processor, and selecting an energy-saving mechanism with most energy conservation amount to perform the task. Compared with the prior art, a dynamic voltage adjustment technology is fully combined with a network energy-saving method, and the total energy consumption of a cluster system is reduced to the greatest extent without influencing system performance.

Description

Research-on-research stream scheduling method based on low energy consumption in a kind of isomeric group

Technical field

The present invention relates to Parallel Scheduling technical field, especially relate to the research-on-research stream scheduling method based on low energy consumption in a kind of isomeric group.

Background technology

Energy consumption problem has become a global problem, and along with the fast development of computer and network technologies, the power managed Yin Qigao operation expense in current high-performance calculation, low reliability, become the focus that people pay close attention to just day by day.How traditional research emphasis always makes system have optimum performance if laying particular emphasis on, be reasonably dispatched on each processor to obtain the shortest overall operation time by task.But in the last few years, increasingly serious along with global energy problem, numerous researchers have just progressively transferred to research emphasis the energy consumption aspect of cluster computing system.

Along with the complexity day by day of problem in science and the continuous expansion of the scale of calculating, scientific workflow (Scientific Workflow) provides the auxiliary scientist of a kind of automatic flow management platform to carry out scientific research and Knowledge Discovery.Job scheduling is a very important problem in scientific workflow technology, and dispatching algorithm directly affects the performance of research-on-research streaming system.Meanwhile, research-on-research streaming system needs high performance computational resource and storage resources, carries out extensive science application program and has consumed huge energy.

Scientific workflow is dispatched with the scheduling of traditional directed acyclic graph (Directed Acyclic Graph, DAG) and is compared, and existing similar part also exists some differences simultaneously.Job scheduling is that research-on-research flow structure is analyzed, and the performance of each resource in dispatch environment is assessed, and met under the condition of scientific workflow operation constraint and user's constraint, operation is reasonably assigned to the process of each resource.A large amount of research organizations conducts in-depth research this problem both at home and abroad, has proposed respectively multiple dispatching algorithm for different application scenarioss, also has some research organizations to sum up scheduling strategy and algorithm.

It is a kind of by the power-saving mechanism of rational method adjusting service voltage that dynamic electric voltage regulates (Dynamic Volgage Scaling, DVS) technology, has been proved to be to design one of effective method of low-power dissipation system.Its basic thought is dynamically to regulate processor to carry out voltage and frequency in the situation that not affecting the normal operation of processor, makes processor not always with ceiling voltage work, thereby plays the object that reduces processor energy consumption.Current most processor is all used the manufacture of CMOS technology, and supports various processor frequency and voltage setting.The power consumption of cmos circuit be proportional to clock frequency and voltage square, i.e. the energy consumption of each clock period be proportional to voltage square.For a task, complete its needed clock period to fix, square being directly proportional of the energy consuming and voltage, just can reduce energy consumption by reducing voltage.Although the linear relationship between clock frequency and voltage, reduce voltage and can turn down clock frequency, the deadline of increase task, but can utilize the free time of task to reduce processor voltage or frequency, under the prerequisite that does not affect other tasks carryings, reduce system energy consumption.In voltage-regulation process, voltage management chip can be carried out small voltage adjustment, and can be within the extremely short time (in tens microseconds) complete the adjustment of voltage.Dynamic electric voltage regulates (DVS) technology to be proved to be to design one of effective method of low-power dissipation system, and has become a kind of power-saving mechanism of comparative maturity.In the last few years, many researchers had successfully applied to DVS mechanism in the middle of the cluster calculating based on energy consumption perception.Although dynamic electric voltage regulation technology has been obtained huge contribution in structure efficiency cluster, it focuses on saving the energy consumption of processor in cluster.

Along with the expansion of network size in group system and the continuous renewal of the network facilities, the problems such as network system high energy consumption, poor efficiency are day by day remarkable.The utilization factor of the network equipment is now all very low, and because the energy consumption of nearly all network equipment is all determined by peak bandwidth, and most equipment is whole day full speed operation, and the network user really need high bandwidth not enough equipment operation of working time 5%, therefore, even the in the situation that of the network free time, the network equipment still according to the standard of peak bandwidth at consumed energy, this has just caused the significant wastage of the energy.Therefore, build GreenNet, reduce network energy consumption and become key issue significant, in the urgent need to address of current computer network field.

Dispatching method major part has in the past carried out improving or optimizing for certain part (system energy consumption comprises processor energy consumption and network energy consumption two parts) of performance or energy consumption.The weak point existing has: the method 1. having is only considered performance and ignored energy consumption completely; Although the method 2. having has not only been considered performance but also has considered the saving of processor energy consumption, has ignored network service energy consumption, even if or considered communication energy consumption, but network delay is not taken into account.Although 3. some algorithm has also been considered network service energy consumption, do not utilize network energy-saving Mode and policy further to save network service energy consumption.

Summary of the invention

Object of the present invention is exactly to provide the research-on-research stream scheduling method based on low energy consumption in a kind of isomeric group in order to overcome the defect that above-mentioned prior art exists, the method is organically combined dynamic electric voltage regulation technology with network energy-saving method fully, reduces to the full extent the total energy consumption of group system in not affecting system performance.

Object of the present invention can be achieved through the following technical solutions:

A research-on-research stream scheduling method based on low energy consumption in isomeric group, comprises the following steps:

1) required profile information of fetch program;

2) obtain task scheduling sequence calculation task parameter;

3) task in task scheduling sequence is divided into groups, obtain task group result;

4) by the duty mapping in each grouping to processor;

5) in each processor, search has the task of free time, and selects the power-saving mechanism that amount of energy saving is maximum to carry out this task.

Step 1) comprise following two steps:

Step 1.1, read in scientific workflow profile information;

Step 1.2, read in parameter configuration files information, comprise processor parameter and network parameter information.

Step 2) comprise following two steps:

Step 2.1, obtain task scheduling sequence;

Step 2.2, calculation task parameter, comprise earliest start time, earliest finish time, optimum processor, best forerunner, allow the deadline and allow at the latest the start time at the latest.

Step 3) comprise the following steps:

Step 3.1, first task from task scheduling sequence start, and carry out depth-first search until task finishes;

Step 3.2, in task search procedure, if the best forerunner of current task does not divide into groups, distributed to the grouping at current task place, and be labeled as and distribute;

Step 3.3, repeating step 3.2, until that all tasks have all been divided into groups is complete.

Step 4) comprise the following steps;

Step 4.1, from first grouping, by the unappropriated processor of all duty mapping in each grouping, and this processor of mark is occupied;

Step 4.2, repeating step 4.1, until each grouping is assigned on processor.

Step 5) comprise the following steps:

Step 5.1, a processor of search, obtain the task on this processor with free time;

Step 5.2, judge that whether this task can be undertaken by DVS mechanism or network speed zoom mechanism energy-conservationly, if can be undertaken by any one mechanism energy-conservationly, selects the maximum mechanism of amount of energy saving to carry out; If only can be undertaken by a kind of mechanism wherein energy-conservationly, select this mechanism to carry out;

Step 5.3, repeating step 5.1 and step 5.2, until all tasks on each processor are searched complete.

Compared with prior art, the present invention is directed to the energy-optimised problem of scientific workflow scheduling in isomeric group, DVS technology and network energy-saving technology are combined, optimization system energy consumption in meeting system resource constraints and system performance requirement.First, in the present invention, the dispatching algorithm based on copying is obtained near-optimization grouping according to system environments (mainly comprising that processor parameter is connected parameter with network); Then, each grouping is dispatched to respectively on a unappropriated processor, and the free time producing by the dependence between exploration task, utilize DVS technology dynamically to adjust processor voltage to save the idle energy consumption of processor, or utilize network energy-saving Mode and policy to be optimized network service energy consumption.

Brief description of the drawings

Fig. 1 is process flow diagram of the present invention.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.

Embodiment

The invention provides the research-on-research stream scheduling method based on low energy consumption in a kind of isomeric group.Wherein, scientific workflow adopts directed acyclic graph to represent that the modeling method of scientific workflow is more conventional modeling method in scheduling research.In the present invention, the set of tasks of scientific workflow can be expressed as G (V, E, t, c), wherein V={v by the directed acyclic graph of a sideband weight ₁, v ₂..., v _nrepresent that task-set, E represent that the set of all directed edges is message set, and t represents the computing time of task on processor, and e represents the dependence between task, and the weight that c is limit represents two call duration times between task.Isomeric group due to what the present invention is directed to, thus for the concrete computing time of the dissimilar each task of processor be different, but execution time on same type processor equates completely.At this, we use t _iwexpression task v _iat processor p _won computing time, use e _ij={ v _i, v _j∈ E represents from task v _ibe transferred to v _jmessage, i.e. v _iv _jpredecessor task, and transmit the required time c of this message _ijrepresent, and a task only after being all finished, its all predecessor tasks can start to be performed.In real network, call duration time c _ijrepresent that message is from sending to the received time interval, it is mainly made up of transmission of messages time and message delay.M in above-mentioned model and n represent respectively the task number in processor number and the task-set using in cluster, i, and j, k all represents mission number, w, p, q all represents processor numbering, therefore 1≤i, j, k≤n, 1≤p, q, w≤m.

The processor node that computing system in an isomeric group has different disposal ability by several forms, and is designated as P={p ₁, p ₂..., p _m, these nodes are connected with each other by express network.In the present invention, adopt identical network for connecting each processor node, i.e. network connection is isomorphism.All processors in system are all supported DVS technology, suppose processor p _ithere is h _iindividual discrete voltage, can be expressed as here for non-ascending order is arranged, its corresponding frequency also be like this.The number of all types of processor discrete voltage determines by itself voltage, and the discrete voltage number of same type processor is identical.

The method specifically comprises several steps as shown in Figure 1:

One, required profile information of fetch program

Step 1.1, read in scientific workflow profile information;

Step 1.2, read in parameter configuration files information, comprise processor parameter and network parameter information.

Two, obtain task scheduling sequence and calculation task parameter

Step 2.1, obtain task scheduling sequence queue; For parallel task collection V, total n=|V| task, from export task, calculate each task priority until entrance task finish.Then arrange according to task priority size ascending order, obtain initiating task scheduling sequence.

At this, we defined notion bottom is expressed as the maximal value of the total computing time from current task to export task, and this value is in order to meet the priority constraint relationship between task.Calculate when bottom according to the original calculation time, and do not consider the intertask communication time.Task v _jbottom value be defined as follows:

Wherein succ (v _j) expression task v _jsuccessor set, t _jexpression task v _jmaximum execution time on all processors.Obtaining after the bottom value of all tasks, just can obtain an initiating task scheduling sequence queue who arranges according to bottom value ascending order.

Step 2.2, calculation task parameter, comprise earliest start time (EST), earliest finish time (ECT), optimum processor (fproc), best forerunner (FP), allow at the latest the deadline (LACT), allow at the latest the start time (LAST).

Task v _joptimum processor can calculate according to following formula:

$\underset{w\&Element;P,q\&Element;P}{\mathrm{fproc}\left(j\right)}=w,s.t.\left(\begin{array}{c}(\mathrm{ECT}(k/w)+t_{\mathrm{jw}})\\ \underset{e_{\mathrm{kj}}\&Element;E,w=\mathrm{fproc}\left(k\right),q\&NotEqual;\mathrm{fproc}\left(k\right)}{<(\mathrm{ECT}(k/q)+c_{\mathrm{kj}}+t_{\mathrm{jq}})}\end{array}\right)---\left(2\right)$

Task v _jearliest start time (EST) recursive calculation from top to bottom on each processor, wherein the earliest start time of entrance task is 0.Computing formula is: otherwise (3)

Task v _jearliest finish time (ECT) equal task v _iactual earliest start time add task v _jexecution time in its optimum processor, be shown below.

ECT(j)＝EST(j)+t _jfproc(j). (4)

Task v _jbest forerunner be defined as follows shown in formula:

$\underset{e_{\mathrm{kj}}\&Element;E,e_{\mathrm{lj}}\&Element;E}{\mathrm{FP}\left(j\right)=k,s.t.}\left(\begin{array}{c}\underset{\mathrm{fproc}\left(k\right)=w}{(\mathrm{ECT}\left(k\right)\&GreaterEqual;\mathrm{ECT}\left(l\right)+c_{\mathrm{lj}})\left|\right|}\\ \underset{\mathrm{fproc}\left(k\right)\&NotEqual;w,\mathrm{fproc}\left(j\right)=w}{(\mathrm{EST}\left(k\right)+t_{\mathrm{kw}}\&GreaterEqual;\mathrm{ECT}\left(l\right)+c_{\mathrm{lj}})}\end{array}\right)---\left(5\right)$

Task v _jthe deadline of permission at the latest: otherwise (6)

Task v _jthe start time of permission at the latest:

LAST(j)＝LACT(j)-t _jfproc(j) (7)

First, from top to down DAG figure is traveled through calculation task v _jat processor p _won earliest start time EST (j/w).Then, the v that can go out on missions _jearliest start time EST (j) in its optimum processor, and the ECT on earliest finish time (j) of calculation task.The above calculation of parameter of current task out after, we can draw current task v _joptimum processor fproc (j).Next, calculation task v _jbest forerunner FP (j), this parameter means task and best forerunner thereof is dispatched on identical processor and can obtains less parallel time; Finally, calculation task v _jthe deadline LACT of permission at the latest (j) and allow at the latest start time LAST (j).

Three, obtain group result

Step 3.1, first task from task scheduling sequence queue start, and carry out depth-first search until task finishes;

In step 3.2, task search procedure, if current task v _ibest forerunner (FP (i)) do not divide into groups, distributed to the grouping at current task place, and be labeled as and distribute;

Step 3.3, repeating step 3.2, until that all tasks have all been divided into groups is complete.

Four, task packet map

Step 4.1, from first grouping, by the unappropriated processor of all duty mapping in each grouping, and this processor of mark is occupied;

Step 4.2, repeating step 4.1, until each grouping is assigned with on a processor.

Five, Energy Saving Strategy is selected

On step 5.1, a processor of search, there is the task of free time;

Step 5.2, for a task with free time, judge that whether this task can be undertaken by DVS mechanism or network speed zoom mechanism energy-conservationly, if can be undertaken by any one mechanism energy-conservationly, selects the mechanism that amount of energy saving is maximum to carry out; If only can be undertaken by a kind of mechanism wherein energy-conservationly, select this mechanism to carry out.

Detailed process is: first judge whether this task can be energy-conservation by DVS mechanism, if can, need to judge whether this task can be energy-conservation by network speed zoom mechanism, if can, calculate respectively the amount of energy saving and the amount of energy saving that uses network speed zoom mechanism that use DVS mechanism, choose the power-saving mechanism that amount of energy saving is maximum and carry out.

If this task is can not be by network speed zoom mechanism energy-conservation or what select is that to utilize dynamic electric voltage regulation mechanism to carry out energy-conservation, utilize DVS mechanism under the prerequisite that does not affect other task execution times by this tasks carrying frequency and lower voltage to near-optimization value; If this task is can be by network speed zoom mechanism energy-conservation and what select is that to utilize network speed zoom mechanism to carry out energy-conservation, utilize speed convergent-divergent network energy-saving mechanism, under the prerequisite that does not affect other task execution times, the message transmission rate between this task and its subsequent tasks is reduced to near-optimization value.

When searching a task v with free time _iafter, first calculate v _icalculating energy consumption compEN1[v in the time not carrying out task power-economizing method _i] and and unallocated subsequent tasks to same processor between communication energy consumption commEN1[v _i] [v _j], then utilize DVS technique computes v _icalculating energy consumption compEN2[v after execution DVS is energy-conservation _i], utilize network speed zoom mechanism to calculate v _icarry out the communication energy consumption commEN2[v after network energy-saving _i] [v _j]; Make compEN3[v at this _i]=compEN1[v _i]-compEN2[v _i], the amount of energy saving of DVS technology is carried out in representative; Make commEN3[v _i] [v _j]=commEN1[v _i] [v _j]-commEN2[v _i] [v _j], the amount of energy saving of network speed convergent-divergent strategy is carried out in expression; Then judge compEN3[v _i] whether be greater than commEN3[v _i] [v _j].If compEN3[v _i] be greater than commEN3[v _i] [v _j], energy-conservation energy-conservation many than carrying out network speed convergent-divergent after explanation execution DVS, therefore utilize DVS technology to carry out energy-conservation, otherwise the energy-saving effect of explanation execution network speed convergent-divergent is better than DVS technology, therefore carries out network speed convergent-divergent strategy.Specific formula for calculation and the computing method of relevant energy consumption are as follows:

(1) calculate energy consumption

At present, circuit in processor adopts CMOS integrated circuit mostly, its energy consumption mainly comprises static energy consumption and dynamic energy consumption, but because static energy consumption institute's point ratio in processor total energy consumption is less, so the main object of the energy-conservation research of dynamic energy consumption Chang Zuowei, therefore the present invention calculates when energy consumption model the only dynamic energy consumption of considering processor building.

The present invention adopts DVS Techniques For Reducing processor energy consumption, because processor all has several different voltage and frequency level, therefore dispatching algorithm can fully select most suitable execution voltage and frequency to come energy-conservation.DVS mechanism be by the frequency adapting and voltage distribution to a series of processors, thereby the free time of eliminating tasks carrying to reach save the object of idle energy consumption.The power consumption of processor can be following formula with given Parametric Representation:

$P=\mathrm{\&alpha;}\&CenterDot;C_L\&CenterDot;V_{\mathrm{dd}}^2\&CenterDot;f_{\mathrm{clk}}---\left(8\right)$

Wherein, α represents switching rate, C _lrepresent load capacitance, V _ddrepresent supply voltage, f _clkrepresent clock frequency.What need to consider at this is how under the prerequisite that does not affect system performance, to select most suitable processor to carry out frequency.The object that dynamic electric voltage regulates must be the task with free time on the non-critical path in DAG, and for being present in critical path of task, its execution time is can not be reformed.DVS technology of the present invention is that the execution frequency/voltage of the task with free time is adjusted to certain near-optimization value.If an available free time of task is utilized DVS technology the execution frequency of this task will be reduced to near optimal and carries out frequency, otherwise fixing under the highest frequency and voltage, this task one carries out.Task v _iat processor p _won activity energy consumption calculation as shown in the formula:

${\mathrm{en}}_{\mathrm{iw}}={\mathrm{PN}}_{\mathrm{highest}-w}\&CenterDot;\frac{f_{\mathrm{iw}}\&CenterDot;V_{\mathrm{iw}}^2}{f_{\mathrm{highest}-w}\&CenterDot;V_{\mathrm{highest}-w}^2}\&CenterDot;{\mathrm{\&tau;}}_{\mathrm{iw}}---\left(9\right)$

Therefore, the formula that is calculated as follows of the total activity energy consumption of all processors:

${\mathrm{EN}}_{\mathrm{active}}=\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{PN}}_{\mathrm{highest}-w}\&CenterDot;\frac{f_{\mathrm{iw}}\&CenterDot;V_{\mathrm{iw}}^2}{f_{\mathrm{highest}-w}\&CenterDot;V_{\mathrm{highest}-w}^2}\&CenterDot;{\mathrm{\&tau;}}_{\mathrm{iw}}\&CenterDot;x_{\mathrm{iw}}---\left(10\right)$

The formula that is calculated as follows of the total idle energy consumption of all processors:

${\mathrm{EN}}_{\mathrm{idle}}=\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}{\mathrm{PN}}_{\mathrm{idle}-w}\&CenterDot;(\mathrm{makespan}-\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}({\mathrm{\&tau;}}_{\mathrm{iw}}\&CenterDot;x_{\mathrm{iw}}))---\left(11\right)$

Wherein x _ikmapping relations between expression task and processor, are defined as follows:

Finally, total calculating energy consumption can be calculated by following formula:

EN＝EN _active+EN _idle (13)

PN in above formula _highest-wand PN _idle-wrepresent respectively processor p _kthe highest activity power consumption and idle power consumption, f _highest-wand V _highest-wrepresent respectively processor p _won the highest execution frequency and carry out voltage, f _iwand V _iwdo not represent task v _iat processor p _won the best carry out frequency and carry out voltage, τ _iwexpression task v _iat processor p _won actual execution time, makespan represents that the time of being finished of workflow application is scheduling length.

(2) communication energy consumption

The communication energy consumption of network is mainly divided into two parts, and a part is the energy consumption while maintaining network equipment work itself, and referred to herein as the static energy consumption of equipment, another part is the data communication energy consumption of network equipment working time, referred to herein as devices communicating energy consumption.In the present invention, we do not consider the static energy consumption of the network equipment, and the emphasis of network energy consumption research is the data communication energy consumption in the time of network equipment working, and therefore, the network energy consumption model that the present invention adopts is mainly to set up for the data communication energy consumption of the network equipment.

In the present invention, interconnection network are isomorphisms, and network all has identical data transmission capabilities and power of communications.The present invention adopts continuous stream speed zoom model to do and reasonably evaluate and save communication energy consumption.Data in this model are to stablize and continuous form transmission, and the energy consuming in network can regulate by the data traffic in route network, also can regulate by the transmission speed reducing on each link, and in the present invention, we take the latter.

Suppose all corresponding energy consumption function f of arbitrary link e _e(s) energy being consumed when, it represents this link with speed s transmission data.Each link can be selected according to the transport load situation of network its actual transfer rate s in [0, R] speed interval, and 0≤s≤R is special circumstances in the time of R=∞.In following energy consumption model, R is bandwidth.

The network energy consumption function model that the present invention uses is as follows:

This is a kind of function model with Start-up costs, and wherein e represents any network linking, σ _erepresent the Start-up costs of energy consumption function.With respect to the energy consumption function model of tape starting expense not, more meet real network environment and actual conditions with the function model of Start-up costs.

Owing to being assigned between two tasks of same processing node, call duration time is negligible, and to be therefore assigned on different processor the total communication energy consumption computing formula between task as follows for the communication energy consumption of whole network:

$\underset{e_{\mathrm{ij}}\&Element;E}{\mathrm{EL}}=\underset{i=1}{\overset{n-1}{\mathrm{\&Sigma;}}}\underset{j=i+1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{e_{\mathrm{ij}}\&Element;E}{({\mathrm{in}}_{\mathrm{ij}}\&CenterDot;f_e\left(s_{\mathrm{ij}}\right))}---\left(15\right)$

Wherein, parameter in _ijbe defined as follows formula:

Finally, the total energy consumption of whole system can be expressed as:

E _total＝EN+EL (17)

Step 5.3, repeating step 5.1 and step 5.2, until all tasks on each processor are searched complete.

Claims (6)

1. the research-on-research stream scheduling method based on low energy consumption in isomeric group, is characterized in that, comprises the following steps:

1) required profile information of fetch program;

2) obtain task scheduling sequence calculation task parameter;

3) task in task scheduling sequence is divided into groups, obtain task group result;

4) by the duty mapping in each grouping to processor;

5) in each processor, search has the task of free time, and selects the power-saving mechanism that amount of energy saving is maximum to carry out this task.

2. the research-on-research stream scheduling method based on low energy consumption in a kind of isomeric group according to claim 1, is characterized in that step 1) comprise following two steps:

Step 1.1, read in scientific workflow profile information;

Step 1.2, read in parameter configuration files information, comprise processor parameter and network parameter information.

3. the research-on-research stream scheduling method based on low energy consumption in a kind of isomeric group according to claim 1, is characterized in that step 2) comprise following two steps:

Step 2.1, obtain task scheduling sequence;

Step 2.2, calculation task parameter, comprise earliest start time, earliest finish time, optimum processor, best forerunner, allow the deadline and allow at the latest the start time at the latest.

4. the research-on-research stream scheduling method based on low energy consumption in a kind of isomeric group according to claim 1, is characterized in that step 3) comprise the following steps:

Step 3.1, first task from task scheduling sequence start, and carry out depth-first search until task finishes;

Step 3.2, in task search procedure, if the best forerunner of current task does not divide into groups, distributed to the grouping at current task place, and be labeled as and distribute;

Step 3.3, repeating step 3.2, until that all tasks have all been divided into groups is complete.

5. the research-on-research stream scheduling method based on low energy consumption in a kind of isomeric group according to claim 1, is characterized in that step 4) comprise the following steps:

Step 4.1, from first grouping, by the unappropriated processor of all duty mapping in each grouping, and this processor of mark is occupied;

Step 4.2, repeating step 4.1, until each grouping is assigned on processor.

6. the research-on-research stream scheduling method based on low energy consumption in a kind of isomeric group according to claim 1, is characterized in that step 5) comprise the following steps:

Step 5.1, a processor of search, obtain the task on this processor with free time;

Step 5.2, judge that whether this task can be undertaken by DVS mechanism or network speed zoom mechanism energy-conservationly, if can be undertaken by any one mechanism energy-conservationly, selects the maximum mechanism of amount of energy saving to carry out; If only can be undertaken by a kind of mechanism wherein energy-conservationly, select this mechanism to carry out;

Step 5.3, repeating step 5.1 and step 5.2, until all tasks on each processor are searched complete.