CN105812461A - Mobile cloud environment context awareness computing migration method - Google Patents

Abstract

The invention relates to a mobile cloud environment context awareness computing migration method. The method includes the following steps that: a design pattern supporting computing dynamic migration in an application is provided; an evaluation model is put forward, so that optimal cloud resources can be automatically selected for computing migration based on the context of a mobile device; and a framework for supporting the above design model and the evaluation model is realized. The method of the invention can be flexibly applied to a complex and changeable mobile cloud environment. Compared with a traditional migration method, the method of the invention can greatly improve the performance of the mobile device and user experience and enhance the endurance of the mobile device.

Description

A kind of mobile cloud environment context aware computation migration method

Technical field

The present invention relates to mobile field of cloud calculation, particularly a kind of mobile cloud environment context aware computation migration method.

Background technology

Along with popularizing of intelligent mobile handheld device, emerge numerous abundant in content application so that performance and the energy consumption problem of mobile equipment become increasingly conspicuous.Computation migration becomes currently a popular solution, and it passes through " calculating " task of Mobile solution from terminal transfer to Cloud Server, thus alleviating terminal unit computation burden, reducing the electrical source consumption of terminal unit simultaneously and promoting its flying power.Traditional moving method great majority adopt the server or cloud specified in advance.But, these traditional methods can not be flexibly applied to mobile cloud environment complicated and changeable, and main cause has following two aspects:

On the one hand, along with the movement of equipment, its context environmental (such as position, network condition) changes constantly.Such as, local at some, mobile equipment can use free WIFI to connect；And local at other, it can only use 3G connection to replace.With regard to speed and energy consumption, the performance of each connection is likely to difference.

On the other hand, along with increasing of cloud, mist, intelligent terminal etc., there is a lot of optional cloud resource, such as public cloud computing service, neighbouring cloudlet.At different conditions, it is necessary to select different cloud resources to migrate.Such as, when network environment is poor, we can consider cloudlet compute-intensive applications being moved to surrounding, postpones by reducing, thus improving performance and the Consumer's Experience of mobile equipment.

Summary of the invention

In view of this, the purpose of the present invention is to propose to a kind of mobile cloud environment context aware computation migration method, context environmental and available cloud resource based on current mobile device, it is possible to by a kind of cost-effective in the way of be dynamically selected the best cloud resource for migrating.

The core concept of the present invention is, based on context environmental and the available cloud resource of mobile equipment, selects the cloud resource of the best dynamically at runtime, and the calculating task code dynamic migration migrated will be needed to run to this resource.The present invention is made up of three parts: a kind of design pattern, an assessment models, and a kind of context aware computation migration framework.First, it is proposed that a kind of design pattern supporting to calculate dynamic migration in application so that all parts in application can be on-demand in Local or Remote interoperability.Wherein, calculate task code can dynamically be migrated and dispose.Secondly, it is proposed that an assessment models, according to the context environmental of mobile equipment and available cloud resource, the resource automatically selecting the best migrates.In order to improve the availability of migration service, we have introduced service pool concept, and each service pool has been polymerized local service, optimal service and the most frequent available service.In client, it would be desirable to data such as monitoring in real time and collection computation task, mobile device context environment, and cloud Resource Calculation abilities, and respectively it is modeled.According to model information and runtime data, we select optimal service and the most frequent available service at proposed decision making algorithm with regard to dynamic.Finally, it is achieved that a kind of context aware computation migration framework is for supporting above-mentioned design pattern and assessment models.

The present invention adopts below scheme to realize: a kind of mobile cloud environment context aware computation migration method, specifically includes following steps；

Step S1: propose a kind of design pattern supporting to calculate dynamic migration in application so that all parts in application can be on-demand in Local or Remote interoperability；

Step S2: propose an assessment models, according to the context environmental of mobile equipment and available cloud resource, the resource automatically selecting the best migrates；

Step S3: propose a kind of context aware computation migration framework for supporting design pattern and the step S2 assessment models of step S1.

Further, the design pattern in described step S1 includes programming model, deployment process and servicing operating mode；

As it is shown in figure 1, described programming model includes a primary module and a service registry storehouse；Described primary module defines the code that can not be migrated, described service registry storehouse is registered all services that can migrate；

Described deployment process comprises the following steps: step S11: mobile node finds target cloud resource, and described target cloud resource is carried out authentication；Step S12: upon successful authentication, mobile node sends all information relevant to current service deployment to described target cloud resource；Step S13: described target cloud resource receives request, and starts to check the deployment whether current environment can support this service, if it can, and required service be absent from, enter step S14, otherwise, enter step S15；Step S14: described target cloud resource obtains corresponding service document from mobile node and disposes, and service description file is returned to mobile node；Step S15: described target cloud resource returns the information of " service exists " or " can not dispose " to mobile node；Fig. 2 describes the whole workflow of mechanism.

As it is shown on figure 3, described servicing operating mode provides a kind of object construction supporting to calculate on-demand long-range execution in application, it is allowed to primary module can effectively call other service regardless of currently they operate in same VM or at different resource nodes.This structure includes three core elements: local adapter, long-range adapter, service pool；

Described service pool is single Virtual Service, but has been polymerized the identical candidate service being deployed in different cloud resource of multiple function, and the quality of described candidate service is different from；When receiving call request, preferentially top-quality service can be selected from service pool to call；

The position that the described local adapter primary responsibility identification service of being called is current, and forward the method to call；When primary module calls the method for Service1, first local adapter finds the service pool of correspondence, and obtains the description file of service best in quality in service pool.From service description file, it can recognize that the position that Service1 is current.If Service1 is at local runtime, then local adapter can call by direct retransmission method so that need not through network stack when primary module calls Service1.If Service1 is in long-range execution, local adapter obtains the url of service from service description file, and forwards the method to call to long-range adapter.

Described HTTP request, when receiving HTTP request, can be resolved by described long-range adapter, obtains Service name, method name and parameter information；And invocation target method；After goal approach has performed, corresponding result can be transferred back to client, merges with former application.In whole process, if the position change of called service, such as having transferred to remote node from this locality, or transferred to another node, caller from a certain remote node, namely primary module is not aware that the change of called service position.

Further, the described code that can not be migrated includes following three classes: realize relevant to using user interface；Employ I/O equipment；Employ arbitrary external module.

Specifically, when exploitation application, programming personnel, it is first necessary to the method in application is classified, is divided into anchored and movable two class.The method of Anchored type must be stayed in primary module, can only perform on the mobile apparatus；Mainly include following a few class: 1) realize relevant to using user interface；2) I/O equipment is employed, accelerometer on such as reading equipment, GPS etc.；3) employ arbitrary external module, such as use network to connect and perform e-commerce transaction.These methods have used some resources that can only could obtain on the mobile apparatus.Performing if these methods are transferred in cloud resource, they can cause execution mistake because can not find required resource.The method that additive method outside these methods is all classified as movable type automatically, namely they both can perform can also perform in cloud resource on the mobile apparatus.The method of movable type must be developed to service, and each service encapsulates independent applied function module；And they are required in service registry storehouse to register.Each service in service registry storehouse has the service pool of correspondence, the service that in this service pool, one group of function of dynamic aggregation is identical.At first, only one of which service in service pool, namely local service.When application runs, primary module can by " Service name. method name (parameter information) " form calls service.

Further, described step S2 specifically includes following steps:

Step S21: respectively computation task, mobile device context environment and cloud Resource Calculation ability are modeled, obtain task-function model, network-resource model, task-resource model；Perform when a task is migrated in long-distance cloud resource, total time T_totalMainly it is made up of two parts: T_total=T_server+T_transfer, wherein, T_severFor server execution time, T_transferFor network latency；

Step S22: structure service pool: according to device context environment and available cloud resource, it is proposed to decision making algorithm dynamically selects optimal service and the most frequent available service；Then optimal service, the most frequent available service and local service are dynamically aggregated into a service pool, when optimal service and the most frequent available service are same service, and only two services in service pool；Meanwhile, service pool can define according to service quality and call order；When service is called, can preferentially call top-quality service in service pool；Described service quality is the execution time；

Step S23: collect model information, including data monitoring and information；

Wherein said data monitoring is in order to monitor the execution of application and current environment in real time；Monitored data include: the title of current network, resource information available under current network, under current network, the transfer rate of all available resources and round-trip delay, current task information, current task move to the total execution time on different resource；

Described information includes actual execution and test execution；Described actual execution is: when mobile equipment enters a network, the service in prioritizing selection service pool truly connects or execution task, and corresponding result is stored in model information；Described test execution is: do not have the resource in service pool to test those.Fast-changing runtime environment is a big characteristic of mobile computing；If be select the resource in service pool, it is more likely that local optimum can be absorbed in every time.Such as, certain resource is likely to be due to Network Abnormal, causes current time performance extreme difference；But this does not represent the next moment, and it will not improve.It would therefore be desirable to do not have the resource in service pool to be also carried out test those.We are provided with information timeliness Time basic time for each cloud resource；If a certain information overaging time, then submitting test assignment to: network-resource model is tested when corresponding network, task-resource model is tested when connecting this resource.If test is basically unchanged, then Time increases every time, otherwise reduces；Same search engine.In order to obtain information required in task-function model, network-resource model, task-resource model, the result of actual execution or test execution all can be saved in data base every time.Additionally, when network changes, it would be desirable to recording the information of a network and current network, the assessment for the most frequent available service provides support.

Further, described in step S21, task-function model is:

Making T represent a group task set, for each the task t in T, its transmission time is calculated by below equation:

$T_{transfer}(t,v,rtt)=max\{\frac{C\left(t\right)}{v},rtt\}$

Wherein, C (t) represents the volume of transmitted data of task t, and it is to be obtained by training in advance, and v represents the message transmission rate size in network, and rtt represents the round-trip delay between equipment and cloud resource；Different Mobile solution have different volumes of transmitted data, including the result of the input data sent and reception.Such as, face recognition application needs bigger volume of transmitted data, and quintet game application only needs relatively small volume of transmitted data.

In described task-function model, all it is independent between all tasks.

Further, described in step S21, network-resource model is:

Make N={n₁,n₂,...,n_h, represent h collection of network, S={s₁,s₂,...,s_mRepresenting m cloud resource collection, V and RTT is the matrix of two h × m, have recorded the expected value of the transfer rate of each resource in heterogeneous networks situation and round-trip delay respectively；The definition of V and RTT matrix is as follows:

Wherein, v_ij, r_ijIt is illustrated respectively in network n_iUnder, equipment is to cloud resource s_jTransfer rate and the expected value of round-trip delay；For each element v in V and RTT_ij, r_ij, they are calculated by following expectation function:

$v_{ij}=W^T\·V^{ij}=\left(\begin{array}{cccc}{w}_1& w_2& ...& w_p\end{array}\right)\left(\begin{array}{c}{v_1}^{ij}\\ {v_2}^{ij}\\ ...\\ {v_p}^{ij}\end{array}\right);$

$r_{ij}=W^T\·{\mathrm{RTT}}^{ij}=\left(\begin{array}{cccc}{w}_1& w_2& ...& w_p\end{array}\right)\left(\begin{array}{c}{r_1}^{ij}\\ {r_2}^{ij}\\ ...\\ {r_p}^{ij}\end{array}\right);$

s.t.w₁+w₂+…+w_p=1；

Vector W^TIn one weight factor of each element representation, adjust the scale of these elements according to different scenes；V^ijHave recorded at network n_iUnder, cloud resource s_jThe historical series of transfer rate, RTT^ijHave recorded at network n_iUnder, cloud resource s_jThe historical series of round-trip delay；Historical series is obtained by actual execution or test execution.

Also defining the Matrix C of a h × h, for recording the variation tendency of customer location, Matrix C definition is as follows:

Wherein, c_ij(i≠j)Have recorded users from networks n_iMove to network n_jNumber of times.

Further, described in step S21, task-resource model is:

Different cloud resources have different disposal ability.The task T={t of given one group of independence₁,t₂,...,t_rAnd one group of cloud resource S={s₁,s₂,...,s_m, matrix E have recorded the different task expected value performing the time in each resource, then the matrix E definition of this r × m is as follows:

Wherein, e_kjExpression task t_kAt cloud resource s_jOn perform the time expected value, for each element e in matrix E_kj, it is calculated each through following expectation function:

$e_{kj}=W^T\·E^{kj}=\left(\begin{array}{cccc}{w}_1& w_2& ...& w_p\end{array}\right)\left(\begin{array}{c}{e_1}^{kj}\\ {e_2}^{kj}\\ ...\\ {e_p}^{kj}\end{array}\right);$

s.t.w₁+w₂+…+w_p=1；

Wherein, E^kjHave recorded task t_kAt cloud resource s_jOn perform the time historical series.

Further, optimal service described in described step S22 refers to the service that the actual efficiency under current network is the highest, for the computation migration of current application, the service that wherein actual efficiency is the highest must is fulfilled for performing total time less than the time calling local service, specifically includes following steps:

Step 1: operationally collecting all of contextual information as input parameter, described contextual information includes current task t_k, essential information model and mobile equipment current network n_i；

Step 2: for given task t_k, assess it in local execution total time, be designated as local_cost；

Step 3: according to input parameter, get cloud resource one group available, represent with S；Then, all available cloud resources are all estimated and select assessed value minimum cloud resource as current network optimal service；Valuation functions is:

$offloading\_\cos t=\underset{\∀s_j\∈S}{\min }(e_{kj}+T_{transfer}(t_k,v_{ij},r_{ij}\left)\right);$

Step 4: if the Network Abnormal of selected cloud resource, then need to select next one and so on according to assessed value；If the network of selected resource is normal, its assessed value is less than local_cost simultaneously, then return this resource, and algorithm terminates；

The most frequent available service described in described step S22 is according to essential information model, the currently available and the most available service that assessment obtains, and namely performs total time less than the time calling local service；It uses when network condition changes, and is used for providing seamless migration service.In order to automatically select out the most frequent available service, we have proposed algorithm 2.The core of algorithm 2 is a performance evaluation function, and this object function is made up of two parts, and a part is for all possible network, is worth the network weight sum moving to this resource, is designated as Worthy；Another part is in the network that all values must migrate, and moves to the temporal summation saved in this resource, is designated as Save.Therefore, in all possible network, task t is specifying the total benefit definition performed in resource as follows:

Benefits (t)=α₁*Worthy(t)+α₂*Save(t)

Wherein, α₁, α₂It is weight factor, in our framework, it is possible to adjust the scale of these elements according to different scenes.As shown in algorithm 2, whole decision process includes following four step:

Step 1: operationally collecting all of contextual information as input parameter, described contextual information includes current task t_k, essential information model and mobile equipment current network n_i；

Step 2: for given task t_k, assess it in local execution total time, be designated as local_cost；

Step 3: all available resource in current network is estimated, and select the maximum cloud resource of benefit as the most frequent available service；

Step 4: if the Network Abnormal of selected cloud resource, then need to select next one and so on according to assessed value；Otherwise returning this cloud resource, algorithm terminates.

Further, the context aware computation migration framework that described step S3 proposes have employed client-server communication model；Wherein, cloud resource is server, and mobile equipment is client, and it is able to access that the service provided on server, as shown in Figure 4；

Administrative mechanism, computation migration supporting mechanism, services selection supporting mechanism when described client realizes running；During described operation, administrative mechanism includes data monitoring and module management；Described computation migration supporting mechanism includes the proxy call module of the resource registering module of client, service pool module, client；Described services selection supporting mechanism includes model information module, service selecting module, service testing module；

1) service call: when carrying out service call, proxy call module (adapter) according to corresponding Service name, method name, parameter information, can call top-quality service from service pool module.

2) service pool configuration: during operation, administrative mechanism is constantly carrying out data monitoring.If it monitors network and changes or call inefficacy, service selecting module will in conjunction with operation time administrative mechanism the currently monitored to data and model information module in the information collected re-start services selection.Meanwhile, resource registering module and service pool module are updated accordingly.

Described server includes the proxy call module of the resource registering module of server end, service library module and server end；All available resource nodes are responsible for and are registered to the resource registering module of described server end；Described service library module is in order to register the service that be there is a need to be deployed on remote resource；The dynamic call serviced when the proxy call module of described server end is in order to process operation.Proxy call module with client is similar, by call request is resolved, dynamically calls on required service from service library module.

Compared with prior art, on the one hand, due to the mobility of equipment, its context environmental changes constantly；On the other hand, along with increasing of cloud, mist, intelligent terminal etc., there is a lot of optional cloud resource, these all bring new challenge to computation migration.Traditional moving method great majority adopt the server specified in advance, and they according to current environment, cannot be dynamically selected that disposal ability is powerful and the good cloud resource of network condition.The present invention is based on the context environmental of current mobile device and available cloud resource, it is possible to by a kind of cost-effective in the way of be dynamically selected the cloud resource of the best for migrating: first, provide a kind of design pattern supporting to calculate dynamic migration in application；Secondly, it is proposed to an assessment models, based on the context environmental of mobile equipment, it is possible to automatically select the cloud resource of the best for computation migration；Finally, it is achieved a kind of framework supports above-mentioned put forward design pattern and assessment models.The present invention can be flexibly applied to mobile cloud environment complicated and changeable.Compared to traditional moving method, it can improve performance and the Consumer's Experience of mobile equipment significantly, and promotes its flying power.

Accompanying drawing explanation

Fig. 1 is computation migration programming model schematic diagram of the present invention.

Fig. 2 is service arrangement workflow diagrams of the present invention.

Fig. 3 is that the present invention services on-demand far call pattern diagram when running.

Fig. 4 is context aware computation migration block schematic illustration of the present invention.

Detailed description of the invention

Below in conjunction with drawings and Examples, the present invention will be further described.

Present embodiments provide a kind of mobile cloud environment context aware computation migration method, specifically include following steps；

Step S1: propose a kind of design pattern supporting to calculate dynamic migration in application so that all parts in application can be on-demand in Local or Remote interoperability；

Step S2: propose an assessment models, according to the context environmental of mobile equipment and available cloud resource, the resource automatically selecting the best migrates；

Step S3: propose a kind of context aware computation migration framework for supporting design pattern and the step S2 assessment models of step S1.

In the present embodiment, the design pattern in described step S1 includes programming model, deployment process and servicing operating mode；

As it is shown in figure 1, described programming model includes a primary module and a service registry storehouse；Described primary module defines the code that can not be migrated, described service registry storehouse is registered all services that can migrate；

Described deployment process comprises the following steps: step S11: mobile node finds target cloud resource, and described target cloud resource is carried out authentication；Step S12: upon successful authentication, mobile node sends all information relevant to current service deployment to described target cloud resource；Step S13: described target cloud resource receives request, and starts to check the deployment whether current environment can support this service, if it can, and required service be absent from, enter step S14, otherwise, enter step S15；Step S14: described target cloud resource obtains corresponding service document from mobile node and disposes, and service description file is returned to mobile node；Step S15: described target cloud resource returns the information of " service exists " or " can not dispose " to mobile node；Fig. 2 describes the whole workflow of mechanism.

As it is shown on figure 3, described servicing operating mode provides a kind of object construction supporting to calculate on-demand long-range execution in application, it is allowed to primary module can effectively call other service regardless of currently they operate in same VM or at different resource nodes.This structure includes three core elements: local adapter, long-range adapter, service pool；

Described service pool is single Virtual Service, but has been polymerized the identical candidate service being deployed in different cloud resource of multiple function, and the quality of described candidate service is different from；When receiving call request, preferentially top-quality service can be selected from service pool to call；

The position that the described local adapter primary responsibility identification service of being called is current, and forward the method to call；When primary module calls the method for Service1, first local adapter finds the service pool of correspondence, and obtains the description file of service best in quality in service pool.From service description file, it can recognize that the position that Service1 is current.If Service1 is at local runtime, then local adapter can call by direct retransmission method so that need not through network stack when primary module calls Service1.If Service1 is in long-range execution, local adapter obtains the url of service from service description file, and forwards the method to call to long-range adapter.

Described HTTP request, when receiving HTTP request, can be resolved by described long-range adapter, obtains Service name, method name and parameter information；And invocation target method；After goal approach has performed, corresponding result can be transferred back to client, merges with former application.In whole process, if the position change of called service, such as having transferred to remote node from this locality, or transferred to another node, caller from a certain remote node, namely primary module is not aware that the change of called service position.

In the present embodiment, the described code that can not be migrated includes following three classes: realize relevant to using user interface；Employ I/O equipment；Employ arbitrary external module.

Specifically, in the present embodiment, when exploitation application, programming personnel, it is first necessary to the method in application is classified, is divided into anchored and movable two class.The method of Anchored type must be stayed in primary module, can only perform on the mobile apparatus；Mainly include following a few class: 1) realize relevant to using user interface；2) I/O equipment is employed, accelerometer on such as reading equipment, GPS etc.；3) employ arbitrary external module, such as use network to connect and perform e-commerce transaction.These methods have used some resources that can only could obtain on the mobile apparatus.Performing if these methods are transferred in cloud resource, they can cause execution mistake because can not find required resource.The method that additive method outside these methods is all classified as movable type automatically, namely they both can perform can also perform in cloud resource on the mobile apparatus.The method of movable type must be developed to service, and each service encapsulates independent applied function module；And they are required in service registry storehouse to register.Each service in service registry storehouse has the service pool of correspondence, the service that in this service pool, one group of function of dynamic aggregation is identical.At first, only one of which service in service pool, namely local service.When application runs, primary module can by " Service name. method name (parameter information) " form calls service.

In the present embodiment, described step S2 specifically includes following steps:

Step S21: respectively computation task, mobile device context environment and cloud Resource Calculation ability are modeled, obtain task-function model, network-resource model, task-resource model；Perform when a task is migrated in long-distance cloud resource, total time T_totalMainly it is made up of two parts: T_total=T_server+T_transfer, wherein, T_severFor server execution time, T_transferFor network latency；

Step S22: structure service pool: according to device context environment and available cloud resource, it is proposed to decision making algorithm dynamically selects optimal service and the most frequent available service；Then optimal service, the most frequent available service and local service are dynamically aggregated into a service pool, when optimal service and the most frequent available service are same service, and only two services in service pool；Meanwhile, service pool can define according to service quality and call order；When service is called, can preferentially call top-quality service in service pool；Described service quality is the execution time；

Step S23: collect model information, including data monitoring and information；

Wherein said data monitoring is in order to monitor the execution of application and current environment in real time；Monitored data include: the title of current network, resource information available under current network, under current network, the transfer rate of all available resources and round-trip delay, current task information, current task move to the total execution time on different resource；

Described information includes actual execution and test execution；Described actual execution is: when mobile equipment enters a network, the service in prioritizing selection service pool truly connects or execution task, and corresponding result is stored in model information；Described test execution is: do not have the resource in service pool to test those.Fast-changing runtime environment is a big characteristic of mobile computing；If be select the resource in service pool, it is more likely that local optimum can be absorbed in every time.Such as, certain resource is likely to be due to Network Abnormal, causes current time performance extreme difference；But this does not represent the next moment, and it will not improve.It would therefore be desirable to do not have the resource in service pool to be also carried out test those.We are provided with information timeliness Time basic time for each cloud resource；If a certain information overaging time, then submitting test assignment to: network-resource model is tested when corresponding network, task-resource model is tested when connecting this resource.If test is basically unchanged, then Time increases every time, otherwise reduces；Same search engine.In order to obtain information required in task-function model, network-resource model, task-resource model, the result of actual execution or test execution all can be saved in data base every time.Additionally, when network changes, it would be desirable to recording the information of a network and current network, the assessment for the most frequent available service provides support.

In the present embodiment, described in step S21, task-function model is:

Making T represent a group task set, for each the task t in T, its transmission time is calculated by below equation:

$T_{transfer}(t,v,rtt)=max\{\frac{C\left(t\right)}{v},rtt\}$

Wherein, C (t) represents the volume of transmitted data of task t, and it is to be obtained by training in advance, and v represents the message transmission rate size in network, and rtt represents the round-trip delay between equipment and cloud resource；Different Mobile solution have different volumes of transmitted data, including the result of the input data sent and reception.Such as, face recognition application needs bigger volume of transmitted data, and quintet game application only needs relatively small volume of transmitted data.

In described task-function model, all it is independent between all tasks.

In the present embodiment, described in step S21, network-resource model is:

Make N={n₁,n₂,...,n_h, represent h collection of network, S={s₁,s₂,...,s_mRepresenting m cloud resource collection, V and RTT is the matrix of two h × m, have recorded the expected value of the transfer rate of each resource in heterogeneous networks situation and round-trip delay respectively；The definition of V and RTT matrix is as follows:

Wherein, v_ij, r_ijIt is illustrated respectively in network n_iUnder, equipment is to cloud resource s_jTransfer rate and the expected value of round-trip delay；For each element v in V and RTT_ij, r_ij, they are calculated by following expectation function:

$v_{ij}=W^T\·V^{ij}=\left(\begin{array}{cccc}{w}_1& w_2& ...& w_p\end{array}\right)\left(\begin{array}{c}{v_1}^{ij}\\ {v_2}^{ij}\\ ...\\ {v_p}^{ij}\end{array}\right);$

$r_{ij}=W^T\·{\mathrm{RTT}}^{ij}=\left(\begin{array}{cccc}{w}_1& w_2& ...& w_p\end{array}\right)\left(\begin{array}{c}{r_1}^{ij}\\ {r_2}^{ij}\\ ...\\ {r_p}^{ij}\end{array}\right);$

s.t.w₁+w₂+…+w_p=1；

Vector W^TIn one weight factor of each element representation, adjust the scale of these elements according to different scenes；V^ijHave recorded at network n_iUnder, cloud resource s_jThe historical series of transfer rate, RTT^ijHave recorded at network n_iUnder, cloud resource s_jThe historical series of round-trip delay；Historical series is obtained by actual execution or test execution.

Also defining the Matrix C of a h × h, for recording the variation tendency of customer location, Matrix C definition is as follows:

Wherein, c_ij(i≠j)Have recorded users from networks n_iMove to network n_jNumber of times.

In the present embodiment, described in step S21, task-resource model is:

Different cloud resources have different disposal ability.The task T={t of given one group of independence₁,t₂,...,t_rAnd one group of cloud resource S={s₁,s₂,...,s_m, matrix E have recorded the different task expected value performing the time in each resource, then the matrix E definition of this r × m is as follows:

Wherein, e_kjExpression task t_kAt cloud resource s_jOn perform the time expected value, for each element e in matrix E_kj, it is calculated each through following expectation function:

$e_{kj}=W^T\·E^{kj}=\left(\begin{array}{cccc}{w}_1& w_2& ...& w_p\end{array}\right)\left(\begin{array}{c}{e_1}^{kj}\\ {e_2}^{kj}\\ ...\\ {e_p}^{kj}\end{array}\right);$

s.t.w₁+w₂+…+w_p=1；

Wherein, E^kjHave recorded task t_kAt cloud resource s_jOn perform the time historical series.

In the present embodiment, optimal service described in described step S22 refers to the service that the actual efficiency under current network is the highest, for the computation migration of current application, the service that wherein actual efficiency is the highest must is fulfilled for performing total time less than the time calling local service, specifically includes following steps:

Step 1: operationally collecting all of contextual information as input parameter, described contextual information includes current task t_k, essential information model and mobile equipment current network n_i；

Step 2: for given task t_k, assess it in local execution total time, be designated as local_cost；

Step 3: according to input parameter, get cloud resource one group available, represent with S；Then, all available cloud resources are all estimated and select assessed value minimum cloud resource as current network optimal service；Valuation functions is:

$offloading\_\cos t=\underset{\∀s_j\∈S}{\min }(e_{kj}+T_{transfer}(t_k,v_{ij},r_{ij}\left)\right);$

Step 4: if the Network Abnormal of selected cloud resource, then need to select next one and so on according to assessed value；If the network of selected resource is normal, its assessed value is less than local_cost simultaneously, then return this resource, and algorithm terminates；

Specific algorithm is as shown in the table:

The most frequent available service described in described step S22 is according to essential information model, the currently available and the most available service that assessment obtains, and namely performs total time less than the time calling local service；It uses when network condition changes, and is used for providing seamless migration service.In order to automatically select out the most frequent available service, we have proposed algorithm 2.The core of algorithm 2 is a performance evaluation function, and this object function is made up of two parts, and a part is for all possible network, is worth the network weight sum moving to this resource, is designated as Worthy；Another part is in the network that all values must migrate, and moves to the temporal summation saved in this resource, is designated as Save.Therefore, in all possible network, task t is specifying the total benefit definition performed in resource as follows:

Benefits (t)=α₁*Worthy(t)+α₂*Save(t)

Wherein, α₁, α₂It is weight factor, in our framework, it is possible to adjust the scale of these elements according to different scenes.As shown in algorithm 2, whole decision process includes following four step:

Step 1: operationally collecting all of contextual information as input parameter, described contextual information includes current task t_k, essential information model and mobile equipment current network n_i；

Step 2: for given task t_k, assess it in local execution total time, be designated as local_cost；

Step 3: all available resource in current network is estimated, and select the maximum cloud resource of benefit as the most frequent available service；

Step 4: if the Network Abnormal of selected cloud resource, then need to select next one and so on according to assessed value；Otherwise returning this cloud resource, algorithm terminates.

Shown in algorithm 2 table specific as follows:

In the present embodiment, the context aware computation migration framework that described step S3 proposes have employed client-server communication model；Wherein, cloud resource is server, and mobile equipment is client, and it is able to access that the service provided on server, as shown in Figure 4；

Administrative mechanism, computation migration supporting mechanism, services selection supporting mechanism when described client realizes running；During described operation, administrative mechanism includes data monitoring and module management；Described computation migration supporting mechanism includes the proxy call module of the resource registering module of client, service pool module, client；Described services selection supporting mechanism includes model information module, service selecting module, service testing module；

1) service call: when carrying out service call, proxy call module (adapter) according to corresponding Service name, method name, parameter information, can call top-quality service from service pool module.

2) service pool configuration: during operation, administrative mechanism is constantly carrying out data monitoring.If it monitors network and changes or call inefficacy, service selecting module will in conjunction with operation time administrative mechanism the currently monitored to data and model information module in the information collected re-start services selection.Meanwhile, resource registering module and service pool module are updated accordingly.

Described server includes the proxy call module of the resource registering module of server end, service library module and server end；All available resource nodes are responsible for and are registered to the resource registering module of described server end；Described service library module is in order to register the service that be there is a need to be deployed on remote resource；The dynamic call serviced when the proxy call module of described server end is in order to process operation.Proxy call module with client is similar, by call request is resolved, dynamically calls on required service from service library module.

The foregoing is only presently preferred embodiments of the present invention, all equalizations done according to the present patent application the scope of the claims change and modify, and all should belong to the covering scope of the present invention.

Claims (9)

1. a mobile cloud environment context aware computation migration method, it is characterised in that comprise the following steps；

Step S1: propose a kind of design pattern supporting to calculate dynamic migration in application so that all parts in application can be on-demand in Local or Remote interoperability；

Step S2: propose an assessment models, according to the context environmental of mobile equipment and available cloud resource, the resource automatically selecting the best migrates；

Step S3: propose a kind of context aware computation migration framework for supporting design pattern and the step S2 assessment models of step S1.

2. a kind of mobile cloud environment context aware computation migration method according to claim 1, it is characterised in that: the design pattern in described step S1 includes programming model, deployment process and servicing operating mode；

Described programming model includes a primary module and a service registry storehouse；Described primary module defines the code that can not be migrated, described service registry storehouse is registered all services that can migrate；

Described deployment process comprises the following steps: step S11: mobile node finds target cloud resource, and described target cloud resource is carried out authentication；Step S12: upon successful authentication, mobile node sends all information relevant to current service deployment to described target cloud resource；Step S13: described target cloud resource receives request, and starts to check the deployment whether current environment can support this service, if it can, and required service be absent from, enter step S14, otherwise, enter step S15；Step S14: described target cloud resource obtains corresponding service document from mobile node and disposes, and service description file is returned to mobile node；Step S15: described target cloud resource returns the information of " service exists " or " can not dispose " to mobile node；

Described servicing operating mode provides a kind of object construction supporting to calculate on-demand long-range execution in application, and this structure includes three core elements: local adapter, long-range adapter, service pool；

Described service pool is single Virtual Service, but has been polymerized the identical candidate service being deployed in different cloud resource of multiple function, and the quality of described candidate service is different from；When receiving call request, preferentially top-quality service can be selected from service pool to call；

The position that the described local adapter primary responsibility identification service of being called is current, and forward the method to call；

Described HTTP request, when receiving HTTP request, can be resolved by described long-range adapter, obtains Service name, method name and parameter information；And invocation target method；After goal approach has performed, corresponding result can be transferred back to client, merges with former application.

3. a kind of mobile cloud environment context aware computation migration method according to claim 2, it is characterised in that: the described code that can not be migrated includes following three classes: realize relevant to using user interface；Employ I/O equipment；Employ arbitrary external module.

4. a kind of mobile cloud environment context aware computation migration method according to claim 1, it is characterised in that: described step S2 specifically includes following steps:

Step S21: respectively computation task, mobile device context environment and cloud Resource Calculation ability are modeled, obtain task-function model, network-resource model, task-resource model；Perform when a task is migrated in long-distance cloud resource, total time T_totalMainly it is made up of two parts: T_total=T_server+T_transfer, wherein, T_severFor server execution time, T_transferFor network latency；

Step S22: structure service pool: according to device context environment and available cloud resource, it is proposed to decision making algorithm dynamically selects optimal service and the most frequent available service；Then optimal service, the most frequent available service and local service are dynamically aggregated into a service pool, when optimal service and the most frequent available service are same service, and only two services in service pool；Meanwhile, service pool can define according to service quality and call order；When service is called, can preferentially call top-quality service in service pool；Described service quality is the execution time；

Step S23: collect model information, including data monitoring and information；

Wherein said data monitoring is in order to monitor the execution of application and current environment in real time；Monitored data include: the title of current network, resource information available under current network, under current network, the transfer rate of all available resources and round-trip delay, current task information, current task move to the total execution time on different resource；

Described information includes actual execution and test execution；Described actual execution is: when mobile equipment enters a network, the service in prioritizing selection service pool truly connects or execution task, and corresponding result is stored in model information；Described test execution is: do not have the resource in service pool to test those.

5. a kind of mobile cloud environment context aware computation migration method according to claim 4, it is characterised in that: described in step S21, task-function model is:

Making T represent a group task set, for each the task t in T, its transmission time is calculated by below equation:

$T_{transfer}(t,v,rtt)=max\{\frac{C\left(t\right)}{v},rtt\}$

Wherein, C (t) represents the volume of transmitted data of task t, and it is to be obtained by training in advance, and v represents the message transmission rate size in network, and rtt represents the round-trip delay between equipment and cloud resource；

In described task-function model, all it is independent between all tasks.

6. a kind of mobile cloud environment context aware computation migration method according to claim 4, it is characterised in that: described in step S21, network-resource model is:

Make N={n₁,n₂,...,n_h, represent h collection of network, S={s₁,s₂,...,s_mRepresenting m cloud resource collection, V and RTT is the matrix of two h × m, have recorded the expected value of the transfer rate of each resource in heterogeneous networks situation and round-trip delay respectively；The definition of V and RTT matrix is as follows:

Wherein, v_ij, r_ijIt is illustrated respectively in network n_iUnder, equipment is to cloud resource s_jTransfer rate and the expected value of round-trip delay；For each element v in V and RTT_ij, r_ij, they are calculated by following expectation function:

$v_{ij}=W^T\·V^{ij}=\left(\begin{array}{cccc}{w}_1& w_2& ...& w_p\end{array}\right)\left(\begin{array}{c}{v_1}^{ij}\\ {v_2}^{ij}\\ ...\\ {v_p}^{ij}\end{array}\right);$

$r_{ij}=W^T\·{\mathrm{RTT}}^{ij}=\left(\begin{array}{cccc}{w}_1& w_2& ...& w_p\end{array}\right)\left(\begin{array}{c}{r_1}^{ij}\\ {r_2}^{ij}\\ ...\\ {r_p}^{ij}\end{array}\right);$

s.t.w₁+w₂+…+w_p=1；

Vector W^TIn one weight factor of each element representation, adjust the scale of these elements according to different scenes；V^ijHave recorded at network n_iUnder, cloud resource s_jThe historical series of transfer rate, RTT^ijHave recorded at network n_iUnder, cloud resource s_jThe historical series of round-trip delay；

Also defining the Matrix C of a h × h, for recording the variation tendency of customer location, Matrix C definition is as follows:

Wherein, c_ij(i≠j)Have recorded users from networks n_iMove to network n_jNumber of times.

7. a kind of mobile cloud environment context aware computation migration method according to claim 4, it is characterised in that: described in step S21, task-resource model is:

The task T={t of given one group of independence₁,t₂,...,t_rAnd one group of cloud resource S={s₁,s₂,...,s_m, matrix E have recorded the different task expected value performing the time in each resource, then the matrix E definition of this r × m is as follows:

Wherein, e_kjExpression task t_kAt cloud resource s_jOn perform the time expected value, for each element e in matrix E_kj, it is calculated each through following expectation function:

$e_{kj}=W^T\·E^{kj}=\left(\begin{array}{cccc}{w}_1& w_2& ...& w_p\end{array}\right)\left(\begin{array}{c}{e_1}^{kj}\\ {e_2}^{kj}\\ ...\\ {e_p}^{kj}\end{array}\right);$

s.t.w₁+w₂+…+w_p=1；

Wherein, E^kjHave recorded task t_kAt cloud resource s_jOn perform the time historical series.

8. a kind of mobile cloud environment context aware computation migration method according to claim 4, it is characterised in that:

Optimal service described in described step S22 refers to the service that the actual efficiency under current network is the highest, computation migration for current application, the service that wherein actual efficiency is the highest must is fulfilled for performing total time less than the time calling local service, specifically includes following steps:

Step 1: operationally collecting all of contextual information as input parameter, described contextual information includes current task t_k, essential information model and mobile equipment current network n_i；

Step 2: for given task t_k, assess it in local execution total time, be designated as local_cost；

Step 3: according to input parameter, get cloud resource one group available, represent with S；Then, all available cloud resources are all estimated and select assessed value minimum cloud resource as current network optimal service；Valuation functions is:

$offloading\_\cos t=\underset{\∀s_j\∈S}{min}(e_{kj}+T_{transfer}(t_k,v_{ij},r_{ij}\left)\right);$

Step 4: if the Network Abnormal of selected cloud resource, then need to select next one and so on according to assessed value；If the network of selected resource is normal, its assessed value is less than local_cost simultaneously, then return this resource, and algorithm terminates；

The most frequent available service described in described step S22 is according to essential information model, the currently available and the most available service that assessment obtains, and namely performs total time less than the time calling local service；Specifically include following steps:

Step 1: operationally collecting all of contextual information as input parameter, described contextual information includes current task t_k, essential information model and mobile equipment current network n_i；

Step 2: for given task t_k, assess it in local execution total time, be designated as local_cost；

Step 3: all available resource in current network is estimated, and select the maximum cloud resource of benefit as the most frequent available service；

Step 4: if the Network Abnormal of selected cloud resource, then need to select next one and so on according to assessed value；Otherwise returning this cloud resource, algorithm terminates.

9. a kind of mobile cloud environment context aware computation migration method according to claim 1, it is characterised in that:

The context aware computation migration framework that described step S3 proposes have employed client-server communication model；Wherein, cloud resource is server, and mobile equipment is client；

Administrative mechanism, computation migration supporting mechanism, services selection supporting mechanism when described client realizes running；During described operation, administrative mechanism includes data monitoring and module management；Described computation migration supporting mechanism includes the proxy call module of the resource registering module of client, service pool module, client；Described services selection supporting mechanism includes model information module, service selecting module, service testing module；

Described server includes the proxy call module of the resource registering module of server end, service library module and server end；All available resource nodes are responsible for and are registered to the resource registering module of described server end；Described service library module is in order to register the service that be there is a need to be deployed on remote resource；The dynamic call serviced when the proxy call module of described server end is in order to process operation.