CN104869154A - Distributed resource scheduling method for balancing resource credibility and user satisfaction - Google Patents

Distributed resource scheduling method for balancing resource credibility and user satisfaction Download PDF

Info

Publication number
CN104869154A
CN104869154A CN201510200625.7A CN201510200625A CN104869154A CN 104869154 A CN104869154 A CN 104869154A CN 201510200625 A CN201510200625 A CN 201510200625A CN 104869154 A CN104869154 A CN 104869154A
Authority
CN
China
Prior art keywords
resource
agent
user
scheduling
work
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201510200625.7A
Other languages
Chinese (zh)
Inventor
江务学
胡选子
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Individual
Original Assignee
Individual
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to CN201510200625.7A priority Critical patent/CN104869154A/en
Publication of CN104869154A publication Critical patent/CN104869154A/en
Pending legal-status Critical Current

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H04L67/104—Peer-to-peer [P2P] networks
    • H04L67/1074—Peer-to-peer [P2P] networks for supporting data block transmission mechanisms
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/50—Reducing energy consumption in communication networks in wire-line communication networks, e.g. low power modes or reduced link rate

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a distributed resource scheduling method for balancing resource credibility and user satisfaction. The method comprises the following steps: submitting work to a work response Agent; the work response Agent returning a work completion result to a user; a work decomposition distribution Agent distributing decomposed subjobs to a work scheduling Agent; a resource search Agent searching for online resources in a network and registering the resources; the work scheduling Agent distributing the subjobs to a resource Agent; the resource Agent determining registration, receiving work sent by the work scheduling Agent and completing the work; and a work monitoring Agent monitoring the state of the resource Agent and redistributing the work when receiving a work handover application sent by the resource Agent. The method has the advantages of comprehensively improving the user satisfaction, reducing the average work return time and improving the operation efficiency of a distributed system.

Description

Plan as a whole the distributed resource scheduling method of resource confidence level and user satisfaction
Technical field
The invention belongs to distributed algorithm technical field, relate to the distributed resource scheduling method planning as a whole resource confidence level and user satisfaction.
Background technology
Distributed computing system make use of number on the Internet, in millions upon millions of idle computational resources, can obtain the computing capability more powerful than centralized computing system.Due to the retardance of the magnanimity of its resource, the isomerism of composition and network service, its resource scheduling is made to become difficult point.The scheduling of resource refers to that the operation by user submits to is reasonably allocated to appropriate resource to meet the demand of user, maximizes the benefit of system cloud gray model.For distributed resource scheduling problem, scholar proposes a series of scheduling strategy.Document [1]proposing a kind of distributed resource scheduling strategy forecast model based on ant group algorithm, mainly in order to improve dynamic and the real-time of Distributed Calculation, improving real-time and the validity of scheduling of resource.Document [2,3]propose the distributed resource scheduling strategy based on confidence level, its core is the load state of prediction resource future time instance and then calculates confidence level, decides the distribution of operation according to confidence level.Document [4,5]propose the hereditary task scheduling algorithm based on user's total satisfaction, this algorithm devises user satisfaction function according to the preference of different user, carrys out Optimized Operation process by genetic algorithm, thus the total satisfaction of user is got a promotion.Document [6,7]propose a kind of equalized scheduling algorithm, this algorithm calculates component level value to operation according to dependence, operation dynamic conditioning is carried out according to this value, key operation is completed as early as possible, reduce the wait between operation, shorten the computational tasks stream time of implementation, result shows, the quick execution of this algorithm to the computational tasks stream dropped in job management system has stronger superiority.Document [8,9]the genetic algorithm about scheduling problem in existing document is studied and relatively after, this article proposes a kind of method for scheduling task based on genetic algorithm, devises a kind of mutation operator different from other algorithms in the algorithm.Document [10]for the problem that fixing process Node distribution formula system dynamic regulation ability is weak, provide a kind of distributed system Task Scheduling Model.Simulated experiment proves, this algorithm has good dynamic regulation ability, can reduce processor load as required, improves task process time delay and more reasonably utilize system resource.Document [11]propose a kind of task scheduling load-balancing algorithm based on fair standard, derive the method for allocating tasks under multinode condition, and improve the load-balancing algorithm based on fair standard under this model.Document [12]in further investigation distributed system load balance scheduling problem basis on, the induction and conclusion universal model of load balance scheduling, has carried out detailed analysis to each factor affecting load balance.Document [13]propose the distributed resource allocation strategy (MADRAP) based on the mobile agent, this strategy improves study mechanism on traditional distributed resource allocation policy grounds, achieves the task balance load of Existence of Global Stable.Document [14]have studied the Agent dispatching algorithm in distributed real-time systems, establish the structural model of distributed Real-time Multi-agent system, and its design feature, load capacity and conventional Agent real-time scheduling are analyzed respectively and discussed.Document [15]discuss the Parallel Scheduling problem of many group jobs in Agent system, propose concept and equilibrium-compression dispatching algorithm (MADTBCSA) of multi-Agent inter-related task of dispatching efficiency.
Above-mentioned research or lay particular emphasis on the research of resource reliability, and have ignored the requirement of user for operation, because the user had in reality is responsive for running time, and some users are responsive for operating cost, and the demand ignoring user will cause the decline of user satisfaction; Or put undue emphasis on user satisfaction, as document [2]the account form of middle satisfaction is static, the change of satisfaction after failing to consider load; Or bias toward Design cooling load equilibrating mechanism and Agent coordination mechanism, and consider the impression of user and the reliability of resource in target function, with failing hommization.
Summary of the invention
The object of the present invention is to provide the distributed resource scheduling method planning as a whole resource confidence level and user satisfaction, the average turnaround time solving existing algorithm operation is high, the problem that running efficiency of system is low.
The technical solution adopted in the present invention is carried out according to following steps:
Step 1: User interface Agent: to operation response Agent submit job, the consumer as resource occurs, takies resource, releasing resource after operation completes during job request success;
Step 2: operation response Agent: accept the job request that user submits to, complete result to user's backtracking;
Step 3: breakdown of operation distribution Agent: accept the operation that operation response Agent sends, breakdown of operation is become analysable minimum particle size, i.e. subjob, and the subjob after decomposing is issued job scheduling Agent;
Step 4: resource searching Agent: resource online in search network, registration resource;
Step 5: job scheduling Agent: job scheduling Agent accepts the subjob being divided into minimum particle size, is assigned in resource agents by dispatching algorithm by subjob;
Step 6: resource agents: send online message to resource searching Agent time online, select resource searching Agent, determine registration, accept the operation that job scheduling Agent sends, fulfil assignment, detect oneself state, send own resource confidence level to resource searching Agent, apply for transferring operation to monitoring operation Agent when scheduling is come down in torrents or operation is failed;
Step 7: monitoring operation Agent: the state of monitoring resource agents, when receiving the transfer job request that resource agents sends, redistributes operation.
Further, the dispatching algorithm in described step 5 is:
If the operation Agent resource in job scheduling Agent is 1,2 ..., m, and resource 1,2 ..., the reliability matrix R=[r of m 1, r 2..., r m] t, matrix is the linear combination of resource confidence level r (t), and cost metrix C=[c 1, c 2..., c n] t, matrix is the linear combination of multiple expense, and the operation 1,2 accepted ..., the satisfaction matrix of n and operation
S = w 11 w 21 L w n 1 w 12 w 22 L w n 2 L L L L w 1 m w 2 m L w nm ;
Wherein, w ijrepresent that i-th user is to the satisfaction of a jth resource, w nmrepresent that nth user is to the satisfaction of m resource, w nmin n and m represent user job number and resource number respectively, and w ijmiddle i and j submeter represents i-th concrete user job and a concrete jth resource, and adopt genetic algorithm to calculate, computational process is as follows:
Chromosome coding:
Vector D=[the d of chromosome to be the element of n dimension be integer 1, d 2, L, d n], d i(i=1 ~ n) is integer, represents that i-th operation has been assigned to d iin individual resource, d ispan be that 1 ~ m, D vector represents current work 1,2 completely ..., the Resources allocation situation of n;
Initialization of population:
After arranging Population Size popNum, generate popNum chromosome, the element in each chromosome is stochastic generation, and the scope of random number is between 1 ~ m;
Fitness function: after initialization completes, assesses fitness chromosomal in population, and i-th operation is assigned to d iin individual resource, the now extent function of user as follows:
s i , d i = w it 1 h d i r d i r b + w ie c b c d i
w it+w ie＝1，
W itand w iebe respectively the preference of user for time performance and economic cost, r jfor the confidence level of a jth resource, r bfor distributed system history resource confidence level average, c jfor a jth resource is to the quotation of user i, c bfor System History resource quotation average;
Fitness evaluating function is as follows:
fitness = Σ i = 1 n s i , d i .
Further, described resource confidence level is
r(t)＝p(t)*i(t)*h(t)*v(t)，
Wherein, t is the time, r (t) is the confidence level of resource in t, it is made up of p (t), i (t), h (t), v (t) four part, p (t) is the performance function of resource, the availability factor that i (t) is resource, the online rate that h (t) is resource, the success rate that v (t) fulfils assignment for resource.
Further, described w itand w iespan is [0,1], according to the difference of user type, arranges different w itand w ie, for performance priority user, w is set it=0.8, w ie=0.2; For economic priority user, w is set it=0.2, w ie=0.8; For the balanced user of compromise, w is set it=0.5, w ie=0.5.
The invention has the beneficial effects as follows the satisfaction comprehensively improving user, reduce the average turnaround time of operation, improve the operational efficiency of distributed system.
Accompanying drawing explanation
Fig. 1 is Multi-agent scheduling strategy frame structure schematic diagram of the present invention;
Fig. 2 the present invention is based on resource dispatching strategy genetic algorithm flow chart.
Embodiment
Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.
Herein for current distributed system scheduling of resource Problems existing, propose a kind of Multi-agent distributed system resource dispatching strategy considering user satisfaction and resource confidence level.This strategy develops simultaneously the satisfaction of user and the confidence level of resource in target function, by the confidence level of resource is integrated in access customer extent function, be optimization aim again with user satisfaction, adopt genetic algorithm for solving, scheduling result can be made to maximize the interests of user and the efficiency of calculating.After the encapsulation of above-mentioned scheduling strategy is entered Agent, propose a kind of Multi-agent Scheduling Framework, the autonomy utilizing Multi-agent outstanding, study idea and socially realize distributed resource scheduling.
Definition resource confidence level:
The confidence level of resource weighs the disposal ability of resource and the index of reliability.It has embodied a concentrated reflection of this resource from the disposal ability of current to the future a period of time and reliability.By setting up the credibility model of resource, making the invoked probability of the resource of high confidence level larger, and calling the resource even not calling low confidence level less, the load state of resource can be made more balanced.The confidence level of resource is dynamic, is change along with its performance, utilance, online rate and the success rate change that fulfils assignment.In scheduling strategy, introduce the confidence level of resource, make scheduling be dynamic, along with the change of resource changes, this distributed system often changed for resource situation is significant.
The confidence level of resource is defined as follows:
r(t)＝p(t)*i(t)*h(t)*v(t)
Wherein, t is the time, and r (t) is the confidence level of resource in t, and it is made up of p (t), i (t), h (t), v (t) four part.P (t) is the performance function of resource, reflects the ability of resource processing operation, if resource is CPU, is then the dominant frequency of CPU.The availability factor that i (t) is resource, the same with p (t), embodiment be the disposal ability of resource, if resource is CPU, be then (occupancy of 1-CPU).The online rate that h (t) is resource, reflects the reliability of resource.What utilize due to distributed system is resource on internet, therefore be difficult to predict that resource is at a time reached the standard grade or off-line, but can obtain h (t) by the rate of reaching the standard grade of adding up resource, the reliable resources degree that rate of reaching the standard grade is higher is higher, and confidence level is also larger.The success rate that v (t) fulfils assignment for resource, similarly, not all can be successfully completed task after resource receives an assignment, the height of its success rate of finishing the work characterizes the height of reliability at every turn.
Definition user satisfaction:
Most of Study of Scheduling only lays particular emphasis on the global index that optimizes the system operation, and have ignored the user satisfaction from individual angle.In fact, for different user, it is different to the time performance of operation and the preference of economic cost, and as in distributed animation rendering system, user A more pays close attention to playing up expense, and user B more pays close attention to render time.For same subscriber, when different when, it is also different to the preference of the time performance of operation and economic cost.When user A faces a urgent need, if its client C require must see rendering effect as early as possible, and and do not mind the reward for this reason paid more, now user A will be more prone to Fast rendering; And when user A is in the face of a non-emergent demand, as the limited fund of its client D, require that A controls budget, and hard requirement is not had to the deadline, at this moment can imagine, A will be more prone to reduce economic cost; When user A faces the client E to time performance and economic cost requirement equilibrium, the scheme showing that A will select compromise balanced of being not difficult.
In view of this, for different users, need design for the extent function of its preference, to meet the demand of different user, and the target function of scheduling strategy should consider the Maximum Satisfaction that makes user overall.Set up user i is assigned to resource j satisfaction s to its operation ijas follows:
s ij = w it r j r b + w ie c b c j ,
w it+w ie＝1，
W itand w iebe respectively the preference of user for time performance and economic cost, span is [0,1].R jfor the confidence level of a jth resource, r jlarger, the satisfaction of user is larger.R bfor distributed system history resource confidence level average.C jfor a jth resource is to the quotation of user i, c jlower, the satisfaction of user is higher.C bfor System History resource quotation average.
According to the difference of user type, different w can be set itand w ie.As for performance priority user, can w be set it=0.8, w ie=0.2; For economic priority user, w is set it=0.2, w ie=0.8; For the balanced user of compromise, w is set it=0.5, w ie=0.5.
Multi-agent scheduling strategy is adopted to analyze:
Namely an important branch of artificial intelligence technology is Multi-agent technology.Namely Agent acts on behalf of, refer to one can in certain circumstances continuously, spontaneously practical function, and to the software entity of relevant Agent and process cross-talk.Agent has independence, can complete its most of function, control its internal state under nobody or other Agent intervene.The another one important attribute of Agent is exactly its social ability, and it can initiatively carry out alternately, to reach target with other Agent.Agent also possesses certain learning ability, can make a change along with environment adaptively.Multi-agent system refers to the system be made up of multiple Agent.It can solve and rely on the insoluble challenge of single Agent.Common Solve problems is carried out by competition, coordination and cooperation between multiple Agent.Traditional dispatching method, as branch and bound method, Dynamic Programming, HEURISTIC ALGORITHM FOR GRAPH SEARCH etc., although mathematically very perfect, but owing to having made a large amount of simplification to true environment, it is made to be difficult to adapt to complicated production environment, the single limitation that result also in its scope of application of while method.Scheduling strategy based on Multi-agent technology can overcome the various shortcoming of above-mentioned algorithm, what rely on due to it is the swarm intelligence effect of Distributed agent, make to the simulation of challenge and disposal ability very outstanding, have powerful adaptive ability concurrently simultaneously, under various circumstances can adaptively modifying oneself state, study relevant knowledge, reach the dispatching effect of expection, there is very significant robustness.
Multi-agent scheduling strategy of the present invention, general frame as shown in Figure 1.Comprise following several Agent:
● User interface Agent: to operation response Agent submit job, the consumer as resource occurs.Resource is taken, releasing resource after operation completes during job request success.
● operation response Agent: the intermediate layer belonging to distributed system and user's handing-over, accepts the job request that user submits to, complete result to user's backtracking.
● breakdown of operation distribution Agent: accept the operation that operation response Agent sends, breakdown of operation is become analysable minimum particle size, i.e. subjob, and the subjob after decomposing is issued job scheduling Agent.
● resource searching Agent: resource online in search network, registration resource.
● job scheduling Agent: job scheduling Agent accepts the subjob being divided into minimum particle size, and selection scheduling algorithm, is assigned to subjob in resource agents and goes.
● resource agents: send online message to resource searching Agent time online, select resource searching Agent, determine registration.Accept the operation that job scheduling Agent sends, fulfil assignment.Detect oneself state, send own resource confidence level to resource searching Agent.Apply for transferring operation to monitoring operation Agent when scheduling is come down in torrents or operation is failed.
● monitoring operation Agent: the state of monitoring resource agents, when receiving the transfer job request that resource agents sends, redistributes operation.
Dispatching algorithm is wherein: the operation resource dispatching strategy adopting genetic algorithm;
Operation resource dispatching strategy is encapsulated in the operation Agent resource 1,2 among job scheduling Agent ..., m, wherein resource refers to such as CPU, internal memory etc.And resource 1,2 ..., the reliability matrix R=[r of m 1, r 2..., r m] t, matrix is the linear combination of the multiple r (t) in step 1.And cost metrix C=[c 1, c 2..., c n] t, matrix is the linear combination of multiple expense, and the operation 1,2 accepted ..., the satisfaction matrix of n and operation
S = w 11 w 21 L w n 1 w 12 w 22 L w n 2 L L L L w 1 m w 2 m L w nm ;
This matrix value using the important parameter as genetic algorithm, to form the extent function of genetic algorithm, wherein, w ijrepresent that i-th user is to the satisfaction of a jth resource.
W nmrepresent that nth user is to the satisfaction of m resource.W nmin n and m represent user job number and resource number respectively, and w ijmiddle i and j submeter represents i-th concrete user job and a concrete jth resource.
Its completing of task is the operation 1,2 that will accept ..., n is assigned to resource 1,2 respectively ..., m.Because scheduling problem belongs to np problem, traditional algorithm is adopted often to be difficult to obtain globally optimal solution.Genetic algorithm is de-clever at numerous dispatching algorithm with the global optimizing ability of its excellence.Chromosome coding, initialization of population, genetic manipulation, Fitness analysis.
The flow process of algorithm is as shown in Figure 2:
1) chromosome coding
Vector D=[the d of chromosome to be the element of n dimension be integer 1, d 2, L, d n].D i(i=1 ~ n) is integer, represents that i-th operation has been assigned to d iin individual resource.D ispan be 1 ~ m.Visible, D vector represents current work 1,2 completely ..., the Resources allocation situation of n.
D=[d 1, d 2, L, d n], d i(i=1 ~ n), d ifor integer, number is n, and span is 1 ~ m, just in time have expressed n user job and is assigned in m resource respectively; User in the present invention is identical with user job implication, and the satisfaction of user is embodied by the satisfaction of user job.
Initialization of population:
After arranging Population Size popNum, algorithm will carry out random initializtion, and generate popNum chromosome, the element in each chromosome is stochastic generation, and the scope of random number is between 1 ~ m.
2) Fitness analysis
After initialization completes, need to assess fitness chromosomal in population.Fitness function reasonable in design is the key of genetic algorithm, because fitness is induction of the Evolutionary direction of population, fitness function may cause convergence in population solution and optimal solution to obtain relatively large deviation improperly.
As in step 2 to the elaboration of satisfaction, i-th operation is assigned to d iin individual resource, cause the satisfaction of now user as follows:
s i , d i = w it r d i r b + w ie c b c d i
Resource reliability matrix R and cost metrix C is as extent function a part for matrix, is dissolved into resource confidence level in the function of user satisfaction, carrys out part most crucial in structure genetic algorithm (for scheduling of resource), i.e. extent function.
Because same resource may be assigned to multiple operation, in fact multiple operation is caused to share the confidence level of same resource.Therefore according to the operation quantity that each Resourse Distribute arrives, need to make adjustment to the extent function of user.Make resource d ithe operation quantity be simultaneously assigned to is the user satisfaction function obtaining revising is as follows:
s i , d i = w it 1 h d i r d i r b + w ie c b c d i
Fitness evaluating function is as follows:
fitness = Σ i = 1 n s i , d i
3) genetic manipulation:
First select, select to refer to according to certain principle and preference, from chromosome population, select individuality breed, common principle is that the individuality that fitness is higher is larger by the chance selected, what adopt herein is selection mechanism based on roulette, and the ratio accounting for overall fitness by ideal adaptation degree is as the selected probability of individuality.
Hybridization is after selecting two or more chromosomes, chooses certain site in chromosome at random, it is divided into two sections, and two chromosomal front and back sections are exchanged, to produce two individualities of new generation.
Variation be with aberration rate be probability again on stochastic generation item chromosome some bit code.Appropriate aberration rate can strengthen the global optimizing ability of algorithm, but excessive, and algorithm can be caused to be difficult to convergence.The hybridization of this genetic algorithm adopts single-point hybridization, and as the statement of blue portion word above, the crossover probability in example all gets P c=0.6.Variation adopts conventional variation method, and mutation probability gets P c=0.1, still generate a new individuality after variation, with the individuality do not made a variation together, participate in the survival of the fittest of next round.
The present invention will be described to enumerate specific embodiment below.
Embodiment 1: based on JACK agent language, achieve the emulation of Multi-agent distributed scheduling strategy.Operation type and the quantity of submission are as shown in table 1.
Table 1 operation type and quantity
As shown in Table 1, according to job run fiducial time (fiducial time is set to the time run in the CPU of dominant frequency 1GHZ) length, homework type is divided into simple, medium and difficult three classes.In every class operation, the composition of user is had nothing in common with each other again.As in simple homework type, the operation of 70% is economical preferential, and performance priority and balanced type then respectively account for 15%.This is that running time is short because simple operation completes fast, and therefore user is more partial to has requirement to its economy.In medium homework type, performance priority, economic preferential, balanced type are similar to and respectively account for 1/3.Finally, for difficult operation, because its running time is longer, user more pays close attention to its performance, and therefore performance priority operation accounts for 70%, and other two classes respectively account for 15%.
Not the same time discharges above-mentioned task to Multi-agent system, but discharges to Multi-agent system at random in the time of 1 hour.
The number arranging resource agents in Multi-agent system is 10.The composition of resource agents is as shown in table 2:
Table 2 resource agent forms
Dominant frequency/Hz Running time
Agent 1～3 500M Benchmark * running time 2
Agent 4～7 1G Benchmark running time
Agent 8～10 2G Benchmark running time/2
This algorithm and Randomized scheduling algorithm and minimum load dispatching algorithm are compared.Obtain average response time and user satisfaction is as shown in table 3:
Table 3 average response time and user satisfaction contrast
From the above results, the average response time of Multi-agent and user's total satisfaction are all better than Randomized scheduling algorithm and minimum load dispatching algorithm, and wherein user's total satisfaction is significantly better than other two kinds of algorithms.
The invention has the beneficial effects as follows that the present invention establishes the assessment models considering user satisfaction and resource confidence level, according to this modelling based on the resource dispatching strategy of genetic algorithm, and propose with Multi-agent to be the distributed system resource scheduling system framework of framework.Comprehensively improve the satisfaction of user, reduce the average turnaround time of operation, improve the operational efficiency of distributed system.
The above is only to better embodiment of the present invention, not any pro forma restriction is done to the present invention, every any simple modification done above execution mode according to technical spirit of the present invention, equivalent variations and modification, all belong in the scope of technical solution of the present invention.
Index list:
[1] Zhou Wenjun, Cao Jian. based on prediction and the cloud computing resources scheduling strategy [J] of ant group algorithm. Computer Simulation, 2012,29 (9): 239-242.
[2]M ITZENMACH ER M.The power of two choices in randomized load balancing[J]IEEE Transactions onParallel and Distributed Systems，2001，12(10)：1094-1104.
[3] Xu Weimin, Cai Xinhuan, Shen Wenfeng, Zhi Fenglin. based on the system of distributed resource scheduling [J] of confidence level. Shanghai University's journal (natural science edition), 2009,15 (1): 77 ~ 80
[4]DAH LIN M.Interpreting stale load info rma tion[J].IEEE Transactions on Para llel and D istributedSystems，2000，11(10)：1033-1047.
[5] Wang Xiaoguang, Wang Yongbin, Yang Xiaogang. based on the hereditary task scheduling algorithm [J] of user's total satisfaction. Communication University of China's journal, 2010,17 (3): 53 ~ 56
[6]Glatard T，Montagnat J，Lingrand D，et al.Flexible and efficient workflow deployment of data-intensiveapplications on grids with MOTEUR.International Journal of High Performance Computing Applications.2008
[7] soup late autumn, Li Honghua. the equalized scheduling algorithm [J] of computational tasks stream in distributed system. computer engineering, 2010,36 (19): 78 ~ 80
[8]TStutzle，M Dorigo.A short convergence p roof for a class of antcolony optimization algorithms[J].IEEETransactions on Evolutionary Computation，2002，6(4)：358-365.
[9] Sun Jun, must civilian ripple. a kind of task scheduling [J] of the distributed system based on genetic algorithm. and computer engineering and application, 2003,39 (21): 105 ~ 106,121
[10] beam root, Qin Yong, Guo little Xue, Liang Huomin. based on the distributed system task scheduling [J] of dynamic multiprocessing node. computer engineering, 2009,35 (9): 31 ~ 33,3
[11] beam root, Guo little Xue, Qin Yong. the load-balancing of distributed system based on fair scheduling algorithm is studied [J]. computer engineering and design, 2008,29 (6): 1362 ~ 1364
[12] Zhang Heng, Chen Chao, once in .P2P network of heap of stone, the searching resource of the mobile agent distributed research [J]. Southwestern Normal University's journal (natural science edition), 2010.35 (1): 164-167.
[13] Zhang Xiaoyu, Li Lixin. based on the distributed resource allocation strategy [J] of mobile Agent. Journal of UEST of China (natural science edition), 2007,36 (6): 1393 ~ 1395
[14] full in summer mountain, Xu Feng. the multi-Agent Scheduling [J] in distributed real-time systems. Nanjing Aero-Space University's journal, 2004,36 (3): 302 ~ 307.
[15] Tao Jianwen, Wang Fengru. a kind of multi-Agent inter-related task Parallel Scheduling Algorithm [J] of optimization. computer application, 2007,27 (3): 756 ~ 758.

Claims (4)

1. plan as a whole the distributed resource scheduling method of resource confidence level and user satisfaction, it is characterized in that carrying out according to following steps:
Step 1: User interface Agent: to operation response Agent submit job, the consumer as resource occurs, takies resource, releasing resource after operation completes during job request success;
Step 2: operation response Agent: accept the job request that user submits to, complete result to user's backtracking;
Step 3: breakdown of operation distribution Agent: accept the operation that operation response Agent sends, breakdown of operation is become analysable minimum particle size, i.e. subjob, and the subjob after decomposing is issued job scheduling Agent;
Step 4: resource searching Agent: resource online in search network, registration resource;
Step 5: job scheduling Agent: job scheduling Agent accepts the subjob being divided into minimum particle size, is assigned in resource agents by dispatching algorithm by subjob;
Step 6: resource agents: send online message to resource searching Agent time online, select resource searching Agent, determine registration, accept the operation that job scheduling Agent sends, fulfil assignment, detect oneself state, send own resource confidence level to resource searching Agent, apply for transferring operation to monitoring operation Agent when scheduling is come down in torrents or operation is failed;
Step 7: monitoring operation Agent: the state of monitoring resource agents, when receiving the transfer job request that resource agents sends, redistributes operation.
2. according to the distributed resource scheduling method planning as a whole resource confidence level and user satisfaction described in claim 1, it is characterized in that: the dispatching algorithm in described step 5 is:
If the operation Agent resource in job scheduling Agent is 1,2 ..., m, and resource 1,2 ..., the reliability matrix R=[r of m 1, r 2..., r m] t, matrix is the linear combination of resource confidence level r (t), and cost metrix C=[c 1, c 2..., c n] t, matrix is the linear combination of multiple expense, and the operation 1,2 accepted ..., the satisfaction matrix of n and operation
S = w 11 w 21 L w n 1 w 12 w 22 L w n 2 L L L L w 1 m w 2 m L w nm ;
Wherein, w ijrepresent that i-th user is to the satisfaction of a jth resource, w nmrepresent that nth user is to the satisfaction of m resource, w nmin n and m represent user job number and resource number respectively, and w ijmiddle i and j submeter represents i-th concrete user job and a concrete jth resource, and adopt genetic algorithm to calculate, computational process is as follows:
Chromosome coding:
Vector D=[the d of chromosome to be the element of n dimension be integer 1, d 2, L, d n], d i(i=1 ~ n) is integer, represents that i-th operation has been assigned to d iin individual resource, d ispan be that 1 ~ m, D vector represents current work 1,2 completely ..., the Resources allocation situation of n;
Initialization of population:
After arranging Population Size popNum, generate popNum chromosome, the element in each chromosome is stochastic generation, and the scope of random number is between 1 ~ m;
Fitness function: after initialization completes, assesses fitness chromosomal in population, and i-th operation is assigned to d iin individual resource, the now extent function of user as follows:
s i , d i = w it 1 h d i r d i r b + w ie c b c d i
w it+w ie＝1，
W itand w iebe respectively the preference of user for time performance and economic cost, r jfor the confidence level of a jth resource, r bfor distributed system history resource confidence level average, c jfor a jth resource is to the quotation of user i, c bfor System History resource quotation average;
Fitness evaluating function is as follows:
fitness = Σ i = 1 n s i , d i .
3. according to the distributed resource scheduling method planning as a whole resource confidence level and user satisfaction described in claim 2, it is characterized in that: described resource confidence level is
r(t)＝p(t)*i(t)*h(t)*v(t)，
Wherein, t is the time, r (t) is the confidence level of resource in t, it is made up of p (t), i (t), h (t), v (t) four part, p (t) is the performance function of resource, the availability factor that i (t) is resource, the online rate that h (t) is resource, the success rate that v (t) fulfils assignment for resource.
4. according to the distributed resource scheduling method planning as a whole resource confidence level and user satisfaction described in claim 2, it is characterized in that: described w itand w iespan is [0,1], according to the difference of user type, arranges different w itand w ie, for performance priority user, w is set it=0.8, w ie=0.2; For economic priority user, w is set it=0.2, w ie=0.8; For the balanced user of compromise, w is set it=0.5, w ie=0.5.
CN201510200625.7A 2015-04-27 2015-04-27 Distributed resource scheduling method for balancing resource credibility and user satisfaction Pending CN104869154A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510200625.7A CN104869154A (en) 2015-04-27 2015-04-27 Distributed resource scheduling method for balancing resource credibility and user satisfaction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510200625.7A CN104869154A (en) 2015-04-27 2015-04-27 Distributed resource scheduling method for balancing resource credibility and user satisfaction

Publications (1)

Publication Number Publication Date
CN104869154A true CN104869154A (en) 2015-08-26

Family

ID=53914671

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510200625.7A Pending CN104869154A (en) 2015-04-27 2015-04-27 Distributed resource scheduling method for balancing resource credibility and user satisfaction

Country Status (1)

Country Link
CN (1) CN104869154A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106878389A (en) * 2017-01-04 2017-06-20 北京百度网讯科技有限公司 Method and apparatus for carrying out scheduling of resource in cloud system
CN106993019A (en) * 2016-11-29 2017-07-28 上海壹账通金融科技有限公司 Distributed task dispatching method and system
CN110908782A (en) * 2019-11-01 2020-03-24 湖北省楚天云有限公司 Genetic algorithm optimization-based packaging type distributed job task scheduling method and system
CN111459669A (en) * 2020-03-30 2020-07-28 中科边缘智慧信息科技（苏州）有限公司 Multi-node task scheduling method under high dynamic resource non-equilibrium condition

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101916404A (en) * 2010-08-06 2010-12-15 沈阳工业大学 Multi-factory cooperative scheduling optimization method during equipment manufacturing
CN102263781A (en) * 2006-02-18 2011-11-30 华为技术有限公司 media resource scheduling method and system
CN102480480A (en) * 2010-11-26 2012-05-30 中国科学院软件研究所 Resource scheduling method under ubiquitous computing environment

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102263781A (en) * 2006-02-18 2011-11-30 华为技术有限公司 media resource scheduling method and system
CN101916404A (en) * 2010-08-06 2010-12-15 沈阳工业大学 Multi-factory cooperative scheduling optimization method during equipment manufacturing
CN102480480A (en) * 2010-11-26 2012-05-30 中国科学院软件研究所 Resource scheduling method under ubiquitous computing environment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
江务学: "统筹资源可信度与用户满意度的Multi-agent分布式资源调度策略", 《西南大学学报（自然科学版）》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106993019A (en) * 2016-11-29 2017-07-28 上海壹账通金融科技有限公司 Distributed task dispatching method and system
CN106993019B (en) * 2016-11-29 2019-11-19 深圳壹账通智能科技有限公司 Distributed task dispatching method and system
CN106878389A (en) * 2017-01-04 2017-06-20 北京百度网讯科技有限公司 Method and apparatus for carrying out scheduling of resource in cloud system
CN106878389B (en) * 2017-01-04 2020-02-07 北京百度网讯科技有限公司 Method and device for resource scheduling in cloud system
US10778807B2 (en) 2017-01-04 2020-09-15 Beijing Baidu Netcom Science And Technology Co., Ltd. Scheduling cluster resources to a job based on its type, particular scheduling algorithm,and resource availability in a particular resource stability sub-levels
CN110908782A (en) * 2019-11-01 2020-03-24 湖北省楚天云有限公司 Genetic algorithm optimization-based packaging type distributed job task scheduling method and system
CN111459669A (en) * 2020-03-30 2020-07-28 中科边缘智慧信息科技（苏州）有限公司 Multi-node task scheduling method under high dynamic resource non-equilibrium condition

Similar Documents

Publication Publication Date Title
Xu et al. Fuzzy optimal allocation model for task–resource assignment problem in a collaborative logistics network
CN102780759B (en) Based on the cloud computing resource scheduling method in regulation goal space
CN107659433B (en) Cloud resource scheduling method and equipment
Rathnayaka et al. Framework to manage multiple goals in community-based energy sharing network in smart grid
CN101719931B (en) Multi-intelligent body-based hierarchical cloud computing model construction method
CN103338241B (en) A kind of new publicly-owned cloud framework and virtual resources adaptive configuring method thereof
CN104065745A (en) Cloud computing dynamic resource scheduling system and method
Jana et al. Dynamic schedule execution in an agent based holonic manufacturing system
CN104102544A (en) Multi QoS (quality of service)-constrained parallel task scheduling cost optimizing method under mixed cloud environment
Durgadevi et al. Resource allocation in cloud computing using SFLA and cuckoo search hybridization
CN104023042B (en) Cloud platform resource scheduling method
CN107220118A (en) Resource pricing is calculated in mobile cloud computing to study with task load migration strategy
CN104869154A (en) Distributed resource scheduling method for balancing resource credibility and user satisfaction
CN106611277A (en) Cloud resource scheduling method in mixed cloud model
Xiang Energy emergency supply chain collaboration optimization with group consensus through reinforcement learning considering non-cooperative behaviours
Moghaddam et al. On coordination of smart grid and cooperative cloud providers
CN110008023A (en) Cloud computing system budgetary restraints Random Task dispatching method based on genetic algorithm
Fu et al. A combined multi-agent system for distributed multi-project scheduling problems
CN110321198A (en) A kind of container cloud platform computing resource and Internet resources coordinated dispatching method and system
Pereira et al. Web-based platform for the management of citizen energy communities and their members
Chen et al. Multi-objective robust optimal bidding strategy for a data center operator based on bi-level optimization
CN106095591A (en) A kind of virtual machine two-stage optimizing management and running platform based on cloud computing
Ni et al. Dynamic pricing mechanism design for electric mobility-on-demand systems
CN108768703A (en) A kind of energy consumption optimization method, the cloud computing system of cloud workflow schedule
CN101127700A (en) Grid resource scheduling method based on network bandwidth and user charge

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
DD01 Delivery of document by public notice
DD01 Delivery of document by public notice

Addressee: Jiang Wuxue

Document name: the First Notification of an Office Action

RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20150826