CN102708003A - Method for allocating resources under cloud platform - Google Patents
Method for allocating resources under cloud platform Download PDFInfo
- Publication number
- CN102708003A CN102708003A CN2011100756388A CN201110075638A CN102708003A CN 102708003 A CN102708003 A CN 102708003A CN 2011100756388 A CN2011100756388 A CN 2011100756388A CN 201110075638 A CN201110075638 A CN 201110075638A CN 102708003 A CN102708003 A CN 102708003A
- Authority
- CN
- China
- Prior art keywords
- task
- resource
- cloud
- banker
- allocation methods
- 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
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for allocating work flow resources under a cloud platform. The method is a quasi banker algorithm and aims to solve the most-worried problem that whether the deadline of a work flow task can be met or not. By the method, a double-layer resource allocation model is constructed according to the actual condition of the cloud platform, wherein a main scheduler allocates tasks to specific zones according to the information of the tasks in a first layer; and local schedulers in the specific zones allocate resources according to the quasi banker algorithm in the second layer. By the model, the resource utilization rate of the cloud platform can be increased, and the load balance of a cloud system can be improved. The invention also discloses a flow chart and detailed steps of the quasi banker algorithm, and the cost of the tasks can be minimized by the algorithm on the premise of ensuring the deadlines of the tasks.
Description
Technical field
The present invention relates to a kind of cloud computing resource allocation methods, relate in particular to the resource allocation methods of workflow task under a kind of cloud platform.
Background technology
Cloud computing (Cloud Computing) is a kind ofly to be distributed in the large-scale data center, can to provide kind of the server resource of clamoring to satisfy the computing platform of field demands such as scientific research, ecommerce dynamically.Cloud computing platform utilizes Intel Virtualization Technology through relevant scheduling strategy; Different demands to the user; Dynamically, transparent its required virtual calculating and storage resources that provide; And when the active user does not use, other users are supplied with in its resource dynamic recovery, just as the generating plant supplies power to the user carries cheap calculating and storage resources, domestic consumer's realization large-scale parallel calculating is operated with mass data becomes possibility.
In reality, the cloud system that company releases generally only provides one type service.Such as Google, the Google Apps of release belongs to the category of SaaS; Google App Engine belongs to the category of PaaS; And a series of new network application programs such as Google docs, Google Maps, Google Calendar etc., also all belong to the category of SaaS.Equally, the EC2 of Amazon Company and S3 belong to the category of IaaS.Certainly, these three kinds of services are can be simultaneous in the cloud platform, and also do not have boundary clearly between them.Wherein: SaaS, it is meant that it is oneself service that the user uses on the cloud platform existing software.In this service, the user can download and install application software, and just comes directly to use software through network, because application software is already installed in cloud system.And, can on any computer, check work on this application software, work on, perhaps with other people collaborative work.PaaS has provided a platform, can let the user on this platform, carry out the exploitation of application program, and this platform can provide relevant support.When application development has been got well, can also on this platform, issue to supply other people to use.IaaS is to go out the resource of the bottom of cloud system to lease, mainly is computational resource and storage resources.In existing cloud system, generally be to lease through the mode of virtual machine.
Unified the cloud resource is made up of the different main frame of the function of a large amount of distributions to the bottom, and the superiors are exactly the various application that the user submits to.The user hopes that the cloud platform accomplishes these application according to oneself requirement and demands of applications, and the cloud supplier hopes the higher and load balance of system performance.At this moment, just need a kind of resource allocation policy of practicality, satisfy the requirement of user and system.
Summary of the invention
In view of this, fundamental purpose of the present invention is to provide a kind of and is satisfying under the prerequisite of using closing time, makes the user effort minimum and makes the higher method of cloud system load balance performance as much as possible.
For achieving the above object, the invention provides a kind of resource allocation methods, this method has been constructed a two-layer resource allocator model according to the actual conditions of cloud platform:
Ground floor be by the master scheduling device with task according to information distribution separately to specific district;
The second layer is that the local scheduler on the given zone carries out resources allocation according to a class banker's algorithm of inventing.
This model comprises: pretreater, master scheduling device and three important component parts of local scheduler, and partition strategy storehouse, cloud task characteristic information storehouse, cloud idling-resource information and local these storage areas of cloud resource information.The each several part function is following:
Pretreater: after workflow task was submitted on the cloud platform by the user, pretreater just was divided into relatively independent first task-set according to the partition strategy storehouse with it, and got into a new formation to first task-set in proper order according to topological sorting.That is, a workflow task has been represented in this formation.And, when being divided into first task-set to a workflow task, the while the proportion of the closing time of this workflow task according to first task, also be allocated to each yuan task.That is, this moment, each first task all had oneself closing time.
The master scheduling device: first yuan task in all first task-set formations is carried out resources allocation, at first confirms the resource requirement of each first task according to cloud task characteristic information storehouse.And architecture and the operating system specific according to first required by task are assigned in the corresponding district.Need notice master scheduling device when this yuan task is finished, then the master scheduling device goes out team with the head of the queue of this yuan task place formation, carries out resources allocation for next head of the queue.
Local scheduler: be assigned to first task in this district by the master scheduling device, can on local scheduler, have formed a formation to be allocated, can carry out resources allocation to it this moment according to the class banker's algorithm of invention.After the good resource of first Task Distribution or first task execute, all to change the attribute information of corresponding cloud resource in the local cloud resource information bank.And, after first task executes, also to notify the master scheduling device to distribute monobasic task down.When finding that from local cloud resource information resource utilization is too high, will consider to notify the master scheduling device, increase resource according to current cloud idling-resource information state to this district.Perhaps resource utilization is low excessively, considers to regain resource.
Wherein, in the class banker's algorithm of invention, considered that the cloud platform is a real-time system, has task to submit to up at any time.So, in local scheduler, be provided with a global clock, at every turn to submitting to first task of coming up to carry out resources allocation in the time period.
The concrete assigning process of this algorithm is following:
A) task sorts by priority, and gives head of the queue unit Task Distribution resource earlier.Find all can satisfy this yuan task closing time, all tasks combinations.From all combinations, select one group of minimum examination of cost to distribute, carry out security inspection.
Can b) security row inspection that is: be upgraded respective resources " length when doing " attribute, find a kind of resource allocation policy that remaining all first tasks can be accomplished in closing time to remaining task.
C) if inspection is passed through, confirm that then the allocation strategy of this yuan task begins to distribute execution, head of the queue goes out team.Otherwise, change e) redistribute.
D) check whether this yuan task queue is empty, if this queue resource has assigned.Change if not a) and distribute for next first task begins examination.
E) cancel new resources more when busy " length " attribute, reselect the less resource of next group cost and carry out security inspection, change b).
Compare with prior art, advantage of the present invention is:
The present invention has proposed a two-layer resource allocator model at the actual environment that is combining the cloud plateform system.This model prevents that through being divided into master scheduling device and local scheduler the load of master scheduling device is overweight.And the master scheduling device can be adjusted the load balancing of total system; And in the class banker dispatching algorithm of invention, guaranteeing to use one through security inspection satisfies closing time surely; Can just satisfy the combination of resources of its closing time for Task Distribution, thereby make user's cost minimum, and make the resource utilization of system higher.
Description of drawings
Fig. 1 is that the workflow task resource allocator model is always schemed under the cloud platform provided by the present invention;
Fig. 2 is a provided by the present invention type of banker's algorithm general flow chart;
Fig. 3 gives a kind of process flow diagram that just satisfies this task combination of resources closing time of Task Distribution in the provided by the present invention type of banker's algorithm;
Fig. 4 is the process flow diagram of security inspection in the provided by the present invention type of banker's algorithm;
Embodiment
Core concept of the present invention is: adopt two-layer resource allocator model, and on second layer local scheduler, use a type banker's algorithm to solve resource allocation problem.
For making the object of the invention, technical scheme and advantage clearer, the present invention is done to describe in detail further below in conjunction with accompanying drawing and specific embodiment.
The resource allocator model of workflow task comprises pretreater, master scheduling device and three important component parts of local scheduler under the cloud platform as shown in Figure 1, and partition strategy storehouse, cloud task characteristic information storehouse, cloud idling-resource information and local these storage areas of cloud resource information.Concrete workflow is following:
A) at first the user is submitted to workflow task t (i) on the cloud platform, and be d (i) closing time of this task of indicating.
B) task is through after the pre-service, is divided into relatively independent and by first task-set t (i1), the t (i2) of topological sorting ... t (ij) ... t (in), and gives the closing time of first Task Distribution by proportion and be respectively d (i1), d (i2) ... d (ij) ... d (in).This yuan task-set has been formed a new formation, and head of the queue is t (i1).
C) then, the master scheduling device can be given head of the queue t (i1) Resources allocation.Suppose to learn that through cloud task characteristic information storehouse t (i1) need be x86 in architecture that operating system is to carry out in the district 1 of linux.Then, the master scheduling device is assigned to t (i1) in the district 1 and carries out.
D) last, the local scheduler in the district 1 will be t (i1) Resources allocation according to the class banker's algorithm of invention.
E) after t (i1) is finished, local scheduler notice master scheduling device, the master scheduling device goes out team with t (i1), is new head of the queue t (i2) Resources allocation.
F) all first tasks are finished in this formation, and then the master scheduling device can notify user job stream task to be finished.
Shown in as shown in Figure 2 type of banker's resource allocation methods main-process stream, if the demand of first task t (i) of this moment is: calculated amount is w1, and memory requirements is w2, and amount of bandwidth is w3, and the external memory demand is w4; And four types of resources on the cloud platform are as shown in the table:
Table 1 computational resource and bandwidth resources
Table 2 memory source and external memory resource
Total type of resource | The unit resource price | The unit resource standard |
Memory source | c2 | CT |
The external memory resource | c4 | CT |
Then for first task t (i), if give the closing time of distributing highest ranking can satisfy this task, then the cost of this yuan task is: C (i)=(w1)/(v13) * (c13)+(w3)/(v33) * (c33)+(w2)/CT* (c2)+(w4)/CT* (c4).
Should distribute the combination of resources of the closing time that to satisfy this task for first task t (i) earlier according to Fig. 3; Examination distributes to first task t (i) then, carries out security inspection according to Fig. 4; If security inspection passes through, then confirm to distribute this combination of resources to t (i); If fail, then distribute the combination of resources that can just satisfy the closing time of this task in the remaining combination of resources for first task t (i) according to Fig. 3 again through security inspection; Try once more to distribute up to through security inspection, then carry out resources allocation for the remaining first task of this formation; All resources are all distributed resource and are finished in this formation, and then the pairing workflow task of this formation is accomplished.If there is not combination of resources can pass through security inspection, then current resource can not satisfy the closing time of all first tasks in this formation.At this moment, local scheduler increases resource to the application of master scheduling device; The master scheduling device is given local scheduler location Resources allocation according to idling-resource information, and local scheduler carries out resources allocation more again and gets final product.
Claims (10)
1. the workflow resource distribution method under the cloud platform, this method has been constructed a two-layer resource allocator model according to the actual conditions of cloud platform:
Ground floor be by the master scheduling device with task according to information distribution separately to specific district;
The second layer is that the local scheduler on the given zone carries out resources allocation according to a class banker's algorithm of inventing.
2. resource allocation methods according to claim 1 is characterized in that: pretreater is divided into relatively independent first task-set according to the partition strategy storehouse with workflow task, and gets into a new formation to first task-set in proper order according to topological sorting.
3. according to any described resource allocation methods of claim 1 to 2, it is characterized in that: when being divided into first task-set to a workflow task,, also be allocated to each yuan task simultaneously the proportion of the closing time of this workflow task according to first task.
4. resource allocation methods according to claim 1 is characterized in that: the different resources of virtual machine of performance, can according to its architecture (x86, mips etc.), operating system (linux, windows, os/2 etc.) and install software be divided into the district.
5. according to any described resource allocation methods of claim 1 to 4; It is characterized in that: the master scheduling device carries out resources allocation to first yuan task in all first task-set formations; Confirm the resource requirement of each first task earlier according to cloud task characteristic information storehouse; Specific according to first required by task then architecture and operating system are assigned in the corresponding district.
6. according to any described resource allocation methods of claim 1 to 5; It is characterized in that: the first task that is assigned to this district by the master scheduling device; On local scheduler, formed a formation to be allocated, can carry out resources allocation to it this moment according to the class banker's algorithm of invention.
7. according to any described resource allocation methods of claim 1 to 6, it is characterized in that: after first task executed, local scheduler will notify the master scheduling device to distribute the following monobasic task of this yuan task place formation.
8. resource allocation methods according to claim 1 is characterized in that: find in the local cloud resource information that resource utilization is too high, will consider to notify the master scheduling device, increase resource according to current cloud idling-resource information state to this district; Perhaps resource utilization is low excessively, considers to regain resource.
9. resource allocation methods according to claim 1 is characterized in that: a global clock should be arranged in local scheduler, and tackle at every turn and submit to first task of coming up to carry out resources allocation in the time period:
A) from t
0To (t
0+ a) constantly in, according to current resource status, to being submitted to first task queue type of carrying out banker resources allocation of local scheduler;
B) then, upgrade resource status;
C) again to (t
0+ a) to (t
0+ 2a) resources allocation is carried out in the interior constantly first task queue that arrives.
10. according to any described resource allocation methods in the claim 1 to 9, it is characterized in that: need use a new attribute of resource, called after " length when hurrying " refers to how long this resource just can become idling-resource in addition.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2011100756388A CN102708003A (en) | 2011-03-28 | 2011-03-28 | Method for allocating resources under cloud platform |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2011100756388A CN102708003A (en) | 2011-03-28 | 2011-03-28 | Method for allocating resources under cloud platform |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102708003A true CN102708003A (en) | 2012-10-03 |
Family
ID=46900818
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2011100756388A Pending CN102708003A (en) | 2011-03-28 | 2011-03-28 | Method for allocating resources under cloud platform |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102708003A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102981890A (en) * | 2012-11-30 | 2013-03-20 | 华南理工大学 | Computing task and virtual machine deploying method within a virtual data center |
CN103473122A (en) * | 2013-08-21 | 2013-12-25 | 上海交通大学 | Workflow system resource scheduling method in cloud computing environment |
CN103970612A (en) * | 2014-05-07 | 2014-08-06 | 田文洪 | Load balancing method and device based on pre-division of virtual machine |
CN104023042A (en) * | 2013-03-01 | 2014-09-03 | 清华大学 | Cloud platform resource scheduling method |
CN105468460A (en) * | 2015-12-02 | 2016-04-06 | 上海电机学院 | Method for monitoring and distributing virtual resources in mixed cloud environment |
CN106445675A (en) * | 2016-10-20 | 2017-02-22 | 焦点科技股份有限公司 | B2B platform distributed application scheduling and resource allocation method |
CN109871993A (en) * | 2019-01-31 | 2019-06-11 | 上海天好电子商务股份有限公司 | The pending function optimization method of government system and terminal based on banker's algorithm |
CN109918169A (en) * | 2019-01-07 | 2019-06-21 | 广东时汇信息科技有限公司 | A kind of cloud experiment training system based on OBE mode |
CN113112139A (en) * | 2021-04-07 | 2021-07-13 | 上海联蔚盘云科技有限公司 | Cloud platform bill processing method and equipment |
CN113497814A (en) * | 2020-03-19 | 2021-10-12 | 中科星图股份有限公司 | Satellite image processing algorithm hybrid scheduling system and method |
-
2011
- 2011-03-28 CN CN2011100756388A patent/CN102708003A/en active Pending
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102981890B (en) * | 2012-11-30 | 2015-10-28 | 华南理工大学 | A kind of calculation task in Visualized data centre and virtual machine deployment method |
CN102981890A (en) * | 2012-11-30 | 2013-03-20 | 华南理工大学 | Computing task and virtual machine deploying method within a virtual data center |
CN104023042B (en) * | 2013-03-01 | 2017-05-24 | 清华大学 | Cloud platform resource scheduling method |
CN104023042A (en) * | 2013-03-01 | 2014-09-03 | 清华大学 | Cloud platform resource scheduling method |
CN103473122B (en) * | 2013-08-21 | 2017-01-25 | 上海交通大学 | Workflow system resource scheduling method in cloud computing environment |
CN103473122A (en) * | 2013-08-21 | 2013-12-25 | 上海交通大学 | Workflow system resource scheduling method in cloud computing environment |
CN103970612A (en) * | 2014-05-07 | 2014-08-06 | 田文洪 | Load balancing method and device based on pre-division of virtual machine |
CN105468460A (en) * | 2015-12-02 | 2016-04-06 | 上海电机学院 | Method for monitoring and distributing virtual resources in mixed cloud environment |
CN105468460B (en) * | 2015-12-02 | 2018-11-09 | 上海电机学院 | Virtual resource monitoring and distribution method under a kind of mixing cloud environment |
CN106445675A (en) * | 2016-10-20 | 2017-02-22 | 焦点科技股份有限公司 | B2B platform distributed application scheduling and resource allocation method |
CN109918169A (en) * | 2019-01-07 | 2019-06-21 | 广东时汇信息科技有限公司 | A kind of cloud experiment training system based on OBE mode |
CN109871993A (en) * | 2019-01-31 | 2019-06-11 | 上海天好电子商务股份有限公司 | The pending function optimization method of government system and terminal based on banker's algorithm |
CN113497814A (en) * | 2020-03-19 | 2021-10-12 | 中科星图股份有限公司 | Satellite image processing algorithm hybrid scheduling system and method |
CN113112139A (en) * | 2021-04-07 | 2021-07-13 | 上海联蔚盘云科技有限公司 | Cloud platform bill processing method and equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102708003A (en) | Method for allocating resources under cloud platform | |
US9542223B2 (en) | Scheduling jobs in a cluster by constructing multiple subclusters based on entry and exit rules | |
CN103970609B (en) | A kind of cloud data center method for scheduling task based on improvement ant group algorithm | |
Polo et al. | Performance-driven task co-scheduling for mapreduce environments | |
Vecchiola et al. | Deadline-driven provisioning of resources for scientific applications in hybrid clouds with Aneka | |
CN102063336B (en) | Distributed computing multiple application function asynchronous concurrent scheduling method | |
CN103870314A (en) | Method and system for simultaneously operating different types of virtual machines by single node | |
CN103023963B (en) | A kind of method for cloud storage resources configuration optimization | |
CN108701059A (en) | Multi-tenant resource allocation methods and system | |
CN104881325A (en) | Resource scheduling method and resource scheduling system | |
CN107346264A (en) | A kind of method, apparatus and server apparatus of virtual machine load balance scheduling | |
CN105446816B (en) | A kind of energy optimization dispatching method towards heterogeneous platform | |
CN103365726A (en) | Resource management method and system facing GPU (Graphic Processing Unit) cluster | |
CN103297499A (en) | Scheduling method and system based on cloud platform | |
Li et al. | An effective scheduling strategy based on hypergraph partition in geographically distributed datacenters | |
Sun et al. | Research and simulation of task scheduling algorithm in cloud computing | |
CN110311965A (en) | Method for scheduling task and system under a kind of cloud computing environment | |
Yang et al. | Multi-policy-aware MapReduce resource allocation and scheduling for smart computing cluster | |
CN110084507B (en) | Scientific workflow scheduling optimization method based on hierarchical perception in cloud computing environment | |
CN107992351B (en) | Hardware resource allocation method and device and electronic equipment | |
Li et al. | Cost-efficient fault-tolerant workflow scheduling for deadline-constrained microservice-based applications in clouds | |
CN105389196A (en) | Method for processing seismic data by utilizing virtualization technology | |
Biswas et al. | A novel resource aware scheduling with multi-criteria for heterogeneous computing systems | |
Malathy et al. | Performance improvement in cloud computing using resource clustering | |
Wang et al. | Maximize throughput scheduling and cost-fairness optimization for multiple dags with deadline constraint |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
DD01 | Delivery of document by public notice |
Addressee: Liu Wenjuan Document name: Notification of Publication of the Application for Invention |
|
DD01 | Delivery of document by public notice |
Addressee: Liu Wenjuan Document name: Notification of before Expiration of Request of Examination as to Substance |
|
DD01 | Delivery of document by public notice |
Addressee: Yan Deying Document name: Notification that Application Deemed to be Withdrawn |
|
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20121003 |