CN104333569A - Cloud task scheduling algorithm based on user satisfaction - Google Patents

Abstract

Disclosed is a cloud task scheduling algorithm based on user satisfaction. The algorithm mainly comprises several phases including parameter normalization of virtual machines and cloud tasks, European style distance calculation, resource selection, European style distance updating, calculation of the user satisfaction of the tasks, calculation of the resource application cost of a single cloud task and the total system cost after all the tasks are executed, and the like. According to the invention, the tasks are distributed to the most suitable resources from the perspective of users, such that the requirements of users for multiple aspects including a CPU, finishing time, bandwidth and the like are better satisfied, and at the same time, the resource application cost of the users is effectively decreased. In cloud calculation, what the user concerns is whether their paid cost reasonably matches obtained service quality, and the demands of the users are satisfied to a quite high degree. Compared to the prior art, the algorithm gives an effective strategy to enable the users to obtain better service quality.

Description

Based on the cloud task scheduling algorithm of user satisfaction

Technical field

The present invention relates to cloud task scheduling algorithm.

Background technology

Cloud computing is the comprehensive development of parallel computation, Distributed Calculation and grid computing, it is a kind of business computation model, on the resource pool that it can be formed the task distribution that in the past need high-performance computer just can complete at a large amount of inexpensive computers, various application system is enable to obtain computing capability, memory space and information service as required.But, while realizing these services, need consideration problem, the use of namely different users to cloud computing resources has different demands, as CPU, internal memory, deadline, bandwidth, cost of use etc., how user is allowed to obtain better service quality by a kind of effective strategy.The task scheduling algorithm of cloud computing is one of approach solved the problem.

Traditional task scheduling algorithm focuses on the efficiency of server, such as, be the method for scheduling task of target with optimal finish time, complete efficiency preferably although have, the resource utilization that computing capability is strong may be caused high, make system load unbalanced; The bandwidth that load-balancing algorithm can provide effective method to come extended network equipment and server, the throughput that increases, Strengthens network data-handling capacity, the flexibility improving network and availability, but, traditional task scheduling algorithm all have ignored the QoS requirement of user task, can not well distribute according to need to resource.

Summary of the invention

The dispatching algorithm of many classics is defined in the research process to cloud task scheduling technique, their many angles from cloud resource provider, consider optimal finish time, lowest energy consumption, node load balancing, Resource Availability and the parameter such as reliability, system availability, and the algorithm that the present invention proposes is emphatically from the angle of user, consider that matching degree, the user of task completion time, cost, cost and service quality use the parameters such as the satisfaction of resource, have also contemplated that the load balancing of system simultaneously.

Cloud computing uses Intel Virtualization Technology to be encapsulated with the form of virtual machine by the physical resource of bottom, allows virtual machine to perform the task of user.Scheduling problem be by the task of user with certain optimization aim for principle and resource map, cloud computing simplifies mating of task and resource, required by task resource is embodied with the form of a virtual machine, so be converted into the search of resource, a certain virtual machine is searched for.

In order to realize dispatching algorithm, first the present invention is described cloud task, virtual machine and classification of task:

● virtual machine seven element group representations:

vm _i＝<id _i,peNum _i,ram _i,bw _i,C _cpu/num,C _mem/MB,C _bw/Mbps> (1)

Seven tuples represent the unit price of ID, CPU number of virtual machine, internal memory, bandwidth and CPU, internal memory and bandwidth respectively.

● cloud task eight element group representations:

t _i＝<id _i,type _i,len _i，exppe _i,expram _i,expbw _i,s _i,cost _i> (2)

Eight tuples represent the ID of cloud task, type, task size respectively, expect CPU number, expect internal memory, the user satisfaction expecting bandwidth, task and the cost of executing the task.

● cloud task type: the present invention mainly considers following qos parameter:

A) deadline: for the cloud task of requirement of real-time, need to complete within the least possible time, corresponding is exactly CPU and these two resources of execution speed with it.

B) bandwidth: when cloud task requires higher to communication bandwidth, such as media stream demand, needs prioritize bandwidth requirement.

C) internal memory: when cloud task is higher to request memory, need to pay the utmost attention to memory requirements.

For different cloud mission requirements, user satisfaction is weighed according to different qos parameters, for this reason, the present invention devises a weight vectors, it illustrates the value degree of recognition of cloud platform for different resource, use weight vectors to adjust the Performance Ratio parameter selecting resources of virtual machine, better improve with this satisfaction that user uses resource.Such as, for real-time or concerning the cloud task of time-sensitive, it is desirable to finish the work with the minimum deadline, therefore need the resource that computing capability is strong, so give the larger weight of CPU.If the weight vectors of the i-th generic task is expressed as:

ei＝[ei ₁,ei ₂,ei ₃] (3)

Wherein ei ₁, ei ₂, ei ₃the weight of corresponding CPU, internal memory, bandwidth respectively, and

${\mathrm{\&Sigma;}}_{j=1}^3{\mathrm{ei}}_j=1---\left(4\right)$

Thought based on the cloud task scheduling algorithm of user satisfaction is as follows: for the cloud task that a pile is given, select the cloud task that in system, current priority is the highest, parameter normalization is carried out to it, then by the cloud task after this normalization and all virtual machines in system (virtual machines all in advance parameter normalization) compute euclidian distances, during compute euclidian distances, according to the type of cloud task and system to the value degree of recognition of parameters, give the weight that different performance parameters is different, current cloud task is tied to Euclidean distance be worth on minimum virtual machine.Load in order to equalizing system ensures the deadline of all tasks simultaneously, when after the bound cloud task of virtual machine, upgrades its Euclidean distance list, reduces next cloud task and is assigned to possibility on identical virtual machine.After current task executes, calculate its user satisfaction and resource use cost, after all tasks carryings are complete, calculate all comprehensive satisfactions of cloud task and the total cost of system.

The present invention needs the technical scheme protected to be characterized by:

Based on a cloud task scheduling algorithm for user satisfaction, it is characterized in that, comprise the steps:

Step 1, resource parameters normalization.

The performance parameter of task and virtual machine is all normalized to [0,1] interval, order set X _ij={ X _1j..., X _tjj is the number of performance parameter, be the set of the similar performance parameter of virtual machine, its normalized value is:

GX _ij＝(curX _ij-minX _ij)/(maxX _ij-minX _ij) (5)

Wherein, i is the number of task, and j is the number of performance parameter, curX _ijfor performance currency, minX _ijfor the minimum value in similar performance parameter sets, maxX _ijfor the maximum in similar performance parameter sets.

Step 2, calculates Euclidean distance.

Be X={X at the parameter vector of virtual machine after parameter normalization process ₁, X ₂, X ₃, the parameter vector of cloud task is Y={Y ₁, Y ₂, Y ₃.Consider CPU, internal memory and bandwidth three performance parameters, obtain weight vectors W={W according to the type of cloud task ₁, W ₂, W ₃.Then the computing formula of Euclidean distance is:

$D_i=\sqrt{{\mathrm{\&Sigma;}}_{j=1}^3{W}_j*{(X_j-Y_j)}^2}---\left(6\right)$

Wherein X _jrepresent the normalized parameter value of a jth resource in i-th virtual machine; Y _jexpression task is to the expected value of jth kind resource; W _jrepresent the weight of jth kind resource.

Step 3, selects resource.

Each task choosing and the minimum virtual machines performing tasks of its Euclidean distance, the method controlling idle virtual machine is adopted to carry out load balance, each virtual machine safeguards an Euclidean distance table, after certain task is successfully assigned to certain virtual machine execution, need to upgrade Euclidean distance table, increase the Euclidean distance between certain class resource and this virtual machine, more new formula is:

D _i'=D _i(1+1/n), n is the number (7) of virtual machine

Step 4, calculates user satisfaction.

After task completes, consider the performance of each task, comprise the deadline of task and the user satisfaction of each task, and the comprehensive satisfaction of all tasks.The user satisfaction of individual task is:

$s_i={\mathrm{\&Sigma;}}_{j=1}^3{W}_j\ln ({\mathrm{act}}_j/{\exp }_j)---\left(8\right)$

Wherein, s _ifor the user satisfaction of task i; W _jfor the weight of jth item performance parameter; act _jfor task is to the actual consumption of jth item performance parameter; exp _jfor cloud task is to user's expected value of jth item performance.

When 0≤| s _i| when≤0.5, then think that the Resourse Distribute of user to cloud task i is felt quite pleased; Work as 0.5<|s _i| when≤1, then think that the Resourse Distribute of user to cloud task i is satisfied; When | s _i| during >1, then think that user is unsatisfied with the Resourse Distribute of cloud task i; When | s _i| value very large time, then think that user is very dissatisfied to the Resourse Distribute of cloud task i.

User's comprehensive satisfaction of all cloud tasks is:

$S={\mathrm{\&Sigma;}}_{i=1}^t\left|s_i\right|---\left(9\right)$

Wherein s _iit is the user satisfaction of i-th task; T is the number of all cloud tasks.In cloud computing system, the value of S is less, illustrates that the satisfaction of the service that all users of this system and cloud computing service provider provide is higher.

Step 5, assesses the cost.

After executing each cloud task, calculate spent cost of executing the task.Virtual machine according to unit to resource charging, task consumption full payment cost _ifor:

${\cos t}_i=C_{\mathrm{cpu}/\mathrm{num}}*P_1+C_{\frac{\mathrm{mem}}{\mathrm{MB}}}*P_2+C_{\mathrm{bw}/\mathrm{Mbps}}*P_3---\left(10\right)$

Wherein, P _ifor resource quantity, C is unit resource price.

After performing all cloud tasks, the total cost of system is:

$C={\mathrm{\&Sigma;}}_{i=1}^t{\cos t}_i---\left(11\right)$

Wherein cost _iit is the cost of i-th task; T is the number of all cloud tasks.In cloud computing system, the value of C is less, illustrates that this system performs the cost that all cloud tasks spend less.

Algorithm of the present invention, from the angle of user, by task matching in most suitable resource, better meets user to many-sided demands such as CPU, deadline, bandwidth, effectively reduces the cost that user uses resource simultaneously.In cloud computing, for user it is of concern that the cost paid and the service quality that obtains whether Proper Match, make the demand of user obtain higher degree and meet.Compared with prior art, The present invention gives a kind of effective strategy allows user obtain better service quality.

Accompanying drawing explanation

Fig. 1 is based on the cloud task scheduling algorithm flow chart of user satisfaction.

Fig. 2 is based on the simulation result of the cloud task scheduling algorithm of user satisfaction.

The simulation result of Fig. 3 optimal finish time dispatching algorithm.

Fig. 4 task completion time contrasts.

Fig. 5 task user satisfaction contrasts.

Fig. 6 tasks carrying Cost comparisons.

Embodiment

In cloud computing, it is not the performance being concerned about very much system for user, they are it is of concern that the cost paid and the service quality that obtains whether Proper Match, for making the demand of user obtain the satisfied consideration of higher degree, the present invention proposes a kind of cloud task scheduling algorithm based on user satisfaction.This algorithm, from the angle of user, by task matching in most suitable resource, better meets user to many-sided demands such as CPU, deadline, bandwidth, effectively reduces the cost that user uses resource simultaneously.Finally, the present invention uses CloudSim platform to be emulated by algorithm, and algorithm and optimal finish time dispatching algorithm comparatively conventional is at present contrasted, the validity of verification algorithm in user satisfaction and resource use cost.

Below in conjunction with the algorithm flow chart of accompanying drawing, the present invention is further illustrated.

Fig. 1 is algorithm flow chart of the present invention.As shown in the figure, algorithm mainly comprises the parameter normalization of virtual machine and cloud task, the calculating of Euclidean distance, resource selection, Euclidean distance upgrade, the user satisfaction of calculation task, calculates several stage such as the resource use cost of single cloud task and the total cost of the complete rear system of all tasks carryings.

Step 1, resource parameters normalization.In order to carry out the calculating of Euclidean distance more easily, the performance parameter of task and virtual machine is all normalized to [0,1] interval by the present invention, order set X _ij={ X _1j..., X _tjj is the number of performance parameter, be the set of the similar performance parameter of virtual machine, its normalized value is:

GX _ij＝(curX _ij-minX _ij)/(maxX _ij-minX _ij) (5)

Wherein, i is the number of task, and j is the number of performance parameter, curX _ijfor performance currency, minX _ijfor the minimum value in similar performance parameter sets, maxX _ijfor the maximum in similar performance parameter sets.

Step 2, calculates Euclidean distance.Be X={X at the parameter vector of virtual machine after parameter normalization process ₁, X ₂, X ₃, the parameter vector of cloud task is Y={Y ₁, Y ₂, Y ₃.The present invention mainly considers CPU, internal memory and bandwidth three performance parameters, obtains weight vectors W={W according to the type of cloud task ₁, W ₂, W ₃.Then the computing formula of Euclidean distance is:

$D_i=\sqrt{{\mathrm{\&Sigma;}}_{j=1}^3{W}_j*{(X_j-Y_j)}^2}---\left(6\right)$

Wherein X _jrepresent the normalized parameter value of a jth resource in i-th virtual machine; Y _jexpression task is to the expected value of jth kind resource; W _jrepresent the weight of jth kind resource.

Euclidean distance between task and virtual machine is less, illustrates that this virtual machine of task choosing can obtain user satisfaction relatively preferably, also can better meet the value need for approval of cloud computing provider to different resource simultaneously.

Step 3, selects resource.Each task choosing and the minimum virtual machines performing tasks of its Euclidean distance, but consider the problem of load balancing of system, avoid all tasks to be assigned on a very capable large virtual machine simultaneously, simultaneously in order to meet the optimization of task completion time as far as possible, the method controlling idle virtual machine is adopted to carry out load balance, so each virtual machine safeguards an Euclidean distance table, after certain task is successfully assigned to certain virtual machine execution, need to upgrade Euclidean distance table, increase the Euclidean distance between certain class resource and this virtual machine, more new formula is:

D _i'=D _i(1+1/n), n is the number (7) of virtual machine

Resource selection process is as follows:

1.For i＝1 to m

2 Select VM by parameter of t _ito VM _i; (the identical performance composition of vector of all tasks)

3 For i＝1 to t

4 For j＝1 to 3

5 Compute GX _ij(parameter normalization process);

6 For i＝1 to t

The Euclidean distance D of 7 calculation tasks and virtual machine _i;

8 Select min D _i；

9 Bind t _ito VM which has the min D _i；

10 End；

Step 4, calculates user satisfaction.After task completes, need the performance considering each task.Comprise the deadline of task and the user satisfaction of each task, and the comprehensive satisfaction of all tasks.The user satisfaction of individual task is:

$s_i={\mathrm{\&Sigma;}}_{j=1}^3{W}_j\ln ({\mathrm{act}}_j/{\exp }_j)---\left(8\right)$

Wherein, s _ifor the user satisfaction of task i; W _jfor the weight of jth item performance parameter; act _jfor task is to the actual consumption of jth item performance parameter; exp _jfor cloud task is to user's expected value of jth item performance.

When 0≤| s _i| when≤0.5, then think that the Resourse Distribute of user to cloud task i is felt quite pleased; Work as 0.5<|s _i| when≤1, then think that the Resourse Distribute of user to cloud task i is satisfied; When | s _i| during >1, then think that user is unsatisfied with the Resourse Distribute of cloud task i; When | s _i| value very large time, then think that user is very dissatisfied to the Resourse Distribute of cloud task i.

User's comprehensive satisfaction of all cloud tasks is:

$S={\mathrm{\&Sigma;}}_{i=1}^t\left|s_i\right|---\left(9\right)$

Wherein s _iit is the user satisfaction of i-th task; T is the number of all cloud tasks.In cloud computing system, the value of S is less, illustrates that the satisfaction of the service that all users of this system and cloud computing service provider provide is higher.

Step 5, assesses the cost.After executing each cloud task, need to calculate spent cost of executing the task.Virtual machine according to unit to resource charging, therefore, task consumption full payment cost _ifor:

${\cos t}_i=C_{\mathrm{cpu}/\mathrm{num}}*P_1+C_{\frac{\mathrm{mem}}{\mathrm{MB}}}*P_2+C_{\mathrm{bw}/\mathrm{Mbps}}*P_3---\left(10\right)$

Wherein, P _ifor resource quantity, C is unit resource price.

After performing all cloud tasks, the total cost of system is:

$C={\mathrm{\&Sigma;}}_{i=1}^t{\cos t}_i---\left(11\right)$

Wherein cost _iit is the cost of i-th task; T is the number of all cloud tasks.In cloud computing system, the value of C is less, illustrates that this system performs the cost that all cloud tasks spend less.

Step 6, algorithm simulating.The cloud environment of the present invention's simulation is made up of 5 virtual machine node, node sets up Resource Broker, and sets up 10 analog subscriber tasks.Paper examines the following aspects in emulation: deadline of all cloud tasks, the user satisfaction of single cloud task, the comprehensive satisfaction of all tasks, the executory cost of individual task, system perform the cost of all cloud tasks, in this emulation, the unit price of resource is 1 while using (actual can arrange voluntarily).Fig. 2 is the simulation result of the cloud task scheduling algorithm based on user satisfaction.Fig. 3 is the simulation result of optimal finish time dispatching algorithm.

As can be seen from Figure 4, in optimal finish time algorithm, what finally complete is task 3, and total deadline is 334.62ms, and based in the cloud task scheduling algorithm of user satisfaction, what finally complete is task 7, and total deadline is 389.92ms.Dispatching algorithm of the present invention is not so good as classical optimal time dispatching algorithm on the deadline, but also relatively.

As can be seen from Figure 5, optimal finish time dispatching algorithm is all better than based on the satisfaction of most of task in the cloud task scheduling algorithm of user satisfaction and total user satisfaction.This effective aspect has revealed the validity of the algorithm designed by the present invention.

As can be seen from Figure 6, executory cost based on most of task in the cloud task scheduling algorithm of user satisfaction all will lower than optimal finish time dispatching algorithm, system synthesis based on the cloud task scheduling algorithm of user satisfaction is originally 25587 units, and the system synthesis of optimal finish time dispatching algorithm is originally 30432 units.The cloud task scheduling algorithm that this effective aspect has revealed based on user satisfaction has advantage on cost-saving.

By above analysis of simulation result, compared to optimal finish time task scheduling algorithm, cloud task scheduling algorithm based on user satisfaction can on the basis ensureing good task completion time, better meet the demand of different user, improve the satisfaction of user, also effectively can save executory cost simultaneously.

Innovative point of the present invention

1) the angle design cloud task scheduling algorithm of the satisfaction of cloud computing resources is used from user.

2) dispatching algorithm is under the condition of the deadline of the task of guarantee, by dynamic Task Assigned Policy, the user of different demand can better be met, and good user satisfaction and the total satisfactory grade of system can be obtained, effectively can save the executory cost of system.

Claims (1)

1., based on a cloud task scheduling algorithm for user satisfaction, it is characterized in that, comprise the steps:

Step 1, resource parameters normalization

The performance parameter of task and virtual machine is all normalized to [0,1] interval, order set X _ij={ X _1j..., X _tjj is the number of performance parameter, be the set of the similar performance parameter of virtual machine, its normalized value is:

GX _ij＝(curX _ij-minX _ij)/(maxX _ij-minX _ij) (5)

Wherein, i is the number of task, and j is the number of performance parameter, curX _ijfor performance currency, minX _ijfor the minimum value in similar performance parameter sets, maxX _ijfor the maximum in similar performance parameter sets.

Step 2, calculates Euclidean distance

Be X={X at the parameter vector of virtual machine after parameter normalization process ₁, X ₂, X ₃, the parameter vector of cloud task is Y={Y ₁, Y ₂, Y ₃; Consider CPU, internal memory and bandwidth three performance parameters, obtain weight vectors W={W according to the type of cloud task ₁, W ₂, W ₃, then the computing formula of Euclidean distance is:

$D_i=\sqrt{{\mathrm{\&Sigma;}}_{j=1}^3{W}_j*{(X_j-Y_j)}^2}---\left(6\right)$

Wherein X _jrepresent the normalized parameter value of a jth resource in i-th virtual machine; Y _jexpression task is to the expected value of jth kind resource; W _jrepresent the weight of jth kind resource;

Step 3, selects resource

Each task choosing and the minimum virtual machines performing tasks of its Euclidean distance, the method controlling idle virtual machine is adopted to carry out load balance, each virtual machine safeguards an Euclidean distance table, after certain task is successfully assigned to certain virtual machine execution, need to upgrade Euclidean distance table, increase the Euclidean distance between certain class resource and this virtual machine, more new formula is:

D _i'=D _i(1+1/n), n is the number (7) of virtual machine

Step 4, calculates user satisfaction

After task completes, consider the performance of each task, comprise the deadline of task and the user satisfaction of each task, and the comprehensive satisfaction of all tasks; The user satisfaction of individual task is:

$s_i={\mathrm{\&Sigma;}}_{j=1}^3{W}_j\ln ({\mathrm{act}}_j/{\exp }_j)---\left(8\right)$

Wherein, s _ifor the user satisfaction of task i; W _jfor the weight of jth item performance parameter; act _jfor task is to the actual consumption of jth item performance parameter; exp _jfor cloud task is to user's expected value of jth item performance;

When 0≤| s _i| when≤0.5, then think that the Resourse Distribute of user to cloud task i is felt quite pleased; Work as 0.5<|s _i| when≤1, then think that the Resourse Distribute of user to cloud task i is satisfied; When | s _i| during >1, then think that user is unsatisfied with the Resourse Distribute of cloud task i; When | s _i| value very large time, then think that user is very dissatisfied to the Resourse Distribute of cloud task i;

User's comprehensive satisfaction of all cloud tasks is:

$S={\mathrm{\&Sigma;}}_{i=1}^t\left|s_i\right|---\left(9\right)$

Wherein s _iit is the user satisfaction of i-th task; T is the number of all cloud tasks;

Step 5, assesses the cost

After executing each cloud task, calculate spent cost of executing the task; Virtual machine according to unit to resource charging, task consumption full payment cost _ifor:

${\cos t}_i=C_{\mathrm{cpu}/\mathrm{num}}*P_1+C_{\frac{\mathrm{mem}}{\mathrm{MB}}}*P_2+C_{\mathrm{bw}/\mathrm{Mbps}}*P_3---\left(10\right)$

Wherein, P _ifor resource quantity, C is unit resource price;

After performing all cloud tasks, the total cost of system is:

$C={\mathrm{\&Sigma;}}_{i=1}^t{\cos t}_i---\left(11\right)$

Wherein cost _iit is the cost of i-th task; T is the number of all cloud tasks.