CN102681899A - Virtual computing resource dynamic management system of cloud computing service platform - Google Patents

Abstract

Cloud computing is the third change of information technology (IT), and deeply affects information acquiring modes of the information-based society. A cloud computing service platform converts computing, network and storage into a resource pool to provide service for a user through the network. According to a virtual computing resource dynamic management technology of the cloud computing service platform, a layered cloud computing infrastructure model is provided; three kinds of objects are defined to describe and manage statuses, attributes, characteristics and behaviors of a telescopic virtual machine resource; the supply from a physical server to the virtual resource pool is automated; and an online dynamical capacity-expanding method is provided, and through real-time acquisition of load status and application running conditions of the system, the virtual computing resource is dynamically expanded under the condition that the service is not interrupted by taking the artificial neural network as a mechanism for judging expansion and triggering of the resource.

Description

The virtual computational resource dynamic management system of cloud computing service platform

One. technical field

Cloud computing is the trend of the times of internet development in the change for the third time that is considered in the IT field after microcomputer, internet.

Cloud computing is the development along with Intel Virtualization Technology, CPU multi-core technology, broadband technology, concurrent operation, Distributed Calculation, grid computing; The continuous expansion of internet scale; The user constantly increases the demand of information, network, calculating, storage, technology and service that service provider grows up for adapting to this demand gradually.

Cloud computing is through will be as ins and outs such as IT information service provider's server, operating system, network, storage, basic software, platform, framework shielding and abstract in can run service forms such as manageable computing power, storage capacity; Dynamically offer information user through network, the user pays by actual use amount.

Cloud computing has following key point: 1) resources shared and share technique framework 2) have distributed computation schema and a memory module; 3) through virtual statistic multiplexing and the flowable resource that realizes data center's hardware resource; 4) the IT resource service of the different stage that comprises software and hardware facilities can be provided for the user; 5) resource dynamic expansion is supplied with as required, charges according to quantity.6) intelligentizedly be applied to resources allocation; 7) application deployment of robotization and resource distribution.

Cloud computing now not merely has been a kind of technical term or Technical Architecture, a kind of especially new service mode and business model.Cloud computing is through doing integration with server, network, storage, various operating system, the software of technical field; And further the data center in the company in the industrial field, unit and the enterprise, equipment, personnel, service etc. are integrated; Make information service accomplish industrialization, scale, specialization, intensification; Promote the utilization rate of information basic resource and the efficient of information service greatly, saved social cost, resource and consumption.

Two. background technology

The core technology of cloud computing has comprised Intel Virtualization Technology, distributed computing technology, cloud platform management technology

2.1 Intel Virtualization Technology

Swift and violent since 2006 based on the virtual development of increasing income of Xen, make Intel Virtualization Technology progressively begin extensively to popularize.Because Intel Virtualization Technology can be installed a plurality of operating systems on a station server, thereby has improved the service efficiency of server greatly.

The logical expressions of virtual representative resource do not receive the restriction of physical resource in theory.The way of realization of Intel Virtualization Technology is in operating system, to add a virtualization layer, and the resource of lower floor is abstracted into the resource of another form, offers the upper strata and uses.Virtual and I/O virtual of virtual, the internal memory of CPU arranged in practice.

This technology though IBM just began one's study as far back as the seventies; Begin in the nineties from VMware but realization is virtual on the X86 framework; Because it is virtual that X86 is difficult to realize; This is because the ISA framework of X86 has 17 responsive instructions not belong to privileged instruction, carries out the instability that these instructions agree be able to cause total system if that is to say virtual machine.

So Vmware has released fully virtualized commercial version product; Promptly need not to revise the operating system of client computer; Through the privileged instruction of intercepting and capturing the client computer execution and the binary code that scans the execution of client computer kernel, become virtual instruction to carry out responsive instruction translation.

Occurring later in 2005 is half virtualization product of increasing income of representative with Xen; Through revising the code of Guest OS; Make its operation that those are relevant with privileged instruction all conversion can issue the Hypercall (hypercalls) of VMM; And Hypercall supports batch processing and asynchronous these two kinds of optimal way, the feasible speed that can obtain being similar to physical machine through Hypercall.

Intel and AMD have released the virtualized product of support hardware after 2007; Through introducing new cpu instruction and operational mode; Let virtual machine and VMM run on different processor modes; Like Root pattern and Operation pattern, when virtual machine was carried out responsive instruction, system automatically switched to the Root pattern and is carried out by VMM.

Also have one type virtual be to be the operating system virtualization product of representative with OpenVZ and Solaris Container, be about to the operating system partition of main frame, on the basis of kernel instance realization virtual, system overhead is littler.But because virtual operating system environment is not provided, many application that need to visit bottom layer driving can't move, and also can't move with the host operating system different operating systems.

Though Vmware provides many effective virtual machine management systems; But because VMware is commercial version; CPU slot number according on the main frame is collected the version expense; Need rely on magnanimity PC server that the user of service is provided like this for Internet firm, it is high using the cost of VMware.

And based on XEN or the such virtualization product of increasing income of KVM; Though the dynamic migration function of VM is also arranged; But the management system that lacks virtual machine when reply magnanimity resources of virtual machine, lacks effective management tool; Can't robotization ground distribute resources of virtual machine and supply resources of virtual machine, lack also that to call resources of virtual machine with the interface of using be application service.

2.2 distributed computing technology

Distributed treatment is a kind of mode of information processing, is a notion relative with centralized processing, and it couples together through many computing machines that communication network will be dispersed in various places, under the management control of control system, accomplishes information handling task in phase.Distributed treatment is usually used in mass data is carried out analytical calculation; It is data and distribution of computation tasks different computing machine to the network; These computing machines are the common calculation task of accomplishing under the scheduling of controller, and the performance of distributed treatment depends primarily on the communication efficiency of data and control.

Distributed treatment is a key link of cloud computing; It can be deployed in virtual on; Solve the collaborative work problem of the extensive server zone of cloud computing data center, form by distributed file system, Distributed Calculation, distributed data base and distributed synchronization mechanism four parts.

Distributed file system is one of architecture of DCE; It presents to the user to the file resource that is dispersed in the network with unified viewpoint; Simplified the complicacy of user capture, strengthened the manageability of compartment system, also prepared condition for further developing Distributed Application.Distributed file system is based upon on the Client basis, by server and client file systems co-operating.Control function is dispersed between the client-server, makes the thing that in centralized file system, is easy to handle such as shared, data security, the transparency etc. become quite complicated.File-sharing can be divided into read to share, sequential write is shared and concurrently write sharedly, sequential write need solve the same viewpoint problem of sharing users in distributed file system, concurrent write to insert in the middle of then need considering upgrade the consistency problem that causes.Aspect data security, the data that need to consider the private ownership of data and conflicting when are recovered.Transparency demand file system is unified complete to users'interfaces, need guarantee location transparency at least, concurrent visit is transparent and fault is transparent.In addition, extendability also is the problem that distributed file system needs emphasis to consider, when increase or minimizing server, and distributed file system perception automatically, and the user is not had any impact.The example of distributed file system has GFS, HDFS, Ceph, Lusture, glusterFS, NFS etc.

Distributed data base is one group of structurized data set, belongs to same system in logic, and a plurality of grounds that physically disperse the active computer network to connect, and unified by a distributed data base management system (DDBMS) management.Compare with centralized or separate data storehouse, distributed data base have the expansion of reliability height, module easily, advantage such as little, the load balancing of operating lag, fault-tolerant ability be strong.Data redundancy is that distributed data base is different from one of principal character of other databases, and it has guaranteed the reliability of distributed data base, also is the basis that walks abreast.Along with the development that with SNS is the web2.0 dynamic website of representative, high concurrent, mass memory and the high bottleneck that has been extended to the traditional relational database, the non-relational database NoSQL that addresses these problems grows up gradually.The example of distributed data and NoSQL database has BigTable in a word, HBase, MongoDB, Redis, Tokyo Cabinet, CouchDB, Cassandra etc.

Distributed Calculation be let several physically independently assembly as an independent system synergistic working.For distributed programmed, core problem is how to become some subroutines that can parallel processing to a big Application procedure decomposition.Have two kinds can treatable method, a kind of is to cut apart calculating, promptly is divided into several modules to the function of application program, by many collaborative completion of machines on the network; Another kind is a partition data, promptly is divided into fritter to data set, is calculated respectively by many computing machines on the network.Large-scale distributed system can take these two kinds of methods usually simultaneously, and solves the problem of collaborative work between each functional module.This type systematic possibly adopt three layers or the multilayer distributed subject architecture that is the basis with the C/S structure, is distributed in presentation logic, service logic and data logic on the different machines, and CORBA, EJB, DCOM are the middleware Technology of three kinds of main flows.Also possibly adopt the architecture based on Web, or claim Web Service, this model is the Enterprise SOA of height profile, and SOAP, WSDL, UDDI are the core technologies of Web Service.

In distributed system; The inconsistent problems of data such as the parallel work-flow to shared resource may cause lost update, reads dirty data, non-repeatable read; At this moment need introduce distributed synchronization mechanism concurrent operations is controlled, most common form is distributed lock mechanism and collision detection.Several kinds of concurrency control methods commonly used are arranged:

Concurrency control method based on lock mechanism

Concurrency control method based on timestamp

Optimistic concurrent control (Optimistic Concurrency Control) method

Concurrency control method based on version

Concurrency control method based on transactions classes

For the cloud computing data center that is made up of extensive low-cost server crowd, distributed synchronization mechanism is the basis of carrying out all upper layer application, is the basic assurance of system's correctness and reliability.Google Chubby and Hadoop ZooKeeper are typical case's representatives of cloud architecture distributed synchronization mechanism, are used for each parts of coherent system, and other distributed systems can be come the synchronization of access shared resource with it.

2.3 the cloud platform management is technological

The major function of cloud computing is that computational resource, storage resources, Internet resources are managed effectively, offers the user with the form of serving.So singly do not need Intel Virtualization Technology and distributed computing technology as the basic fundamental element, offer the service platform that the user calls computational resource as one, more need handle the integration of virtual resource and physical resource; Solve resource demand assigned problem, the problem of dynamic expansion, the problem of automatic management; The problem of application deployment; The problem of data management, and problems such as user management, carrying out safety backup, this just needs a perfect cloud platform management system to accomplish.

Technology such as the Intel Virtualization Technology that more than relates to, Distributed Calculation, distributed storage all are the basic fundamental assemblies of cloud computing; But to large-scale network system, database, basic software etc. be offered the user as computing power, storage capacity etc. with the form of serving; The administrative skill that also needs the cloud platform realizes the deployment of robotization, dynamic expansion, the support of using, visit of data or the like with these assembly organization and managements.

2.3.1 the service mode of cloud platform and COS

And cloud computing is a kind of service mode of innovation, and the service mode of cloud platform is according to can be divided into three kinds of IaaS, PaaS and SaaS according to the technical system framework:

IaaS

The bottom of cloud computing platform is an IT infrastructure; To be exactly service provider comprise that with IT infrastructure server, network, storage, operating system become the manageable virtual resource of a kind of program after virtual, calculate and stores service for the user provides through network to Infrastructure as a Service.The EC2 of typical example such as Amazon, Gogrid and Rackspace also are cloud computing IaaS service providers.

PaaS

The PaaS platform is the platform of a kind of exploitation and trustship network application; This platform provides several kinds of network programming language and framework; The user can use this programming language and framework to write the application that can expand and stretch, in cloud computing platform, moves, and need not the restriction whether the consideration program receives resource.This generally is that service provider designs the application call that a kind of programming framework API lets the user write and operates on the own inner cloud computing facility.The Apps Engine of Google for example.

SaaS

Software is promptly served, Google Gmail for example, Google Docs.This is a kind ofly through the internet pattern of software service to be provided, and on the server of oneself, the user need not download and install this locality with the application software unified plan in manufacturer, can use through browser.The user can be according to own actual demand, by the service of ordering what and time length to manufacturer's defrayment.The user need not buy software again; And need not software is safeguarded; The service provider understands full powers and administers and maintains software; Software vendor also provides the off-line operation and the local datastore of software, the software and services that lets the user can use it to order whenever and wherever possible when the client provides internet, applications.

Cloud computing platform can be divided into privately owned cloud and publicly-owned cloud again according to COS.In enterprises is fire wall with deployed, for enterprises IT is privately owned cloud with producing what service was provided; Dispose beyond the fire wall, what public service is provided for the public is publicly-owned cloud; For the cloud computing client, privately owned cloud is arranged oneself, again a part of data and calculating are put into publicly-owned cloud platform simultaneously, the use pattern of this cloud computing is mixed cloud.

2.3.2 the administrative skill analysis of cloud platform

For IaaS, the management of cloud platform is unused physical resource is concentrated and after management gets up, to be considered how the virtual resource that these are abstract offers the user, and therefrom create economic benefit with Intel Virtualization Technology.Need on the one hand IT infrastructure is managed, comprise shielding hardware differences, monitors physical resource user mode, dynamic assignment virtual resource etc.; Interface with user interactions also need be provided on the other hand, comprise api interface, the configuration interface that virtual resource is provided that standard is provided, provide service catalogue to supply the user to search available service, real time monitoring and statistical function etc. are provided.

PaaS is served; On the cloud platform, need to dispose technology such as distributed storage, distributed data base, distributed synchronization mechanism and distributed computing model; After making platform just possess the basic capacity of distributed software development; PaaS cloud service merchant also need consider how this development platform is offered the user, as aspect the user interface, code library, programming model, DLL, development environment etc. need be provided.Basic functions such as the storage of code library package platforms, calculating, database are used when supplying user's developing application.Programming model has determined the Application Type of user based on the cloud platform development, and it depends on the distributed computing platform of platform selecting.The PaaS operation management system need solve the supply and the problem of management of storage required in the user application operation process, calculating, network foundation resource, and needing dynamically increases or reduce running example according to the application program practical operation situation.Be to guarantee the reliability service of application program, system also need consider the mutual isolating problem between different application, lets their reliability services in the sandbox environment of safety.

As far as SaaS, because service itself makes up on the internet, the user possesses networked capabilities can online use.No matter the operation management system of any service, all need solve system management problem and user managements such as charging that product relates to, authentication, safety, monitoring in the operation process.In addition, to the difference of business characteristic, each service operation management system also need solve different problems separately.

The SaaS pattern had just existed before cloud computing occurs, but the SaaS pattern does not open up new prospect in China all the time yet.A lot of SaaS companies have all closed down, and also have only Salesforce one family more successful in the U.S..

The main flow of existing cloud computing service platform is exactly two types of PaaS product and IaaS.

PaaS service is to be representative with Google and Microsoft; Main concurrent operation, distributed arithmetic, the distributed storage technology of adopting, the client who is suitable for has large-scale processor active task, like data excavation, data analysis, search etc.; The user need learn a kind of new multiple programming method or programming tool; And use this new tool to come again coding to accomplish calculation task, though the user can shorten the operation time of application exponentially, but original application can not carry out the transition to new system smoothly.Such characteristics are user's application to be broken up rub broken being distributed to then in the resource pool, send to the server process that each is in diverse geographic location with calculating as task, obtain the result and return.

IaaS service is to be representative with Amazon and RackSpace; The main Intel Virtualization Technology that adopts; Be applicable to that single calculation task is not a super large;, single application wastes computational resource but monopolizing server, simultaneously towards numerous such application and calculation task, perhaps towards numerous such clients.Such service offers a plurality of calculation tasks or a plurality of user service with hardware server resource virtualizing subregion, and manages the resource pool that numerous virtualized servers constitutes.For the user; Original application can be moved in the environment of cloud; Again do not write application and need not learn new programming tool; Other application program need be installed yet, thereby can protect original investment, such characteristics be with the resource in the resource pool as required dynamic assignment give the application need resource.

On market, the cloud computing service of IaaS has accounted for the overwhelming majority at present.This be because, market is relatively ripe, business model is clear relatively, user's application is moved in the cloud computing environment more easily, and first kind of pattern is applicable to that more the developer directly writes new cloud and uses in cloud environment.

The IaaS service is the cloud computing pattern of on architecture, constructing, and needs the virtualization product of bottom, like VMware; Xen, KVM etc., wherein Xen/KVM is the product of increasing income; Redhat is integrated Xen and KVM; These softwares are called as HypeVisor or VMM, have realized the virtual machine resource management on the separate unit physical machine, comprise virtual management and the Virtual Machine Manager of CPU, internal memory, I/O.But how numerous virtualized servers and network, storage etc. are managed as a resource pool, constituted a cloud computing platform, intelligent automaticization ground is for using the resource that provides dynamic on-demand to distribute; Independent virtualization software is insurmountable; This needs the management software of a cloud computing to accomplish, and perhaps is called the cloud platform management system and accomplishes, and this respect does not have standard; Do not have ripe product and scheme, rare successful case can be used for reference.

As seen from the above analysis; The PaaS platform service is the most complicated; Both needed large-scale data center and network system platform to support; Perfect distributed application program development environment, DLL, code library, programming model, distributed data base etc. need be provided again, and having only Google and Microsoft really to have the ability to provide with Apps Engine at present is the PaaS service of core.Domestic also only have Sina to test SAE.

The PaaS platform service to as if the application developer, audient's face is less, and the developer need be from the new multiple programming mode of new study, more increases to use difficulty.

The business model discomfort of PaaS is very clear, says for the COS of Apps Engine, and chargeing according to request call amount, CPU use amount application programs is difficult to realize, and owing to the developer towards the minority, real income is not high yet.

The real success of cloud computing operation, that clear profit model is arranged is IaaS, is representative with Amazon, RackSpace, many companies such as GoGrid are this pattern of self-replication all.And most users is in EC2 that uses Amazon and S3 service on the present cloud computing service market; For example maximum in the world video display lease service merchant Netflix made the transition for the mode with video flowing be that the user provides VOD and download service; And they also from original self-built from O&M data center to the cloud service of using based on Amazon EC2, because EC2 can provide linear expansion.

2.3.3 the IaaS platform management is technological

Amazon EC2 is typical IaaS service, but the virtual machine of installing operating system is provided, and the user can own set up applications, also can use AMI (Amazon Machine Image), can make up the customized application of a high complexity with Amazon EC2.EC2 allows user control environment parameter, underlying operating system, storage and network demand, says that technically it belongs to the very service of bottom, and the user can adjust needed most of thing.

Service provides a platform to the EC2 of Amazon as IaaS, and its service provides calling of resources of virtual machine and configuration service, but calling interface is a series of api routine interfaces of Amazon oneself definition.Many third party's platform development merchants can develop the third party and use on the Amazon platform, realize that the dynamic expansion of resource etc. are disposed in the robotization of user application.If the user selects then need learn the api routine interface of Amazon EC2 service from service, own again compile script application also needs user oneself to accomplish with establishment virtual machine environment, the deployment of application program.

Also have Joyent, Gogrid, Rackspace, Savvis, the CloudEx of Terremark and A-1. Net etc. with the similar IaaS cloud of Amazon EC2 platform.Because the interface and the communication standard of lack of uniform, the interoperability between these platforms is very poor, the user can't be between IaaS service provider seamless migration.Abiquo company has released the cloud computing platform AbiCloud that increases income; This platform provides unified cloud core; Can be used for creating publicly-owned cloud and privately owned cloud with extended capability, the various interface that is used for integrated third party's instrument and software also is provided, the various clouds of particular demands are satisfied in the construction of assistance services provider; In the IaaS standardization and the effort of making aspect increasing income attempt, but that these trials are not sufficient to guarantee the IaaS operation management system is available conveniently.The service provider needs as required, selects research and development voluntarily or on the existing project foundation of increasing income, makes up the IaaS operation management system that satisfies enterprise's particular demands.

The basic function that these IaaS platforms itself are realized is that a better simply virtual machine distributes; Even but its inner system realizes also very complicacy like this, Amazon never disclosed the realization details of its background system, but some open source software is claimed the repertoire of having realized Amazon EC2; Such as OpenNebula; Nimbus, OpenStack, Eucaptus.

In a word, the management of cloud computing service platform is a complicated problems, and industry does not also form relevant standard at present, does not have to bring the system that direct deployment is used yet, and cloud service provider needs to realize separately.

2.3.4 the increase income product and the problem thereof of IaaS platform management system:

Xen?Cloud?Platform

Xen Cloud Platform is the project by the developer community initiation of the hypervisor Xen that increases income; Be devoted to offer complete Infrastructure platform of service provider, possess open and measured API, multi-user's lease; SLA ensures, the characteristics of chargeing as required.This project originated in 09 year, was at present in the quick exploitation, and system is not mature enough, and comparatively complicacy of configuration is installed.

Eucalyptus

Eucalyptus is the software that is used to realize cloud computing part infrastructure, lets the user make full use of the server zone of oneself.Eucalyptus only realizes basic simple Virtual Machine Manager function, and its main target is not the powerful of management function, but with basic virtual management APIization.It has following characteristic: with EC2 interface compatibility (main interface) with Rocks cluster management instrument; Install and dispose all very simple and safe intercommunication; Use the working method of SOAP and WS-Security superposing type; Need not basic " cloud management " instrument of modifying target Linux environment, realization system and nusrmgr.cpl can become single cloud with a plurality of cluster configuration.

Nimbus

Nimbus is the cloud computing project of increasing income under the grid middleware Globus, and Nimbus is towards the science computation requirement, realizes that through one group of instrument of increasing income infrastructure promptly serves the cloud computing solution of (IaaS).In the Nimbus platform; The assembly that comprises has: the Workspace node manager, based on the remote protocol of WSRF realize, remote protocol realizations based on EC2, cloud computing client, WorkspacePilot integration virtual machine etc. are towards the application component of different aspects; Each assembly of Nimbus project is very lightweight and possess self completeness in design, can make up through multiple isomery mode.

OpenNebula

OpenNebula is that construction is privately owned, public, the instrument of increasing income of mixed cloud.Framework, interface, assembly flexibly are provided.Key property comprises: support Xen, and Kvm, Vmware, virtual platform can dock EC2 and ElasticHost, uses libvirt, supports EC2 and OGC OCCI interface

Libvirt

Libvirt is that a cover is free, the C function library of the virtual instrument of main flow under the support Linux that increases income, and it is intended to is that the various virtual instrument that comprises Xen provides a cover convenient, DLL reliably, supports the binding with multiple main flow development language such as C.All exploitation forms based on libvirt for the virtual management instrument virt-manager (graphically) that gives tacit consent on the current main-stream linux platform, virt-install (command mode) etc.Characteristics: primary C language interface, and the binding interface of other active languages, meet CIM and QMF standard, support Xen, Qemu, KVM, LXC, OpenVZ, UML, Virtualbox, VMWare ESX, but telemanagement virtual machine, virtual network.

These open source softwares have been realized the most function of Amazon EC2 really, and its essence still is the distribution of computational resource and storage resources then, user's service certainly; User's management and authentication etc.; And do not have really to realize the dynamic expansion of the desired automatic management of cloud computing platform, resource, the functions such as robotization deployment of application, at the Amazon platform, these all rely on third party's platform development merchant to provide; Like RightScale etc., but the software of these companies all is business-like.The user can not freely obtain.

2.3.5 the commercially produced product of IaaS platform management system and problem thereof:

The commercial version of cloud plateform system management software has: VMware vSphere/vCloud, Dell VIS, Platform ISF, Heroku, RightScale, BMC, IBM, HP, OpSource etc.

VMware

It mainly is vSphere and vCloud that VMware is used for virtual product with the cloud computing platform management.

It is virtual that vSphere is mainly used in server end, plays purposes such as Server Consolidation and resource optimization through on a physical server, fictionalizing many virtual machines.VSphere can be two parts: first VMM (virtual management device Hypervisor) part, and it two is the vCenter that are used to integrate and manage VMM, and major function has resource and virtual machine inventory management; Task scheduling; Log management, warning and practice management, deploying virtual machine be provided with etc.

VCloud Director is the cloud computing plan of VMware, and two branches are arranged, the IaaS solution of first VMware vCloud Express by name, and it two is PaaS solutions of VMware Platform as a Service by name.

VCloud Director is based on the virtualization capability of VMware vSphere; And expanded the resource pool function of VMware vCenter so that " VDC (Virtual Data Center; virtual data center) " can create in IT department; Resource pool and predefined operating strategy, SLA and the pricing mechanism promptly formed by calculating, network and storage resources, and for the user provide based on the computational resource of VDC with can application deployment on it.VCloud support resource in design isolates and many tenants, and vCloud has introduced two very notions of core for this reason: the first is used for VDC that resource is isolated; It two is the tissues (organization) that are used to support many tenants' mechanism.

VDC is a set that comprises resources such as the calculating that is used for cloud computing and storage, and the keeper is combined into same tissue through rule (Policy) with a plurality of users.

The maximum businessman of share is exactly VMware on the cloud platform management system market at present, but its product line is a lot, in order to protect its virtualization product in the past; Its new product towards cloud computing data center all increases on existing product; Platform product and data center products are also separated, and the division of the isolation of resource and tissue all needs manually-operated, and function can be very complicated; Needing special VMware managerial personnel could manage, and this has all increased the IT handling cost.

Dell?VIS

The VIS of Dell framework can the dynamic adjustments application load also be integrated into unified resource pool with calculating, storage and networked asset.The VIS framework is made up of modular assembly, can be integrated each other with the existing IT environment of client.The VIS framework comprises following three primary clusterings:

● senior infrastructure manager (AIM); Support the keeper according to application load distribution server, storage and Internet resources; AIM can extract hardware and virtualization layer from data center, and therefore, the client can be absorbed in the single resource pool of configuration and the multiple different techniques of non-management.

● the self-service creator of VIS, can standardization and automatically perform the deployment flow process of application, and then the deployment time of operating load is foreshortened to a few minutes.The self-service creator of VIS can strengthen the IT control, quickens the IT flow process simultaneously, thereby saves time and resource.

● Dell VIS Director, support that the user checks the problem in virtual dependence and the quick identification virtual environment comprehensively.This module comprises advanced report, suppose and trend analysis, capacity and utilization rate report, absorption of costs and refuse to pay solution.Dell also is important producer of cloud computing platform management software field, and VIS can be responsible in the dynamic adjustments application, but its hardware robotization deployment can only run on the hardware platform of Dell oneself, does not support the hardware platform of isomery.

Platform?ISF

Platform has also released cloud computing platform management product ISF in the recent period; Be the robotization construction to the privately owned cloud of enterprises, can integrated various application environments such as J2EE, webSphere; Test/Dev or the like; And multiple resource allocation policy is arranged, and key business can reserve resource, and product itself is done finely.But Platform has done science and calculates the HPC field since the more than ten years in the past; Just forward the cloud computing field recently to; So his product has the vestige that science is calculated naturally, be assigned in the different virtual machines like the control application load, this can be used in fact, and the LSF even load is balanced to be substituted.

More multiaspect is to the enterprises large-scale application for this product, rather than internet web uses.

The price of ISF is very expensive in addition, collects the version expense of ＄ 795 U.S. dollars according to each CPU slot of server.For Internet firm, mostly adopt the cheap PC server of magnanimity, this price is that the no one can bear.

BMC

BMC releases a cloud computing life cycle management scheme, comprises the service catalogue that strategy promotes, and service is customized, and is used for the Self-Service of cloud computing resource request and control; Be suitable for the architecture of privately owned cloud and public cloud service, for comprising calculating, network, storing, be applied in interior whole service stack and carry out dynamic-configuration; Flow based on the secure network service of performance classification and the i.e. cloud management work of usefulness; The performance performance of performance monitoring instrument supervision cloud service; Realization flow based on ITIL.

Cloud computing life cycle management scheme is under the jurisdiction of BMC cloud service Managed Solution series; Cloud platform through this scheme foundation; Can realize the robotization to many tenants cloud architecture physics and virtual dividing, load balance, fire wall and service level management can be connected on each container automatically; Can realize " Self-Service " cloud managing portal, and simplify the operation, seamless integrated planning, allotment and management public and privately owned cloud architecture.

Flow process is built up but BMC does software operation; So his cloud computing management software lays particular emphasis on the business service workflow management more; What the boss of BMC said also that BMC does is the management of cloud computing; That does does not provide cloud computing, but the physics of integration and virtual computational resource must will be arranged earlier, just can have the service of cloud computing to provide.So the software of BMC is to having used like the client's of Amazon EC2 cloud computing; Promptly manage the Amazon EC2 of use, so this cover software is not to the user that will develop and manage cloud computing infrastructure and cloud service is provided with the cloud management software of BMC.

Domestic from research and development: Huawei, A-1. Net, big fortune, tide

Also numerous and confused lofty tone such as domestic a lot of IT giants such as Huawei, association, tide, Founder, dawn, Alibaba, A-1. Net sets foot in the cloud computing industry at present; Drop into huge fund and attempt to occupy a tiny space, a lot of companies even tell whole family company separately and come specially to do this business in the cloud computing industry of China.

Huawei has developed a cover cloud computing platform management software on the basis of open source software Eucalyptus, just attempting now to cooperate with telecommunications three big operators with China Mobile, UNICOM, releases their cloud computing software and hardware solution.Tide and dawn etc. be also in exploitation cloud operating system, contains virtual, resource management, task scheduling, load balancing, task management, safety certification, charging etc., but also on market, do not release.

What can see on the market is the cloud cable release of A-1. Net, and this product is still semi-automatic at present, and is also very immature aspect user self-help service, resources configuration management, dynamic expansion.

The cloud computing platform management system is done by open source software by most of company of China at present; Because this market is immature, what the client do not meet yet is the product of cloud computing, so be easy to the market at the initial stage of obtaining; But open source software also has a lot of problems; These several open source softwares all are the basic functions that can only provide virtual machine to distribute, and distribute automatically and the resource dynamic expanded function based on the resource of strategy, also do not have the robotization supply of hardware platform.And these open source softwares have made the less functions of using such as a lot of network pool and storage pool, yet in practical application, network pool can rely on LVS to realize, storage pool also can lean on distributed file system to realize.

And open source software itself can not directly bring the through engineering approaches software of doing enterprise operation, and bug is a lot, does not pass through pressure test, lacks the mechanism of managing risk and handling problem, and reliability is not high, can't bear the through engineering approaches task.The community itself that increases income simultaneously always can not provide technical support timely and help yet.

For example the system that on market, pushes away now of Huawei is based on Eucalyptus and does not just increase income fully, has a lot of Premium Features just not increase income such as load balancing and expansion.

In a word in the market business software and based on the software of increasing income such-and-such problem is arranged all: lack automatic management; Homogeney is high, and function is too complicated, needs too many human configuration; Lack the robotization supply; Lack the resource dynamic expansion, external commercialization cloud computing management software is not Chinesizing also, and domestic software does not also develop.

Three. summary of the invention

This invention is the virtual calculating money dynamic management system of a cloud computing service platform, has solved the above-mentioned system that increases income and the variety of issue of commercial system, and supply, configuration, distribution, expansion, the management of virtual resource can be provided.Be absorbed in quick deployment, supply, the resource of cloud computing resource layer-management, resource high effective integration and share, the elastic telescopic of resource, dynamic trigger mechanism, the linear expansion of resource pool.

3.1 characteristics:

● can generate shared physical resource pond, between different departments and different application, share

● the layer architecture model of corresponding service logic

● support the virtual drivings of main flow such as Xen, KVM, VMware, OpenVZ, adopt Libvirt to call

● support virtual meter operating systems such as Windows, Solaris, Linux

● support Kickstart, SystemImager, TFTP, PXE, agreements such as DHCP

● generate flowable virtual resource pond, eliminate Single Point of Faliure, increase stability

● intelligentized resource reduces artificial auxiliary to using allocation strategy.

● the physical server of robotization has strengthened the agility of service response to the fast supply in virtual resource pond.

● the application deployment of robotization

● the resource dynamic expansion can not influence the service of application program.

● adaptive dilatation trigger mechanism, scalable activation threshold value.

● the application program load balancing is integrated

● distributed buffer memory

● dissimilar application programs is integrated, J2EE, HPC, Test/Dev

● administration interface is succinct, and command line interface is provided.

3.2 function:

Provide patterned friendly interface let user oneself definition or select required computational resource, the user can build an IT of enterprise or internet, applications framework with the mode that mouse pulls on the cloud platform.

System can dynamically generate and pay the computational resource of meeting consumers' demand in several minutes.

The automatic analysis user demand of cloud platform management system, the strategy good according to predefine distributes resource requirement.

The user need not to install the configuration operation system, system's installation of complete operation system in all empty machines.

The user need not oneself configuring application program is installed, and management system is disposed the application program that the user submits to automatically, and load balancing service, rank management, Account Administration, fault-tolerant management, backup management etc. are provided.

The cloud plateform system provides basic IT services such as buffer memory service, web service, mail service, directory service, VPN service for all users.

The cloud plateform system provides dynamic expansion service as required for the user, and user's written program need not to consider the resource scaling problem, the load pressure problem that mass user concurrency and flow bring.

The user need not to learn new programming tool and programming framework, and prior application program can seamlessly be moved on the cloud platform.

The cloud plateform system can be accomplished hardware server initialization and virtual automatically, and configuring virtual machine is installed automatically, forms the virtual resource pond.

System realizes automatic management, has reduced manually-operated.Allocated resource hierarchical levels and system dynamics expanding policy when only needing initially install in system, in the service afterwards, but system robotization ground distribute and extended resources, the interpolation physical server get final product in resource pool and the operator only needs.

System has improved server operational efficiency and service efficiency in a word, has strengthened platform stable property, has improved the automatic management ability, has reduced the service response time.

3.3 principle of work:

The technological frame figure of a typical internet, applications is as shown in Figure 1.The specific function that on behalf of batch processing, each square frame accomplish.All be on the same service logic level at the square frame on each horizontal direction, the logic goes of whole application is from top to bottom by shown in the direction of arrow.Large-scale internet enterprise or other IT enterprises all are a plurality of identical like this or similar application deployments to be arranged on such framework, and the program assembly of each application is all bound specific server resource.

According to the logic flow of business, this framework can abstractly be shown in Figure 2, promptly is divided into following different level to resource so.

The server hardware resource is divided into a plurality of logical levels through virtual back according to predefined strategy.

Each level is abstracted into the resource pool that a virtual machine is formed.

The resources of virtual machine configuration of each level is all inequality.

Virtual machine configurations all within the level are identical.

Each level is all supported different service, and each business is the resources of virtual machine pond of sharing this level in specific level.

Each business has the trend of logic business and data at vertical level.

3 kinds of objects of definition in system, Aggregate, Vol, vTask.

Aggregate is to liking a kind of polymerization of representing all similar virtual machines (VM), the virtual machine set of the level that each Aggregate is corresponding as shown in Figure 3.

Vol is the virtual machine set of a certain business of representative in the distribution of this level to liking the sub-set object on Aggregate.

The vTask object is represented the set of virtual business required vol on each level, has comprised the service logic flow process.

Relation between the three is as shown in Figure 3.

The characteristics of each object:

Aggr：

● Aggr is the object of the polymerization of a kind of same type virtual machine of representative, and each virtual machine in polymerization all has identical attribute, such as identical CPU quantity and frequency, memory size, hard-disk capacity etc.

● when system generates the instance of Aggr, only need to specify a series of physical machine names with identical virtual machine configuration, system is about to the instance that continuous VM on these physical machine distributes to Aggr.

● the bottom of Aggr is a physical machine.

● Aggr can manually expand by adding machine.

Vol

● Vol is the container of a kind of loading VM in logic, can change the size of container arbitrarily through increasing or reduce VM, so do not receive the hardware constraints of physical machine.

● the size of Vol is decided by resource requirements such as how many data volume transmission of application need, how many data storage, rather than is decided by that how many virtual machines are arranged on the physical machine.

● the VM in the Vol is distributed on the different physical machine, and is as shown in Figure 4, can be to different machines, and a machine is when breaking down with assignment of traffic, and an also traffic affecting part only, and can not influence whole service.

● Vol can automatic online increase and minimizing.

● Vol produces for certain specifically is applied on certain Aggr instance.

vTask

● the corresponding application of vTask

● vTask is made up of the vol that is distributed in the different Aggr

● vTask defines the logic goes of the vol in the different Aggr.

The relation of physical machine, virtual machine, Aggr object, Vol object and vTask object is as shown in Figure 5:

Each object all has a global property in the system, is used for the metamessage of data recording in the description object.

System is used as a dead end to each VM and manages; All VM through system deployment have a data record, and record name, IP address, operating system version, CPU number, internal memory, hard disk size, VM are in the position of physical machine and name of physical machine or the like.

Each Aggr safeguards the allowable resource pond of a VM, is made up of doubly linked list; Also safeguard the chained list of a vol.

Each Vol safeguards a VM chained list that has distributed.

Each vTask safeguards the Aggr chained list of a distribution.

System also safeguards the tabulation of total physical machine chained list, Aggr and the tabulation of vTask.

System monitors each VM that belongs to Aggr with heartbeat, if VM is unreachable because server is delayed machine, this VM can be stamped unavailable mark, when VM recovers, and the system recovery mark.

When enough hour of the allowable resource pond that Aggr goes up VM, add physical machine and can expand Aggr, newly-increased VM can join the afterbody of the bi-directional list in subsequent use VM pond.

The benefit of this design is that the level of resources utilization improves flexible configuration, resource sharing.

Fig. 6 is the configuration before adopting, and each application all is limited on the physical machine, and rich resource can not be shared by the application of narrow resources or performance with performance.

The application that Fig. 7 demonstration is disposed through this new design can be shared the resource and the performance of VM polymerization.

When system realized, physical resource was by virtual resource pool for being made up of virtual machine, and virtual machine is divided into each Aggr object, on the Aggr object, produced the telescopic VOL variable of being made up of VM.On each Aggr a plurality of Vol can be arranged, but the Vol dynamic retractility, and when the Aggr space was not enough, the keeper can add VM expansion Aggr.

For example on the Aggr instance, generate the instance of vol,, on aggr1, generate the vol1 instance, require 10 VM of system assignment through order line create vol vol1 on aggr1 vm 10.Select VM to generate Vol on can be in the Aggr1 different physical machine of system.Any like this physical machine machine of delaying does not influence the application on the Vol, has increased availability.

After generating the vol object instance, the instance of we appointing system formation object vTask promptly according to demands of applications, is specified the vol of different levels.Create vTask vtask1 aggr1.vol1 for example, aggr2.vol1, aggr4.vol1.

Here we generate the instance of a vTask object, and vtask1 comprises last vol1 of vol1, aggr2 on the aggr1 last vol1, aggr4 etc.

Four. description of drawings

Fig. 1, the Technical Architecture of the typical IT of enterprise software or internet, applications.

Fig. 2, the service logic framework of layering.

Fig. 3, three expressions of object on layer architecture.

Fig. 4, the virtual machine on the Vol object stride physical machine and distribute.

Fig. 5, the relation of three objects and physical machine, virtual machine.

Fig. 6 adopts native system application model before.

Fig. 7 adopts native system application model afterwards.

Fig. 8, the graph of a relation of 5 sub-systems.

Fig. 9, the doubly linked list of the virtual machine standby pool of Aggr

Figure 10, communication between program process

Figure 11, system generates the algorithm of Vol instance

Figure 12, the algorithm of system assignment virtual machine (VM)

Figure 13, the neural networks training algorithm

Five. embodiment

The cloud platform management system is made up of 5 sub-systems, and is as shown in Figure 8:

1) resource robotization deployment system

2) virtual machine management system

3) application deployment system

4) resources allocation and management system

5) resource dynamic expanding system

5.1 resource robotization deployment system:

Resource robotization deployment system is accomplished physical resource initialization, installation, is disposed and be converted into virtual resource pool.

Robotization is disposed and also to be the Provisioning system, is that a cover is based on Kickstart's and use Perl, the system that Python and script are developed.

After switch is carried out port initialization, with physical machine access network port, behind the machine startup; Automatically through PXE guiding boot, find the network segment at place, according to the requirement that configures; Automatically select Kickstart profile, accomplish automated system and install and configuration.

VME operating system, host operating system is installed when installing; According to the configuration template configuring virtual machine that defines; The configuration application service; Enter into the virtual resource standby pool at last, automatically the virtual machine in the resource pool is divided into different polymerization clusters with predefined rule base, the corresponding level of each polymerization according to layer architecture.

5.1.2 early-stage preparations

The early-stage preparations that robotization is disposed need manually to accomplish, and comprising:

● the configuration of BIOS and firmware comprises the upgrading of firmware, the configuration of time, VT enable, Hyperthreading disable, PXE enable, Raid config or the like.

● configure the switch network port and the network segment, different application servers is connected to the different network segments.

● configure DNS, promptly add IP and server host name.

5.1.3 operating system is installed automatically

But the configuration of automatic configuration system parameter comprises host name behind installing operating system, Username, Password, IP, kernel Parameters, ulimit, disable services, enable cache etc.

This system supports Kickstart, Systemlmager, PXE, DHCP, agreements such as TFTP.

System supports the Image reflection, and leaves in the Systemlmager server.The first station server of using is installed by kickstart and is accomplished, and other server configures is accomplished robotization by Systemlmage and disposed.The time ratio Kickstart that is accomplished the robotization deployment by Systemlmager shortens 3 to 5 times, improves installation effectiveness greatly.

5.1.4 the configuration of virtual machine and installation

The Hypervisior of virtual machine selects KVM for use.

KVM requires the necessary first enable VT of server B IOS.VT is the CPU operational mode of Intel expansion, can move the sensitivity instruction of virtual machine at hardware-level.Owing to used VT; KVM can simplify its code as Hypervisior greatly; And utilize the function of Linux Kernel self to realize the environment that virtual machine requires; So KVM is smaller and more exquisite than Xen, and self is as Kernel model operation, need be as Kernel. of the other loading of Xen

The distribution of the CPU of virtual machine:

Each virtual machine can distribute one to two physical cpu kernel

The Memory Allocation of virtual machine:

According to different business and demands of applications, virutal machine memory can be divided into 4G, 8G, 12G, 16G, 24G.

The hard disk of virtual machine distributes:

The hard disk of virtual machine directly is distributed on the physical machine hard disk piece subregion, can improve 3 to 5 times of I/O read or write speeds.

The installation of virtual machine:

Physical machine and virtual machine mixed deployment are arrived physical machine in the postinstall stage that physical machine is installed with the virtual machine image copy that compresses, and decompress(ion) contracts all virtual machines are installed then, can save 5 to 10 times time.

The configuration of virtual machine:

According to rule base partition virtual machines network, CPU, internal memory and fdisk.

5.1.5 patch upgrading (patching)

The Patching system is used for unifying to do upgrade service for the various operating systems of ends of the earth data center.

The patch of system of Patching system maintenance and software and the Repository Server of bag, can from outside off the net at various software packages and patch, be that the system of varying environment provides service in Intranet.

The version of the patch that the Patching system management is dissimilar, and can be each Image patch installing or upgrading of Image server.

The Patching system selects for use mrepo Server to preserve patch file, and mrepo supports Yum and up2date.

5.1.6 the generation in virtual resource pond

The process that generates the virtual resource pond is in table, to add record, formation object and pointer chained list etc.:

1) generate physics hangar table record:

The physical machine table:

ID

Hostname

IP?Addr

?System

Version

Partition

VM?List

2) generate the virtual machine table record:

The virtual machine table:

Virtual machine:

ID，Hostname，IP，System，Version，Host?ID，Iocation，Availibility，Vol，Aggr，vTask，User，CPU，Memory，Size，Timestamp

3) through rule base, generate default Aggr object.The Aggr object is as shown in the table:

The Aggr object	Attribute
		Aggr0	LVS
Aggr1	Varnish
		Aggr2	Web
Aggr3	Java
		Aggr4	PHP
Aggr5	Application class			1
		Aggr6	Application class		2
Aggr7	Application class			3
		Aggr8	Memcache
Aggr9	The exploitation class

These are the predefined Aggr objects of system, Aggr object and attribute thereof that the also customized justice of user is new.

4), resources of virtual machine is referred to different Aggr set through rule base

Virtual machine distributes on physical machine and shows as shown in the tablely, supposes 10 physical machine, every 10 virtual machines:

Physical machine 1

Physical machine 2

Physical machine 3

Physical machine 4

Physical machine 5

Physical machine 6

Physical machine 7

Physical machine 8

Physical machine 9

Physical machine 10

P1V1

P2V1

P3V1

P4V1

P5V1

P6V1

P7V1

P8V1

P9V1

P10V1

P1V2

P2V2

P3V2

P4V2

P5V2

P6V2

P7V2

P8V2

P9V2

P10V2

P1V3

P2V3

P3V3

P4V3

P5V3

P6V3

P7V3

P8V3

P9V3

P10V3

P1V4

P2V4

P3V4

P4V4

P5V4

P6V4

P7V4

P8V4

P9V4

P10V4

..

..

..

..

..

..

..

..

..

..

P1V10

P2V10

P3V10

P4V10

P5V10

P6V10

P7V10

P8V10

P9V10

P10V10

[0246] System generates the Aggr object instance, and generates Aggr storehouse table:

ID

Attribute

The physical machine list pointer

The VM list pointer

The Vol list pointer

vTask

After generating each Aggr object and virtual machine sorted out respectively, system generates a virtual machine standby pool for each Aggr, representes with pair cycle index chained lists, and is as shown in Figure 9.All virtual machines all in standby pool, when the system assignment virtual machine, find the next virtual machine of Free list head head pointer indication at the beginning, and it is shifted out the bidirectional pointer chained list.When system recoveries is virtual, it is added the afterbody of bi-directional list.

5.2 resources allocation and management system

This system management, tracking also distribute resources of virtual machine, accept following three system's sent request.

● from the request of virtual machine management system

● from the request of application deployment system

● from the request of resource dynamic expanding system

5.2.1 receive request from virtual machine management system:

System finds first the assignable VM behind the freelist head in Aggr9, the ID of VM is returned, and this VM is deleted from the standby pool circular list of Aggr9.

5.2.2 receive request from application deployment system:

The parameter of application deployment system transmission is the number of servers of application name, application architecture level, every layer of needs.

After system receives parameter, use vTask instance of generation, insert vTask storehouse table for this; According to required framework level; For vTask distributes corresponding Aggr instance and allocation order,, on each corresponding Aggr instance, generate the instance of Vol according to every layer of required number of servers.

vTask?ID

Aggr?ID

Vol?ID

VM ID chained list

Host ID chained list

When generating the instance of Vol, use algorithm shown in figure 11.

System returns VM name and the IP address thereof in the Vol instance of Web layer at last.

5.2.3 when system receives the request from the resource dynamic expanding system:

To monitor certain application (vTask) all very high in the load of the virtual machine (vol) of certain framework layer (Aggr) when the resource dynamic expanding system; Analyze through the dilatation trigger mechanism; Decision is during dilatation, the virtual machine that can please look for novelty to resource allocation system, and the parameter of transmission is Aggr ID and Vol ID.

Resource allocation system finds the free head in standby resources pond in the instance of Aggr, find next available VM ID, judges whether the host id at this VM place exists in Vol; If exist, then look for next available VM ID, increase VM ID in Vol; Revise the VM table simultaneously; Indicate its vol, aggr and vTask return vol and give the resource dynamic expanding system.

The algorithm of realizing is shown in figure 12.

5.3 resource dynamic expanding system

The load of resource dynamic expanding system monitoring application program and virtual machine is sent to resources allocation and management system with vol ID when the decision dilatation, the latter increases VM and behind Vol, returns.

5.3.1 monitoring

Through heartbeat monitoring system accessibility; Through load at monitoring host computer and the Agent that in physical machine and virtual machine, disposes monitoring virtual machine; Comprise data such as CPU, internal memory, process number, I/O, swap, ping, serve port linking number, log, regularly collect these information.

5.3.2 trigger mechanism

To each vTask; Aggr begins from the second level, the monitor message of all virtual machines among the systematic collection vol, and whether next step will judge to vol dilatation or capacity reducing; Though we can do judgement rule of thumb for some rules of system; But the variation of various parameters is very complicated in actual motion, if be cured to rule in the program, then can't adapt to the variation of actual conditions.And when input parameter is a lot, even a lot of rules also may not necessarily the various possibilities of limit.

And artificial neural network may address this problem well, and artificial neural network is to belong to artificial intelligence field, can simulate NONLINEAR CALCULATION well.Here adopt the artificial neural network of sandwich construction, the input and output relation of network can be expressed as Y=XxW simply, and wherein X is the one dimension input variable, and Y is the one dimension output variable, and W is two-dimentional weight matrix.

Regulate the matrix full weight so that network can adapt to practical application, this process is called the study or the training of artificial neural network.Here select the training of BP (being Back-Propagation) learning algorithm; The study of BP network is a kind of study that supervision is arranged, and when network training, from actual application environment, selects training sample (data); Training data is paired input and output; To import sample data and be added to the network input end, the actual output of corresponding output sample and network is compared obtains error signal simultaneously, reconnects the adjustment of intensity with this control.

In when training, once all training samples to all reading in internal memory, every pair of training sample training network is used in circulation successively then, up to error less than defined threshold.We obtain strength of joint, threshold value and the study constant equivalence of each layer of network then.

Specific algorithm is shown in figure 13.

Here in order being judged accurately, must to receive a large amount of training samples, just to collect performance and the load parameter of lot of V M, and need the artificial result who provides dilatation or capacity reducing, can be long so collect the time of sample.But in case collect enough sample datas all, because the high-performance of modern computer, the time of training is not long.

When actual motion, system is input to the input end of neural network, network output＜-1,0,＞1 three numerical value with the Monitoring Data of each VM of collected Vol.＜-1 just represents capacity reducing, and 0 representative is kept, and＞1 represents dilatation.

If have among the Vol output of 80% VM be＞1, just decision is to the Vol dilatation.

5.3.3 dynamic capacity-expanding

When degree of declaring decision dilatation or the capacity reducing of system through artificial neural network, system increases or reduces VM to the resource allocation system application, transmits Aggr and gives resource allocation system with the Vol parameter, and resource allocation system returns the vol parameter, and VM newly-increased or minimizing is wherein arranged.

If newly-increased VM, the VM configuration that system copies keeps in advance copies among the new VM.

System finds all VM of the vol among the Aggr (i-1); Watch service or the application of each VM; If standards service such as LVS, Apache, PHP, Java; In the service configuration table, find out the configuration file of this service, the IP of the new VM name of increase, the order of appointment makes service read the configuration file of revising again in according to the service correspondence table.

For example, move ipvsadm-a-t$VIP:$port-r$RS-g to LVS

Service

Configuration file

Startup command

The service correspondence table

If application program:

1) in the application configuration table, finds corresponding configuration template, increase VM and IP record

2) Agent among the VM can watch configuration template with the mode of poll

3) in a single day find document change (version or timestamp), Agent reads configuration template again and generates new application profiles,

4) read the application configuration file again through the signal transfer mode notification applications of Inter-Process Communication, like kill-hup Pid

5) configuration file is read in application again

Concrete realization is shown in figure 10

If application program can not be read configuration file again, then must use the shared drive mode notification applications of Inter-Process Communication.

5.3.4 dynamic capacity reducing

If load reduces; Capacity reducing is judged through artificial neural network by system; Then system finds all VM of the vol of Aggr (i-1) among the iTask of vol place; Revise the configuration file of each VM with the mode of similar dilatation, last VM and IP thereof are deleted from configuration file, and with the mode notification service of similar dilatation or use and read configuration file again.

System passes to resource allocation system with the ID of Aggr and vol, and the latter revises the VM tabulation of vol, and last VM is taken away from vol, adds the afterbody of the doubly linked list of standby pool freelist head to, and returns the pointer of vol.

5.3.5 fault and recovery

Fault category has:

● the server hardware fault, damage like CPU, internal memory, motherboard, hard disk, network interface card

● the server system fault, like the operating system collapse, virtual HyperVisor collapse etc.

● the system failure of VM, like VME operating system collapse, service stopping

● the software fault that VM is last, like software bug, under attack

Unavailable classification:

● all VM are unavailable on the hardware server

● part of V M is unavailable

● a VM is unavailable

If system obtains the physical machine machine of delaying through supervisory system detecting, the storehouse table through physical machine finds its VM that has, and the available flag position of corresponding VM table and relevant VM during vTask, Aggr, Vol show is set to unavailable, and timestamp is set.

If unavailable time was above one hour; System find unavailable VM place vTask-＞Aggr (i-1)-＞last all VM of vol; According to the service configuration table, find configuration file, save as previous release; Revise configuration file with fault VM and IP deletion, read configuration file again according to the order notification service or the application of the appointment of service configuration table.

If unavailable time surpasses as 24 hours, system is the physical machine record deletion, deletes the VM pointer in VM record and the relevant chained list simultaneously.

Recover in 24 hours if unavailable time surpasses one hour, system judges whether application program and data lose, if lose, then equally with the VM record deletion; If do not lose; System find the VM place vTask-＞Aggr (i-1) of recovery-＞last all VM of vol, according to the service configuration table, find the previous release of configuration file; Cover current version, read configuration file again according to the order notification service or the application program of the appointment of service configuration table.

5.4 application deployment system:

The external application service that this system both can serve also can be dug certificate, data analysis etc. like search, filtration, data to inner application program service.

This system is through foreground web service interface; The number of servers of the application name that the collection user submits to, application architecture level, every layer of needs; Information is given to resources allocation and management system; Resources allocation and management system generate the instance of the Vol of vTask and every layer of Aggr, and return the ID. of vTask, Aggr and Vol

Application deployment system is found out web layer Aggr, and the web URL that returns VM in its vol gives the user.

The user uploads through URL and disposes and configuring application program configuration database.

System generated the configuration template reflection after the user accomplished, and deposited in vol.image.

5.5 virtual machine management system:

Virtual machine management system is for programmer's management and distribute exploitation to use virtual machine.

The development of virtual machine is managed distribution in Aggr9.

Development of virtual machine and programmer are one to one.

After system receives the request of user's submission through the web interface, generator program person's record.

System

Version

Architect

Programming

Application

Department

User

DB

Duration

VM

The workflow part and the workform management system interface of system and office automation; Produce a workflow; Arrive first the software department manager, arrive the person in charge of network system portion again, the person in charge produces a shop order after ratifying and putting on record; Arrive network system portion workform management system, the employee of network system department selects this shop order.

System manager's login management web interface is selected Next virtual machine in the subsequent use virtual resource pond from Aggr9 (development environment), distribute to the programmer.

The next VM of the standby resources pond doubly linked list Freelist head of Aggr9 deletes in system, and this VM ID is filled in the VM item of programmer's record.

When application and development completion or programmer leave office, the recyclable developing engine of system manager, system-kill programmer record adds VM ID to the afterbody of Aggr9 standby resources pond doubly linked list.

Claims (7)

1. the virtual computational resource dynamic management of a cloud computing service platform is technological, it is characterized in that:

Have: the robotization physical server is to the deliverability of virtual server, according to the resources of virtual machine layer-management of applied logic architecture definition, the smart allocation of computational resource, line dynamic extended resources.

2. aforesaid right requires 1 described system, possesses the robotization deliverability, it is characterized in that:

Have: extensive physical server passes through network DHCP, TFTP, PXE, Kickstart, SystemImager robotization installing operating system, and guides the configuration of virtual machine and the installation of operating system automatically;

, copy to each virtual machine disk partition, and decompress and accomplish the dummy machine system installation, from video imager download VME operating system compression reflection through network real-time through the configuration of shell script complete operation system;

When host operating system is accomplished, when only needing dummy machine system to install, expand or reconfigure, in main frame, use Kickstart to accomplish re-mounting of designated virtual machine.

3. aforesaid right requires 1 described system, possesses the managerial ability of virtual computational resource, it is characterized in that:

The resource pool management technology of layering; According to the virtual machine layering of the logical architecture of business with the robotization supply; Like load balancing layer, global buffer layer, web layer, application layer, database caches layer, database layer etc., each level generates one by the resources of virtual machine pond of data object structure with the tabulation expression.

4. aforesaid right requires 2 described abilities, it is characterized in that:

Defined 3 kinds of objects, Aggregate, Vol, vTask:

The polymerization of all similar virtual machines of Aggregate object representation; The virtual machine polymerization of the corresponding level of the instance of each Aggregate object; Each virtual machine in polymerization all has identical attribute; Bottom is a physical machine, can realize automatic expansion through the new physical machine of robotization supply;

Vol is to liking the sub-set object on Aggregate; The virtual machine set of representing a certain business on this Aggregate example hierarchy, to be assigned with; Vol is the container of a kind of loading VM in logic, can be through increasing or reduce the size that VM changes container arbitrarily; The size of Vol is decided by resource requirements such as how many data volume transmission of application need, how many data storage, rather than is decided by that how many virtual machines are arranged on the physical machine; VM in the Vol is distributed on the different physical machine, can with assignment of traffic to different machines, when a machine breaks down, can not influence whole service; Vol can automatic online increase and minimizing, as long as Aggr has enough spaces; Vol produces for certain specifically is applied on certain Aggr instance;

The vTask object is represented the set of a business required vol object on each Aggregate level, has comprised the service logic flow process;

The corresponding application of vTask is made up of the vol that is distributed in the different Aggr; VTask defines the logic goes of the vol in the different Aggr.

5. aforesaid right requires 1 described system, it is characterized in that:

The smart allocation ability that possesses virtual computational resource; Receive the services request of application deployment, virtual machine application and dynamic expansion subsystem, generate the instance of three objects of corresponding level, each level has the resource allocation policy of oneself; Application layer finds next assignable resources of virtual machine according to decentralization principle in resource pool; Guarantee this virtual machine on different physical machine, partition virtual machines and is deleted this virtual machine in the VM of Vol variable tabulation from resource pool.

6. aforesaid right requires 1 described system, it is characterized in that:

The dynamic expansion ability that possesses virtual computational resource, the dynamic capacity-expanding trigger mechanism:

Through the load of real-time monitoring application program and virtual machine, input judges whether dilatation and capacity reducing through the two-dimensional matrix of artificial neural network training; When degree of declaring decision dilatation or the capacity reducing of system through artificial neural network; System increases or reduces VM to the resource allocation system application, if newly-increased VM, the VM configuration that system copies keeps in advance copies among the new VM; System finds all VM of the vol among the Aggr (i-1); Watch service or the application of each VM, if standards service such as LVS, Apache, PHP, Java, in the service configuration table, find out the configuration file of this service; The IP of the new VM name that increases, the order of appointment makes service read the configuration file of revising again in according to the service correspondence table.

7. aforesaid right requires 1 described system, it is characterized in that, possesses the Virtual Machine Manager technology:

Virtual machine management system is for programmer's management and distribute exploitation to use virtual machine.The programmer submits demands such as the required service of required virtual machine configuration, quantity, application and development, construction cycle to through the web interface to system, after system receives request, and generator program person's record.The system manager lands management interface, from resource pool, distributes next assignable satisfactory virtual machine, and is divided in programmer's record.Delete the virtual machine that distributes on the system backstage from resource pool, when program development is accomplished or the programmer leaves office, system will regain the virtual machine that distributed in resource pool, and delete program person's record.

Images