CN102546379B - Virtualized resource scheduling method and system - Google Patents

Abstract

The invention discloses a virtualized resource scheduling method and a system. A request from an external management user is received, or a scheduling request is triggered according to a preset strategy; according to the resource requirement information of a virtual machine and a physical server in the current system, a virtualized resource scheduling scheme is calculated and obtained; according to physical server resource flowing information contained in the received resource scheduling scheme, the resource of the relevant physical server is scheduled; according to the volume information contained in the virtualized resource scheduling scheme, the resource channel of the volume is built with the corresponding physical server; and scheduling resource is obtained from a mapping physical server to provide scheduled virtual resource for an external business entity. According to the virtualized resource scheduling method and the system, the resource scheduling efficiency can be improved, and the resource can be optimally scheduled in the whole situation.

Description

A kind of method of virtual resources scheduling and virtual resources dispatching patcher

Technical field

The present invention relates to virtual resources dispatching technique, particularly a kind of virtual resources method of dispatching and virtual resources dispatching patcher.

Background technology

Since Amazon and within 2006, release elastic calculation cloud (EC2, Elastic Compute Cloud) platform and highly successful after, industry has started one and provides shared data center infrastructure based on virtual flexible resource pond, the whirlwind that the brand-new business model (the publicly-owned cloud of IaaS pattern) internally integrating (privately owned cloud) shared resource or externally lease service is studied.

The combination of Intel Virtualization Technology and elastic calculation cloud platform brings brand-new resource consolidation and using forestland, wherein, the distribution according to need of resource and dynamic flow are for improving the utilance of elastic calculation cloud platform resource, the service quality improving elastic calculation cloud service and the total cost of ownership tool that reduces elastic calculation cloud user is of great significance.

In cloud computing resource pool, each physical server by running virtualization software, thus is virtualized into some separate virtual machines, as the unit of Business Entity carrying.Share the hardware resource of this physical server between each virtual machine that this physical server virtualized is corresponding, that is, the hardware resource of this physical server can be shared by each virtual machine of correspondence, namely can dispatch between local resource.In prior art, local resource mainly comprises cpu resource and memory source, carries out brief description below to local resource scheduling.

Local cpu resource regulating method:

Fig. 1 is the existing virtual machine monitor structural representation carrying out cpu resource scheduling.See Fig. 1, virtual machine monitor comprises:

Interception module, for intercepting and capturing the frequency adjustment instruction that multiple client operating system sends, and obtains each self-corresponding expected frequency of all frequencies adjustment instruction;

Acquisition module, for obtaining the load information of each self-corresponding virtual cpu of all expected frequencies according to expected frequency;

Distribution module, for distributing true CPU resource according to the load information of virtual cpu, further, the true CPU resource that the heavier virtual cpu of load is assigned to is more.

Local memory source dispatching method:

Be different from the feature that local cpu resource regulating method focuses on Optimized Operation Strategy Design, virtual platform carries out local memory source scheduling, also be faced with virtual machine internal internal memory service condition obtain not easily with the practical difficulty such as memory requirements prediction, therefore, on scheduling strategy, the hypothesis of identical service priority is had based on each virtual machine, and to minimize this locality with interruption times for optimization aim, by arranging the heuristic search algorithm of carrying out iteration between two between multiple virtual machines that this physical server is corresponding, memory source scheduling is carried out according to heuristic search algorithm result.

That studies along with Intel Virtualization Technology and resource-sharing deepens continuously, cross over physical server border in global scope, to realize dynamically sharing of resource become virtual resources with Real-Time Scheduling and share the trend of dispatching and developing, but from above-mentioned, existing virtual resources dispatching method, substantially the multiple virtual machine internal being confined to physical server corresponding carry out scheduling of resource, virtual resources scheduling model is too simple, do not consider in global resource scheduling scheme, the factors such as performance cost very important during the long-range use of resource and network capacity restriction, lack virtual global resource, optimize the ability of global resource scheduling.

With resource to be that object carries out the visual angle of fine granularity Optimized Operation between Business Entity different from traditional scheduling of resource, virtual machine (vm) migration dispatching method adopts the mode of Business Entity migration to realize the global configuration of all kinds of resource, virtual resources system is according to each physical server resource situation, take virtual machine as thread, virtual machine is dispatched between each physical server, like this, resource can be shared between multiple physical server.Such as, but in this virtual machine (vm) migration dispatching method, take virtual machine as thread, thread granularity is comparatively thick, and, do not participate in scheduling less than the resource of a thread, make dispatching efficiency lower, resource can not get effective Optimized Operation in the overall situation; And, each scheduling of resource, may need to reschedule original scheduled resource, scheduling is comparatively complicated, make single dispatching office relate to resource type complexity, such as, need the integrated decision-making relating to cpu resource, memory source, disk resource etc., and being subject to all many condition restrictions such as physical server resource concrete configuration, its optimized scheduling problem can not Direct Modeling be continuous planning problem; Further, above-mentioned prior art all adopts Static Design thinking, does not consider resource distribution and service deployment scheme in resource pool Construction and operation life cycle as a whole, and the dynamically changeable factor such as business load pressure is for the impact of dispatching effect.

Summary of the invention

In view of this, main purpose of the present invention is the method proposing the scheduling of a kind of virtual resources, improves scheduling of resource efficiency, realizes the Optimized Operation of resource in the overall situation.

Another object of the present invention is to propose a kind of virtual resources dispatching patcher, improve scheduling of resource efficiency, realize the Optimized Operation of resource in the overall situation.

For achieving the above object, the invention provides the method for a kind of virtual resources scheduling, the method comprises:

Receive from external management user request, or, triggering dispatch request according to pre-setting strategy, according to the resource requirement information of virtual machine in current system and physical server, calculating and obtaining virtual resources scheduling scheme;

According to the physical server resource flow information comprised in the virtual resources scheduling scheme received, the resource of scheduling related physical server;

According to the capacity information that virtual resources scheduling scheme comprises, set up the chnnels of resources of this capacity with respective physical server;

Obtain scheduling resource from the physical server mapped, the Business Entity for outside provides the virtual resource of scheduling.

The resource requirement information of physical server comprises Gong amount or the demand information of physical server node, adopts greedy matching algorithm to calculate and obtains virtual resources scheduling scheme.

The greedy matching algorithm of described employing calculates and obtains virtual resources scheduling scheme and specifically comprises:

Gains resources pond physical server node set and resource production and marketing relation information s _i, wherein, s _irepresent Gong amount or the demand of i-th physical server node;

All local resources physical server node that supply exceed demand is rearranged queue O from big to small by output;

All local resources physical server node that supply falls short of demand is rearranged queue I from big to small by demand;

Physics server node i and physical server node j is taken out respectively from queue O and queue I:

If | s _i| > | s _j|, by { i → j:s _jadd virtual resources scheduling scheme, upgrade s _i=s _i+ s _j, i is reinserted queue I;

If | s _i| < | s _j|, by { i → j:s _iadd virtual resources scheduling scheme, upgrade s _j= _si+ s _j, j is reinserted queue O;

Export virtual resources scheduling scheme.

The resource requirement information of described physical server comprises the long-range allocation and transportation cost information of resource further, adopts transport optimizing algorithm calculate and obtain virtual resources scheduling scheme.

Described employing transport optimizing algorithm calculates and obtains virtual resources scheduling scheme and specifically comprises:

The Gains resources physical server node that supply exceed demand can supply resource a _i;

The Gains resources physical server node b that supply falls short of demand _j;

Obtain the resource long-range allocation and transportation cost c of i-th resource physical server node that supply exceed demand and the jth resource physical server node that supply falls short of demand _ij;

Calculate minimum value; In formula,

$\begin{array}{cc}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}=a_i& (i=1,...,m),\end{array}$ $\begin{array}{cc}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}=b_j& (j=1,...,n);\end{array}$

M is the resource physical server nodes that supply exceed demand, and n is the resource physical server nodes that supply falls short of demand.

The resource requirement information of described physical server comprises the long-range allocation and transportation cost information of resource further, adopts shortest path algorithm calculate and obtain virtual resources scheduling scheme.

The resource requirement information of described physical server comprises the long-range allocation and transportation cost information of resource and network system load information further, adopts minimum cost flow algorithm calculate and obtain virtual resources scheduling scheme.

When resource bottleneck is physical server end, by the long-range allocation and transportation cost information of logic pond method Gains resources.

When resource bottleneck is that network connects, by the long-range allocation and transportation cost information of physics pond method Gains resources.

The resource requirement information of described physical server comprises network bandwidth restricted information further.

Described employing minimum cost flow algorithm calculates and obtains virtual resources scheduling scheme and specifically comprises:

Physical server each in resource pool is categorized as the physical server of physical server that resource has more than needed and resource anxiety;

Network between physical server in resource pool is connected bandwidth and limits the flow upper limit being mapped as arc between nodes in real time, obtain and determine capacity limit c _ij;

Using bandwidth consumption for the impact of operation system as network diagram interior joint to the resource cost of transportation b between arc (i, j) _ij;

Obtain minimum value.

When communication performance is limited to physical server, adopt logic pond method Gains resources cost of transportation information.

When communication performance is limited to network topology and interconnect architecture, adopt physics pond method Gains resources cost of transportation information.

By the augmenting chain of unit of adjustment's flow least cost is obtained described minimum value as the augmenting chain that expense is minimum.

A kind of virtual resources dispatching patcher, this system comprises: distributed business network DSN multi-service resource manager, scheduling of resource decision-making device, distributed virtual machine Resource Scheduler, multiple virtual machine and multiple physical server, wherein,

DSN multi-service resource manager, for providing business interface according to type of service Business Entity externally, the resource mapping of the virtual machine that maintenance service type is corresponding and physical server, receive and ask from external management user, or, triggering dispatch request according to pre-setting strategy, exporting scheduling of resource decision-making device to;

Scheduling of resource decision-making device, for providing user interface, receiving scheduling request, according to the resource requirement information of virtual machine in current system and physical server, calculates and obtains virtual resources scheduling scheme, being sent to distributed virtual machine Resource Scheduler;

Distributed virtual machine Resource Scheduler, for according to the physical server resource flow information comprised in the virtual resources scheduling scheme received, dispatches the resource of related physical server;

Virtual machine, for obtaining scheduling resource from the physical server mapped, the Business Entity being outside by DSN multi-service resource manager provides the virtual resource of scheduling;

Physical server, for being virtualized into multiple separate virtual machine by running virtualization software, according to the capacity information that virtual resources scheduling scheme comprises, sets up the chnnels of resources of this capacity with respective physical server.

Described DSN multi-service resource manager is further used for obtaining Business Entity load information, implements user and asks shunting and adjustment.

Described virtual resources scheduling scheme comprises: do not shut down virtual machine (vm) migration, local resource flowing and strange land resource flow.

Described scheduling of resource decision-making device comprises: mapping block and computing module, wherein,

Mapping block, for providing user interface, actual schedule scene map is fundamental plan model by receiving scheduling request, according to the resource requirement information of virtual machine in current system and physical server, determine the network description parameter sets of optimizing scheduling target and correspondence;

Computing module, for the network description parameter sets determined according to mapping block, calculates and the virtual resources scheduling scheme of Gains resources flowing, is sent to distributed virtual machine Resource Scheduler.

Described distributed virtual machine Resource Scheduler comprises: monitoring modular and enforcement module, wherein,

Monitoring modular, for extracting the parameter of fundamental plan model need and the state parameter for determining mapping block employing fundamental plan problem model from real network environment;

Implement module, for according to the physical server resource flow information comprised in the virtual resources scheduling scheme received, dispatch the resource of related physical server.

Described monitoring modular comprises parameter acquiring submodule and status monitoring submodule, wherein,

Parameter acquiring submodule, for extracting limiting parameter needed for plan model and cost parameter from real network environment;

Status monitoring submodule, extracts for determining that mapping block adopts the state parameter of fundamental plan problem model from real network environment for completing.

As seen from the above technical solutions, the method of a kind of virtual resources scheduling provided by the invention and virtual resources dispatching patcher, receive and ask from external management user, or, dispatch request is triggered according to pre-setting strategy, according to the resource requirement information of virtual machine in current system and physical server, calculate and obtain virtual resources scheduling scheme; According to the physical server resource flow information comprised in the virtual resources scheduling scheme received, the resource of scheduling related physical server; According to the capacity information that virtual resources scheduling scheme comprises, set up the chnnels of resources of this capacity with respective physical server; Obtain scheduling resource from the physical server mapped, the Business Entity for outside provides the virtual resource of scheduling.Like this, make full use of cross-platform fine granularity Resources Sharing Mechanism, improve resource utilization and scheduling of resource efficiency, achieve the Optimized Operation of resource in the overall situation.

Accompanying drawing explanation

Fig. 1 is the existing virtual machine monitor structural representation carrying out cpu resource scheduling.

Fig. 2 is embodiment of the present invention virtual resources dispatching patcher structural representation.

Fig. 3 is the method first embodiment schematic flow sheet of virtual resources of the present invention scheduling.

Fig. 4 is the method second embodiment schematic flow sheet of virtual resources of the present invention scheduling.

Fig. 5 is the schematic flow sheet that embodiment of the present invention resource global registration problem adopts greedy matching algorithm calculation optimization scheduling scheme.

Fig. 6 is that the embodiment of the present invention solves minimum cost flow rudimentary algorithm min_flow (G) schematic flow sheet.

Fig. 7 is global resource optimized scheduling case study in embodiment of the present invention resource pool life cycle and solving model structural representation.

Embodiment

For making the object, technical solutions and advantages of the present invention clearly, the present invention is described in further detail below in conjunction with the accompanying drawings and the specific embodiments.

The object of the embodiment of the present invention is to provide and suits distributed business network (DSN, Distributed Service Network) the global resource scheduling strategy of privately owned cloud platform resource pond actual environment, make full use of cross-platform fine granularity Resources Sharing Mechanism, while raising resource utilization and scheduling of resource efficiency, by the condition analysis for remote resource access restriction and loss, utilize optimization programming theoretical, balance scheduling strategy performance boost (resource utilization effect of optimization) and cost overhead, and the dynamical feedback combined for resource pool resource distribution and service condition, thering is provided can the self adaptation of Stepwise refinement and the adjustable resource regulating method of complexity according to actual needs.Its core concept is:

In schedule virtual resources system, first calculate according to the resource requirement information of virtual machine in current system and physical server and obtain virtual resources scheduling scheme;

Then, consider resource pool network environment and specifically limit with under the prerequisite of remote resource access cost, utilize optimization programming thinking to solve global resource optimized scheduling problem.Resource pool network environment specifically limits and comprises: the real-time network bandwidth restriction etc. between physical server; The Business Entity resource consumption needs cost that remote resource access cost flows for reflecting global resource, such as, long-distance inner resource access, disk resource access etc.;

Further, the resource distribution corresponding for the different phase of resource pool Construction and operation and service deployment demand, in conjunction with composite factors such as resource load pressure real-time change, for the state that related service entity setting up is different, scheduling problem that dynamic conditioning adopts analyzes granularity and stipulations derivation algorithm, thus dispatching effect and response time, assess the cost between obtain balance;

Finally, analyzing the difference of resource bottleneck present position for current system network state, is the state that related service entity setting up is different, the limiting parameter acquisition methods that dynamic conditioning adopts, thus obtains balance between modeling effect and intermediate computations cost.

Fig. 2 is embodiment of the present invention virtual resources dispatching patcher structural representation.See Fig. 2, this system comprises: DSN multi-service resource manager, scheduling of resource decision-making device, distributed virtual machine Resource Scheduler, multiple virtual machine and multiple physical server, wherein,

DSN multi-service resource manager, for providing business interface according to type of service Business Entity externally, the resource mapping of the virtual machine that maintenance service type is corresponding and physical server, receive and ask from external management user, or, triggering dispatch request according to pre-setting strategy, exporting scheduling of resource decision-making device to;

In the embodiment of the present invention, in order to improve resource-sharing efficiency, reduce the complexity of scheduling of resource, DSN multi-service resource manager is classified according to the Business Entity of type of service to outside, for each class Business Entity safeguards corresponding resources of virtual machine, the resource mapping of managing virtual machines and physical server.

Outside Business Entity is positioned at Core Feature layer, and for providing service application scene, Core Feature layer upwards provides calling interface to each type telecommunications application software.Such as, with the networking telephone (VoIP, Voice over Internet Protocol) the voice class calling service interface that is representative, the content share class calling service interface etc. that is representative with Streaming Media (Streaming).

Core Feature layer inner utilization point-to-point transmission (P2P, Peer to Peer) distributed technology, the a large number of users Requests routing of self terminal is to each peer node comprised in superposition (Overlay) network of distributed sound exchange and distributed content exchange in the future, each peer node uses the virtual resource provided by infrastructure layer, and namely virtual machine provides service.

Type of service comprises: the distributed reciprocal exchange of business of DSN and content exchange business.

DSN multi-service resource manager is positioned at the infrastructure layer of below Core Feature layer; infrastructure layer provides the abstract network abilities such as calculating, storage, scheduling to Core Feature layer; system-level Intel Virtualization Technology is utilized to realize the flexible division to physical server resource in virtual machine mode; and fine-grained resource dynamic scheduling decision and enforcement between different business, different virtual machine, different physical server can be realized by technology such as no shutdown virtual machine (vm) migration, resource flows, the virtual resources dispatching patcher of the composition embodiment of the present invention.

DSN multi-service resource manager, also for obtaining Business Entity load information, being implemented user and being asked shunting and adjustment.That is, dynamically adding/exiting of peer node in Core Feature layer be in charge of by DSN multi-service resource manager; Simultaneously according to the resource dispatching strategy received, according to the real time load of each peer node, implement user and ask shunting and adjustment, realize load balancing and the disaster tolerance mechanism of application layer.

Scheduling of resource decision-making device, for providing user interface, receiving scheduling request, according to the resource requirement information of virtual machine in current system and physical server, calculates and obtains virtual resources scheduling scheme, being sent to distributed virtual machine Resource Scheduler;

In the embodiment of the present invention, dispatch request is from system manager or Core Feature layer, and virtual resources scheduling scheme is the mapping adjustable strategies of physical resource to resources of virtual machine.

Scheduling of resource decision-making device is by controlled trigger mechanism and execution mechanism, be in each physical server of disposing of core net in operator or implement virtual resources scheduling scheme between physical server, concrete virtual resources scheduling scheme comprises: local resource flowing and strange land resource flow etc.Strange land resource flow is namely under the prerequisite keeping existing system virtual machine and physical server mapping relations, the resource of physical server free time is dispatched between each virtual machine, such as, the resource of physical server A free time can be dispatched between the virtual machine that physical server A and physical server B is corresponding.

Virtual resources scheduling scheme in the embodiment of the present invention can be used as the part of infrastructure layer in DSN fusion architecture (virtual resources dispatching patcher), for completing the global resource dynamic dispatching that the P2P such as DSN VoIP, DSN Streaming business is carried out across physical server on the unified resource pond that DSN fusion architecture provides.

Scheduling of resource decision-making device comprises: mapping block and computing module, wherein,

Mapping block, for providing user interface, actual schedule scene map is fundamental plan model by receiving scheduling request, according to the resource requirement information of virtual machine in current system and physical server, determine the network description parameter sets of optimizing scheduling target and correspondence;

Computing module, for the network description parameter sets determined according to mapping block, calculates and the virtual resources scheduling scheme of Gains resources flowing, is sent to distributed virtual machine Resource Scheduler.

Distributed virtual machine Resource Scheduler, for according to the physical server resource flow information comprised in the virtual resources scheduling scheme received, dispatches the resource of related physical server;

In the embodiment of the present invention, according to virtual resources scheduling scheme schedules resource, specifically see relate art literature, can not repeat them here.

Distributed virtual machine Resource Scheduler comprises: monitoring modular and enforcement module, wherein,

Monitoring modular, for extracting the parameter of fundamental plan model need and the state parameter for determining mapping block employing fundamental plan problem model from real network environment, comprises parameter acquiring submodule and status monitoring submodule,

Parameter acquiring submodule, for extracting limiting parameter needed for plan model and cost parameter from real network environment;

Status monitoring submodule, extracts for determining that mapping block adopts the state parameter of fundamental plan problem model, i.e. current available resource distribution situation from real network environment for completing.

Implement module, for according to the physical server resource flow information comprised in the virtual resources scheduling scheme received, dispatch the resource of related physical server.

Virtual machine, for obtaining scheduling resource from the physical server mapped, the Business Entity being outside by DSN multi-service resource manager provides the virtual resource of scheduling;

Physical server, is positioned at physical layer, for being virtualized into multiple separate virtual machine by running virtualization software, according to the capacity information that virtual resources scheduling scheme comprises, sets up the chnnels of resources of this capacity with respective physical server.

In the embodiment of the present invention, each physical server runs multiple virtual machine, in each virtual machine, run logic function softwares such as being used for operation system request process.

In the embodiment of the present invention, the dispatch request triggering virtual resources scheduling can be that system manager manually initiates, and also can be automatically trigger according to pre-setting strategy.

Fig. 3 is the method first embodiment schematic flow sheet of virtual resources of the present invention scheduling.See Fig. 3, this flow process comprises:

Step 301, keeper according to actual needs, determines artificial initiation global resource allocation optimized dispatch request, initiates global resource dynamic regulation request (GloSch_req ()) to DSN multi-service resource manager;

Step 302, DSN multi-service resource manager receives global resource dynamic regulation request (GloSch_cal_req (P)), and request resource scheduling decision device calculates global resource dynamic scheduling scheme (global adaptation scheme D) according to Current resource pond available resources distribution situation;

In this step, available resources distribution situation refers to the available resources on physical server, can upwards be reported to DSN multi-service resource manager by distributed virtual machine Resource Scheduler.Certainly, in practical application, can also the resource requirement information of combined with virtual machine and physical server, or other relevant information calculates, about calculating global resource dynamic scheduling scheme, to be follow-uply described in detail again.

Step 303, scheduling of resource decision-making device calls the Optimization scheduling algorithm pre-set, and calculates global resource dynamic scheduling scheme, and is sent to distributed virtual machine Resource Scheduler;

In this step, about Optimization scheduling algorithm, to be follow-uply described in detail, global resource dynamic scheduling scheme is sent to distributed virtual machine Resource Scheduler by GloFlow_req () message by scheduling of resource decision-making device again.

Step 304, distributed virtual machine Resource Scheduler, according to the global resource dynamic scheduling scheme received, initiates corresponding physical server resource flow instruction (VM_flow (u, v, vol)) to physical server;

In this step, distributed virtual machine Resource Scheduler global resource dynamic scheduling scheme is split into for specific physical server between physical resource flowing instruction set, send to corresponding physical server to execution.In resource flow instruction, u, v are physical server mark, and vol is the capacity setting up resource flow between physical server u and physical server v.

Step 305, physical server receives resource flow instruction (VM_flow (u, v, vol)), according to the information that instruction is carried, capacity of setting up between physical server u and physical server v is the resource flow channel of vol, and to the information of distributed virtual machine Resource Scheduler feedback operation success or not (Success/Failure);

In this step, by (VM_flow_confirm ()) message to distributed virtual machine Resource Scheduler feedback prompts information.

Step 306, distributed virtual machine Resource Scheduler receives information, amendment available resources statistics, and returns adjustment object information to DSN multi-service resource manager;

In this step, adjustment object information is the information of reception.

Step 306, DSN multi-service resource manager receives adjustment object information, amendment available resources statistics, and returns adjustment object information to keeper.

Fig. 4 is the method second embodiment schematic flow sheet of virtual resources of the present invention scheduling.See Fig. 4, this flow process comprises:

Step 401, DSN multi-service resource manager is monitoring resource pool each physical server available resources situation P periodically, computational resource allocation-business demand matching degree;

In this step, periodically determined whether automatically to trigger global resource dynamic scheduling request to its physical server Resourse Distribute monitored and service condition matching degree by DSN multi-service resource manager.The Resourse Distribute calculated-business demand matching degree is dispatching effect, considers by resource utilization and Operationbenefit two aspect.Concrete comprehensive calculation method is not limit, and manually can be determined, specifically see relate art literature, can not repeat them here by keeper according to resource pool concrete condition.

Step 402, judges that Resourse Distribute-business demand matching degree is less than the threshold value pre-set, and triggers global resource allocation optimized dispatch request (GloSch_cal_req (P));

Step 403, scheduling of resource decision-making device calls the optimization Optimization scheduling algorithm pre-set, and calculates global resource dynamic scheduling scheme, and is sent to distributed virtual machine Resource Scheduler;

In this step, global resource dynamic scheduling scheme is sent to distributed virtual machine Resource Scheduler by GloFlow_req () message by scheduling of resource decision-making device.

Step 404, distributed virtual machine Resource Scheduler, according to the global resource dynamic scheduling scheme received, initiates corresponding physical server resource flow instruction (VM_flow (u, v, vol)) to physical server;

Step 405, physical server receives resource flow instruction (VM_flow (u, v, vol)), according to the information that instruction is carried, capacity of setting up between physical server u and physical server v is the resource flow channel of vol, and to the information of distributed virtual machine Resource Scheduler feedback operation success or not (Success/Failure);

In this step, by (VM_flow_confirm ()) message to distributed virtual machine Resource Scheduler feedback prompts information.

Step 406, distributed virtual machine Resource Scheduler receives information, amendment available resources statistics, and returns adjustment object information to DSN multi-service resource manager.

After according to the resource requirement information determination virtual resources scheduling scheme of virtual machine and physical server in Current resource pond available resources distribution situation or current system, if multiple physical server can both provide resource requirement, as previously mentioned, the embodiment of the present invention is different due to heterogeneous networks scale and resource bottleneck present position in the resource pool under network environment, such as, for the situation that Business Entity resources requirement is larger, its resource bottleneck present position may be physical server side, the different stipulations modes of set planning problem may be caused, thus obtain the different special cases of this problem, and, even if network environment is given, along with the difference of bearer service type and the dynamic change of real time load, resource bottleneck present position also may be moved.Therefore, in the embodiment of the present invention, provide a kind of for abstract problem modeling method enough accurate under specific environment further, thus conciliate realization balance between annual reporting law complexity in expectation dispatching effect.Expect dispatching effect can carry out scheduling of resource according to result of calculation after resource-demand matching degree weigh, consider by resource utilization and Operationbenefit two aspect, concrete comprehensive calculation method is not limit, and manually can be determined by keeper according to resource pool concrete condition.

For this reason, the mapping block of the embodiment of the present invention and the core concept of computing module when completing particular problem stipulations and solving are in the dispatch application of real resource pond, first select and can to accept with actual match degree and the minimum planning solution annual reporting law of complexity, make every effort to shorten scheduling scheme computing time, the tolerable bottom line pre-set is exceeded once dispatching effect, dispatching effect asks reject rate to characterize by physical resource service efficiency and user, under the prerequisite namely do not changed in mode input parameter, after repeatedly carrying out overall scheduling operation, dispatching effect still lower than pre-set threshold value more progressively refinement model is portrayed for the scheduling problem of resource pool environment, or, by keeper when resource pool environment occurs significantly to change, such as, new business online implementing, number of servers changes, network environment adjustment etc., manual adjustment problem portrays model.Thus carry out Optimized Operation effect by the programming evaluation algorithm that the complexity corresponding to problem model after being switched to refinement is higher.

Such as, can suppose that network interconnection resource is infinitely sufficient in small-scale, single service resources pond, capacity (bandwidth c, import-restriction) and cost (b, optimization aim) be 0, now, on the basis of resource requirement information considering virtual machine and physical server in current system, carry out resource global registration, then namely adopt greedy matching algorithm calculation optimization scheduling scheme.

Fig. 5 is the schematic flow sheet that embodiment of the present invention resource global registration problem adopts greedy matching algorithm calculation optimization scheduling scheme.See Fig. 5, this flow process comprises:

Step 501, input resource pool physical server node set and resource production and marketing relation information;

In this step, using each physical server as network diagram node set, resource production and marketing relation information s _irepresent Gong amount or the demand of i-th physical server node.

Step 502, rearranges queue O by all local resources output node that supply exceed demand from big to small by output;

In this step, the resource physical server node that supply exceed demand is called output node.

Step 503, rearranges queue I by all local resources consumption node that supply falls short of demand from big to small by demand;

Step 504, circulation performs step 502 and step 503, until consumption node set is empty;

Step 505, from queue O and queue I, take out node i and node j respectively:

If | s _i| > | s _j|, by { i → j:s _jadd resource overall scheduling scheme D, upgrade s _i=s _i+ s _j, i is reinserted queue I;

In this step, if the available resources of node i meet the remote access demand of node j, then by the decision of " providing j requirement resource to j by i " write resource overall scheduling scheme D (j being shifted out demand queue), again upgrade the available resources quantity of i, still insert supply queue (being taken out queue above).

If | s _i| < | s _j|, by { i → j:s _iadd resource overall scheduling scheme D, upgrade s _j=s _i+ s _j, j is reinserted queue O.

In this step, if node i available resources are not enough to the remote access demand of support node j, then by the decision of " providing i can provide quantity resource to j by i " write resource overall scheduling scheme D (i having been shifted out supply queue), what again upgrade j needs resource quantity, still inserts demand queue (being taken out queue above).

Step 506, exports resource overall scheduling scheme D.

Along with the expansion of networking scale, in extensive, single service resources pond of network environment complexity, further, the cost optimization of the long-range allocation and transportation of resource can be considered, divide work two kinds of situations to consider below.

The first situation: network system underloading

When network system underloading, the capacity limit of the long-range allocation and transportation of resource can not be considered, and select logic pond or physics pond method according to optimizing scheduling effect, logic pond method does not consider the impact of switch on grid, physics pond method considers that switch is on the impact of grid, only carries out monitoring and estimating to allocation and transportation cost.

Like this, can adopt general allocation and transportation cost acquisition methods, now, the further refinement of resource global registration problem, refinement obtains transportation problem model.

Below the Mathematical Modeling of transportation problem and derivation algorithm are described.

Suppose to have certain goods and materials to need allocation and transportation, the dosage unit of this goods and materials can be weight, packing unit or other.There will be a known that m place (physical server) can be used for should this goods and materials, the i.e. place of production, with i=1, and 2 ..., m represents, has n place to need this kind of goods and materials, with namely selling, with j=1, and 2 ..., m represents, if Gong the amount in m the place of production, namely output is a _i(i=1,2 ..., m), the requirement on n pin ground, namely sales volume is b _j(j=1,2 ..., n), from i-th place of production to jth, the unit goods and materials freight rate on a pin ground is c _ij.If use x _ijrepresentative is allocated and transported to the Board Lot of the goods and materials on a jth pin ground from i-th place of production, then, under the condition of the co-ordination of supply and marketing, make total freight charges expenditure minimum, can be expressed as following transportation problem Mathematical Modeling:

$\min \underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{c}_{\mathrm{ij}}{x}_{\mathrm{ij}}$

$s.t.\left\{\begin{array}{cc}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}=a_i& (i=1,...,m)\\ \underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}=b_j& (j=1,...n)\\ & x_{\mathrm{ij}}\&GreaterEqual;0\end{array}\right.$

The classic algorithm of transportation problem optimal solution is table dispatching method, and O (m+n) greedy algorithm of approximate optimal solution comprises MinimumElement Method and Vogel method etc.Specific algorithm see relate art literature, can not repeat them here.

In addition, in practical application, can also according to actual conditions comprehensive selection other planning problem models more simple.Such as, between simple use physical server, local area network (LAN) connection jumping figure simplifies logic pond method, just obtains another special case " critical path problem ".

Below the Mathematical Modeling of critical path problem and derivation algorithm are described.

The most frequently used algorithm of shortest path of asking has two kinds: one to be ask certain a bit to the dijkstra's algorithm of beeline between other points; Another kind is the matrix algorithm asking beeline between any two points in network diagram, wherein,

Dijkstra's algorithm:

If with d _ijrepresent the distance of two consecutive points i and j in network diagram, suppose that i and j is non-conterminous, make d _ij=∞, obviously, d _ii=0, if Ls _irepresent the beeline from s point to i point, then obtain the shortest path from s point to certain 1 t, namely ask specified point to (s, t) the Dijstra algorithm min_distance_pair (G of shortest path between, s, t), as follows by step during dijkstra's algorithm:

A, from a s, because of L _ss=0, this value is labeled in the other little square frame of s, represents s point label;

B, from s point, find out in the point adjacent with s, apart from minimum one, be set to r, by L _sr=L _ss+ d _srvalue be labeled in the other little square frame of r, show a r also label;

C, point from label, find out and put adjacent all non-label point p with these, if having, and L _sp=min _p{ min _r{ min{L _ss+ d _sp; L _sr+ d _rp, then to p piont mark, and by L _spvalue be labeled in the other little square frame of p point;

D, repetition step C, till t point obtains label.

The matrix algorithm of beeline between any two points

If with d _ijrepresent the distance of two consecutive points i and j in network diagram, suppose that i and j is non-conterminous, make d _ij=∞, obviously, d _ii=0, the matrix D (0) obtained thus shows the direct beeline from i point to j point, but from i point to the shortest path not necessarily i → j of j point, and when solving shortest path between any two points with matrix algorithm, step is as follows:

A1, from D (0);

In this step, element d _ij(0) be direct beeline from i point to j point, if k=1.

B1, utilize D (k-1) to construct new matrix D (k), provide the beeline that in network, any two points is counted through middle node when being no more than (2k-1).

In this step, method is: make d _ij(k+1)=min{d _ir(k-1)+d _rj(k-1) }, wherein, k=k+1.

C1, repetition step B1, until there is D (m+1)=D (m) during k=m, each element value of matrix D (m) is designated as each shortest distance between points.

In this step, suppose in network, there be n point, then generally calculate iterations m and meet following formula:

$m-1\&le;\frac{\lg (n-1)}{\lg 2}\&le;m$

The second situation: network system load weight

When system load acquires a certain degree, then programming evaluation is carried out in the monitoring acquisition of enabling for capacity parameter.Now, by above-mentioned " transportation problem " further refinement, refinement obtains " minimum cost flow problem ", and the Mathematical Modeling of minimum cost flow problem and derivation algorithm illustrate and to be follow-uply described again, according to resource bottleneck present position difference, are divided into again two kinds of situations:

If one resource bottleneck is physical server end, then can consider to use the method increase of logic pond for acquisition and the use of allocating and transporting cost parameter;

If two, resource bottleneck is that network connects, the modeling analysis using the method increase of physics pond for allocation and transportation cost parameter can be considered, and use the acquisition analysis of physics pond method for capacity parameter instead.

Like this, scheduling of resource decision-making device under DSN Resource Fusion platform architecture is according to the resource usage monitoring situation in resource pool, real-time adjustment for the problem modeling of optimization scheduling of resource, and adjusts used physical planning derivation algorithm in conjunction with optimizing scheduling result and expectation matching degree.

In practical application, when class of business increase, network-type dense traffic formed on a large scale, multi-service resource pond time, resource pool network environment can be considered further specifically limit and remote resource access cost two aspect factor, can consider to adopt " minimum cost flow problem ".For this reason, need first to complete the mapping from actual schedule scene (SNA) to fundamental plan model (scheduling scheme model), be described respectively below.

One, the mapping of fundamental plan model:

In the embodiment of the present invention, when can be considered factors such as " network bandwidth restrictions " and " operation system resource consumption needs cost " by following concrete steps, in virtual resources pond, between each physical server, carry out the problem of resources optimization scheduling, solved by " minimum cost flow " planning problem each factor is mapped as under network graphics drawing.That is, be fundamental plan model by the actual schedule scene map comprising each influencing factor, this fundamental plan model is minimum cost flow plan model.

First, physical server each in resource pool is categorized as to the physical server of physical server that resource has more than needed and resource anxiety: by abstract for the Optimal Scheduling Problem of global resource at some resource output points, namely the physical server that resource is more than needed and some resources consumptions point, i.e. resource Transport between the physical server of resource anxiety;

In this step, when overall scheduling mechanism is triggered, DSN multi-service resource manager obtains the service condition statistical information to Resources allocation of current available resource reserves and each local service virtual machine on each physical server, calculate, obtain the resources requirement s corresponding to the node i of physical server in network diagram _i.

Secondly, the network between the physical server in cloud computing resource pool is connected bandwidth and limits the flow upper limit that abstract (mapping) is arc between nodes in real time, obtain and determine capacity limit c _ij;

In this step, concrete defining method may be including but not limited to:

Under communication performance is limited to endpoint service device scene, do not consider the logic pond method of group-network construction; With,

Under communication performance is limited to network topology and interconnect architecture scene, consider the physics pond method of group-network construction.

Specifically described below.

One, bandwidth is restricted to the flowmeter factor conversion method of resource restriction

In the embodiment of the present invention, product resource in minimum cost flow problem is the similar resource adopting unified metric standard with input variables such as the flow upper limits on arc, and in concrete resource pool overall scheduling problem, product is the hardware resource shared across physical server, such as, internal memory etc., and with arc flowmeter factor reflection for network connecting band tolerance make, not resource of the same race, linear module is also inconsistent, such as, hardware resource linear module is M, arc stream metric unit is M/s, for solving this problem, can adopt and with the following method bandwidth be limited the restriction converted to teleengineering support resource:

(1) network equipment is utilized to obtain any two node i, the real-time network bandwidth restriction d between j _ij.Under real resource scheduling scenario, if any hardware condition, can directly use the end-to-end bandwidth Real-Time Monitoring numerical value of two ends physical server as d _ijinput.

(2) for the teleengineering support resource of considered unit-sized, data access in the unit interval between its consumer to supplier is determined and the bandwidth m that consumes.Such as, for teleengineering support internal memory, given operation system type and shared drive size, can estimate that within a period of time, resource user carries out long-distance inner access for resource provider and the date transfer on network consumed, thus estimate m.The method of estimation comprises:

The remote access frequency is estimated: obtain the local internal storage access frequency of service logic node in given operation system and average each visit data amount/local memory size by the local internal storage access frequency, be multiplied by long-distance inner size again with these two numerical value products to obtain long-distance inner and access the clean traffic, again divided by cost on network communication ratio (remote access protocol and IP network data packet head expense), thus obtain estimating numerical value; Or,

Under the different long-distance inner allocated size of direct measurement, the statistic of the network access data traffic, after carry out curve fitting, obtain the quantitative relation of the two, then calculate according to it.

(3) any two nodes, such as, physical server node i, flow restriction c between j _ijbe defined as:

c _ij＝d _ij/m

Two, the logic pond method of group-network construction is not considered

Under this situation, the unlimited abundance for current needs of the network interconnection bandwidth in resource pool between each physical server.Now, the bandwidth restriction across physical server shared resource is determined according to the smaller value of self bandwidth more than needed of two endpoint service devices.

Three, the physics pond method of group-network construction is considered

Under this situation, the concrete group-network construction in comprehensive resources pond, the bandwidth restriction introduced for internal switch is considered.Now, the bandwidth restriction across physical server shared resource is determined according to the minimum value of the bandwidth current more than needed of each related exchange equipment providing local exchange to serve between two end points physical servers, network cable.Such as, current mainstream switches are furnished with traffic monitoring function, the bandwidth more than needed in real time of each port of energy Real-Time Monitoring, gets the Monitoring Data minimum value between end points physical server on network path.

Under actual network environment, the real-time statistics of traffic monitoring mechanism realization to inter-node link allocated bandwidth d and utilization ratio v of switching equipment self can be utilized, thus, can directly use d (1-v) numerical value as d _ijinput.

Determine allocation and transportation cost b _ij

3rd step, the cost of consumed resource that global resource scheduling mechanism is introduced such as, bandwidth consumption for the impact of operation system as network diagram interior joint to the resource cost of transportation b between arc (i, j) _ij.

Corresponding aforesaid flow restriction modeling method, determine that the concrete methods of realizing allocating and transporting cost includes but not limited to following two classes equally:

Under communication performance is limited to endpoint service device scene, do not consider the logic pond method of group-network construction; With,

Under communication performance is limited to network topology and interconnect architecture scene, consider the physics pond method of group-network construction.

Specifically described below.

One, the logic pond method of group-network construction is not considered

If the unlimited abundance of network interconnection bandwidth relative requirements in resource pool between each physical server, now, across the bandwidth consumption cost of physical server shared resource from originally can for two of an operation system local backup end points physical server self bandwidth more than needed.Concrete grammar is as follows:

(1) efficiency gains of unit bandwidth and the lost revenue of local service system minimizing unit bandwidth is increased according to physical server node local service system, the long-range allocation and transportation cost of its local unit bandwidth resource of comprehensive estimate.The marginal cost of the long-range allocation and transportation of operator comprehensive resources used, its actual connotation can define as required, and such as, numerical value is taken advantage of, get higher value, get smaller value etc.

(2) according to the long-range allocation and transportation cost of the local unit bandwidth resource of two physical server nodes, comprehensive estimate unit resource allocation and transportation therebetween cost.Comprehensive two the node local coots of operator used, its actual connotation can define as required, and such as, numerical value is taken advantage of, get higher value, get smaller value etc.

Two, the physics pond method of group-network construction is considered

Under this situation, the network interconnection limited bandwidth in resource pool between each physical server.Now, across the bandwidth consumption cost of physical server shared resource also from the bandwidth more than needed of corresponding switching equipment and network cable.Specifically, its modeling method is as follows:

(1) efficiency gains of unit bandwidth and the lost revenue of local service system minimizing unit bandwidth is increased according to physical server node local service system, the long-range allocation and transportation cost of its local unit bandwidth resource of comprehensive estimate.The marginal cost of the long-range allocation and transportation of operator comprehensive resources used, its actual connotation can define as required, and such as, numerical value is taken advantage of, get higher value, get smaller value etc.

(2) according to the long-range allocation and transportation cost of the local unit bandwidth resource of two physical server nodes, comprehensive estimate unit resource allocation and transportation therebetween cost.Comprehensive two the node local coots of operator used, its actual connotation can define as required, and such as, numerical value is taken advantage of, get higher value, get smaller value etc.

(3) according to the local unit bandwidth resource allocation and transportation cost of corresponding physics server node, the unit resource allocation and transportation cost between telephone net node and adjacent server nodes is determined.

(4) unit resource between basis and direct neighbor physical server node allocates and transports cost sum, comprehensively determines the unit resource allocation and transportation cost between two neighboring switch nodes.Wherein comprehensive two the node local coots of operator used, its actual connotation can customize as required, and such as, numerical value adds, get higher value, get smaller value etc.

Like this, the Optimal Scheduling Problem of global resource is summed up as the minimum cost flow problem in set up network model, can utilize related algorithm, such as, minimum cost flow derivation algorithm formulate cost performance/the resource Version Dispatching of suboptimum.

Below minimum cost flow problem and derivation algorithm thereof are described.

Minimum cost flow problem describes: set network to have n node (n as natural number), f _ijfor the flow on arc (i, j), represent that node i is supplied to the resource quantity of the long-range use of node j; c _ijfor the capacity (real-time bandwidth restriction) of this arc, b _ijfor on arc (i, j) by expense during unit discharge, s _irepresent Gong the amount of i-th node or demand, when i is to put in network diagram, s _i> 0, the physical server node that namely resource is more than needed; When i is the sink in network diagram, s _i< 0, i.e. the physical server node of resource anxiety; When i is intermediate transit point (in network diagram other nodes), s _i=0, namely neither send out point also non-sink, such as, the resources requirement of the LAN Monitoring System equipment such as switch is always zero.

When network needs balance altogether time, by the physical distribution of each point to each sink (or press maximum stream flow from each be maneuvered into each sink), make the problem that expense of always allocating and transporting is minimum, following linear programming model can be summed up as:

$\min \underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{b}_{\mathrm{ij}}{f}_{\mathrm{ij}}$

When asking minimum cost flow, on the one hand, flow can be adjusted by finding augmenting chain, and differentiate whether reach maximum stream flow, but then, minimum in order to ensure the expense of the flow flower that every step is whole, need to find out the minimum augmenting chain of each step expense, to ensure that the flow that finally provides or maximum stream flow are also that expense is minimum, augmenting chain is make the flow adjustment path that in network, given point increases to sink resource allocation and transportation quantity.

If b (f) is the expense of feasible flow f, be f ' (f ' > f) along the flow after augmenting chain adjustment, corresponding expense is b (f '), then expense difference Δ b (f):

$\mathrm{\&Delta;b}\left(f\right)=b\left(f^{\&prime;}\right)-b\left(f\right)=\underset{{\mathrm{\&mu;}}^{+}}{\mathrm{\&Sigma;}}{b}_{\mathrm{ij}}({f^{\&prime;}}_{\mathrm{ij}}-f_{\mathrm{ij}})-\underset{{\mathrm{\&mu;}}^{-}}{\mathrm{\&Sigma;}}{b}_{\mathrm{ij}}({f^{\&prime;}}_{\mathrm{ji}}-f_{\mathrm{ji}})=\mathrm{\&theta;}[\underset{{\mathrm{\&mu;}}^{+}}{\mathrm{\&Sigma;}}{b}_{\mathrm{ij}}-\underset{{\mathrm{\&mu;}}^{-}}{\mathrm{\&Sigma;}}{b}_{\mathrm{ji}}]$

Make ratio [Δ b (f)/θ] minimum, also by the augmenting chain of unit of adjustment's flow least cost as the minimum augmenting chain of expense.

In practical application, if when every bar arc is occurred as forward arc or opposite arc, other as flexible strategy mark at this arc by the expense of the unit discharge of this arc, then find the minimum augmenting chain of expense, one can be converted into again and ask a point to the critical path problem of sink.It is therefore, above-mentioned that to ask the step of minimum cost flow to sum up as follows.

Fig. 6 is that the embodiment of the present invention solves minimum cost flow rudimentary algorithm min_flow (G) schematic flow sheet.See Fig. 6, this flow process comprises:

Step 601, from zero stream f ₀start;

In this step, f ₀feasible flow, the feasible flow that when be also corresponding flow being zero, expense is minimum.

Step 602, to feasible flow f _kstructure weighted network W (f _k);

In this step, structure weighted network W (f _k) specifically comprise:

To o < f _ij< c _ijarc (i, j), when it is forward arc, if be b by the expense of unit stream _ij, during for opposite arc, corresponding expense is b _ji=-b _ij.Like this, at weighted network W (f _k) in node i and node j point between, can provide arc (i, j) and (j, i) respectively, the flexible strategy that arc (i, j) is corresponding are b _ij, namely on arc (i, j) by expense during unit discharge, the flexible strategy that arc (j, i) is corresponding are b _ji, namely on arc (j, i) by expense during unit discharge, wherein, i≤k, j≤k.

To 0 < f _ij=c _ijarc (i, j), because this arc stream amount is saturated, can only as opposite arc in augmenting chain, namely at weighted network W (f _k) in, only provide arc (j, i), the flexible strategy that namely arc (j, i) is corresponding are b _ji.

To f _ijthe arc (i, j) of=0 can only be forward arc, namely at weighted network W (f in augmenting chain _k) in, only provide arc (i, j), the flexible strategy that namely arc (i, j) is corresponding are b _ij.

Step 603, at the weighted network W (f of structure _k) in, obtain the augmenting chain that expense is minimum;

In this step, obtaining the augmenting chain that expense is minimum, namely obtaining from sending out the shortest path of point to sink.

Step 604, is adjusted to the maximum of permission by flow on augmenting chain minimum for the expense of acquisition, obtains new flow f _k+1.

In this step, flow on augmenting chain corresponding for the shortest path of acquisition is adjusted to the maximum of permission, obtains a new flow f _k+1, wherein, f _k+1> f _k.

Step 605, returns and performs step 602, until at weighted network W (f _k+m) in there is not augmenting chain, this f _k+mfor required maximal flows at lowest cost.

About other explanations of minimum cost flow rudimentary algorithm, specifically see relate art literature, can not repeat them here.

Fig. 7 is global resource optimized scheduling case study in embodiment of the present invention resource pool life cycle and solving model structural representation.See Fig. 7, by adopting adaptive approach, can be used for instructing the global resource optimized scheduling case study within the whole resource pool Construction and operation cycle and solving, specifically:

When small-scale, single service resources pond, can abstractly be resource matched problem;

When popularization, networking complexity increase extensive, the single service resources pond of formation, can abstractly be transport optimization problem;

And when class of business increase, network-intensive business formed on a large scale, multi-service resource pond time, can abstractly be minimum cost flow problem.

From above-mentioned, in the embodiment of the present invention, in schedule virtual resources system, consider resource pool network environment and specifically limit with under the prerequisite of remote resource access cost, utilize optimization programming thinking to solve global resource optimized scheduling problem;

For the different phase of resource pool Construction and operation, by arranging different states for scheduling entity, in conjunction with resource pool resource distribution and the composite factor such as service deployment, load pressure, the case study granularity that dynamic conditioning adopts and stipulations derivation algorithm, thus dispatching effect and response time, assess the cost between obtain balance;

Present position for resource bottleneck is different, proposes the logic pond parameter acquiring method under the resource-constrained scene of server end and the physics pond parameter acquiring method under Internet resources limitation scene respectively.

There is following technological merit:

One, for feature and the demand of global resource scheduling mechanism in schedule virtual resources system, considers resource pool network environment and specifically limits and remote resource access cost, set up the case study based on " minimum cost flow " problem and solution throughway.

They are two years old, for the different phase of resource pool Construction and operation, the artificial trigger scenario of design management person, implement the self-adaptative adjustment of scheduling problem modeling analysis and derivation algorithm, thus balance is obtained between algorithm complex and dispatching effect: first select and can to accept with actual match degree and the minimum planning solution annual reporting law of complexity makes every effort to shorten scheduling scheme computing time, tolerable bottom line is exceeded once dispatching effect, recycling embodiment of the present invention method progressively refinement portrays model for the scheduling problem of resource pool environment, thus carry out Optimized Operation effect by the programming evaluation algorithm that the complexity corresponding to problem model after being switched to refinement is higher.

Its three, the logic pond parameter acquiring method under server end resource-constrained scene and the physics pond parameter acquiring method under Internet resources limitation scene are proposed.Based on this, for monitoring modular arranges state, different for the present position of resource pool resource bottleneck under reality operation scene, the parameter acquiring method that dynamic conditioning uses, thus obtain Quan Heng at model and between solving complexity and dispatching effect.

The foregoing is only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement and improvement etc., all should be included within protection scope of the present invention.

Claims (20)

1. a method for virtual resources scheduling, it is characterized in that, the method comprises:

Receive from external management user request, or, triggering dispatch request according to pre-setting strategy, according to the resource requirement information of virtual machine in current system and physical server, calculating and obtaining virtual resources scheduling scheme;

According to the physical server resource flow information comprised in the virtual resources scheduling scheme received, the resource of scheduling related physical server;

According to the capacity information that virtual resources scheduling scheme comprises, set up the chnnels of resources of this capacity with respective physical server;

Obtain scheduling resource from the physical server mapped, the Business Entity for outside provides the virtual resource of scheduling.

2. the method for claim 1, is characterized in that, the resource requirement information of physical server comprises Gong amount or the demand information of physical server node, adopts greedy matching algorithm to calculate and obtains virtual resources scheduling scheme.

3. method as claimed in claim 2, it is characterized in that, the greedy matching algorithm of described employing calculates and obtains virtual resources scheduling scheme and specifically comprises:

Gains resources pond physical server node set and resource production and marketing relation information s _i, wherein, s _irepresent Gong amount or the demand of i-th physical server node;

All local resources physical server node that supply exceed demand is rearranged queue O from big to small by output;

All local resources physical server node that supply falls short of demand is rearranged queue I from big to small by demand;

Physics server node i and physical server node j is taken out respectively from queue O and queue I:

If | s _i| > | s _j|, by { i → j:s _jadd virtual resources scheduling scheme, upgrade s _i=s _i+ s _j, i is reinserted queue I;

If | s _i| < | s _j|, by { i → j:s _iadd virtual resources scheduling scheme, upgrade s _j=s _i+ s _j, j is reinserted queue O;

Export virtual resources scheduling scheme.

4. method as claimed in claim 3, it is characterized in that, the resource requirement information of described physical server comprises the long-range allocation and transportation cost information of resource further, adopts transport optimizing algorithm calculate and obtain virtual resources scheduling scheme.

5. method as claimed in claim 4, it is characterized in that, described employing transport optimizing algorithm calculates and obtains virtual resources scheduling scheme and specifically comprises:

The Gains resources physical server node that supply exceed demand can supply resource a _i;

The Gains resources physical server node b that supply falls short of demand _j;

Obtain the resource long-range allocation and transportation cost c of i-th resource physical server node that supply exceed demand and the jth resource physical server node that supply falls short of demand _ij;

Calculate minimum value; In formula,

$\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}=a_i(i=1,...,m),\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{x}_{\mathrm{ij}}=b_j(j=1,...,n);$

M is the resource physical server nodes that supply exceed demand, and n is the resource physical server nodes that supply falls short of demand;

Described x _ijbe that i-th resource physical server node that supply exceed demand is allocated and transported to the Board Lot of the resource of the jth resource physical server node that supply falls short of demand.

6. method as claimed in claim 3, it is characterized in that, the resource requirement information of described physical server comprises the long-range allocation and transportation cost information of resource further, adopts shortest path algorithm calculate and obtain virtual resources scheduling scheme.

7. method as claimed in claim 3, it is characterized in that, the resource requirement information of described physical server comprises the long-range allocation and transportation cost information of resource and network system load information further, adopts minimum cost flow algorithm calculate and obtain virtual resources scheduling scheme.

8. method as claimed in claim 7, is characterized in that, when resource bottleneck is physical server end, by the long-range allocation and transportation cost information of logic pond method Gains resources.

9. method as claimed in claim 7, is characterized in that, when resource bottleneck is that network connects, by the long-range allocation and transportation cost information of physics pond method Gains resources.

10. method as claimed in claim 7, it is characterized in that, the resource requirement information of described physical server comprises network bandwidth restricted information further.

11. methods as described in any one of claim 7 to 10, it is characterized in that, described employing minimum cost flow algorithm calculates and obtains virtual resources scheduling scheme and specifically comprises:

Physical server each in resource pool is categorized as the physical server of physical server that resource has more than needed and resource anxiety;

Network between physical server in resource pool is connected bandwidth and limits the flow upper limit being mapped as arc between nodes in real time, obtain and determine capacity limit c _ij;

Using bandwidth consumption for the impact of operation system as network diagram interior joint to the resource cost of transportation b between arc (i, j) _ij;

Obtain minimum value;

Wherein, described f _ijfor node i is supplied to the resource quantity of the long-range use of node j.

12. methods as claimed in claim 11, is characterized in that, when communication performance is limited to physical server, adopt logic pond method Gains resources cost of transportation information.

13. methods as claimed in claim 11, is characterized in that, when communication performance is limited to network topology and interconnect architecture, adopt physics pond method Gains resources cost of transportation information.

14. methods as claimed in claim 11, is characterized in that, by the augmenting chain of unit of adjustment's flow least cost is obtained described minimum value as the augmenting chain that expense is minimum.

15. 1 kinds of virtual resources dispatching patchers, is characterized in that, this system comprises: distributed business network DSN multi-service resource manager, scheduling of resource decision-making device, distributed virtual machine Resource Scheduler, multiple virtual machine and multiple physical server, wherein,

DSN multi-service resource manager, for providing business interface according to type of service Business Entity externally, the resource mapping of the virtual machine that maintenance service type is corresponding and physical server, receive and ask from external management user, or, triggering dispatch request according to pre-setting strategy, exporting scheduling of resource decision-making device to;

Scheduling of resource decision-making device, for providing user interface, receiving scheduling request, according to the resource requirement information of virtual machine in current system and physical server, calculates and obtains virtual resources scheduling scheme, being sent to distributed virtual machine Resource Scheduler;

Distributed virtual machine Resource Scheduler, for according to the physical server resource flow information comprised in the virtual resources scheduling scheme received, dispatches the resource of related physical server;

Virtual machine, for obtaining scheduling resource from the physical server mapped, the Business Entity being outside by DSN multi-service resource manager provides the virtual resource of scheduling;

Physical server, for being virtualized into multiple separate virtual machine by running virtualization software, according to the capacity information that virtual resources scheduling scheme comprises, sets up the chnnels of resources of this capacity with respective physical server.

16. systems as claimed in claim 15, is characterized in that, described DSN multi-service resource manager is further used for obtaining Business Entity load information, implement user and ask shunting and adjustment.

17. systems as claimed in claim 16, it is characterized in that, described virtual resources scheduling scheme comprises: do not shut down virtual machine (vm) migration, local resource flowing and strange land resource flow.

18. systems as described in any one of claim 15 to 17, it is characterized in that, described scheduling of resource decision-making device comprises: mapping block and computing module, wherein,

Mapping block, for providing user interface, actual schedule scene map is fundamental plan model by receiving scheduling request, according to the resource requirement information of virtual machine in current system and physical server, determine the network description parameter sets of optimizing scheduling target and correspondence;

Computing module, for the network description parameter sets determined according to mapping block, calculates and the virtual resources scheduling scheme of Gains resources flowing, is sent to distributed virtual machine Resource Scheduler.

19. systems as claimed in claim 18, it is characterized in that, described distributed virtual machine Resource Scheduler comprises: monitoring modular and enforcement module, wherein,

Monitoring modular, for extracting the parameter of fundamental plan model need and the state parameter for determining mapping block employing fundamental plan problem model from real network environment;

Implement module, for according to the physical server resource flow information comprised in the virtual resources scheduling scheme received, dispatch the resource of related physical server.

20. systems as claimed in claim 19, it is characterized in that, described monitoring modular comprises parameter acquiring submodule and status monitoring submodule, wherein,

Parameter acquiring submodule, for extracting limiting parameter needed for plan model and cost parameter from real network environment;

Status monitoring submodule, extracts for determining that mapping block adopts the state parameter of fundamental plan problem model from real network environment for completing.