CN103514047A - Task load balancing method used for mobile social network - Google Patents

Abstract

The invention discloses a task load balancing method used for a mobile social network. The task load balancing method used for the mobile social network comprises the steps that (1) each user judges whether other users are located in a communication range in each time slot, (2) d users are selected at random from all the users in the communication range, (3) loads of the d users are compared, (4) a task is allocated to the user with the lowest load among the d users, (5) if the number of the users in the communication range is smaller than d, the task is allocated to the user with the lowest load, (6) if no user is located in the communication range, allocation in a next time slot is waited, (7) when the task is completely allocated to all the users in one time slot, the allocation in the next time slot is started, and (8) the operation is finished. The task load balancing method used for the mobile social network achieves task load balance in pure distributed environments, takes the capacity level and computing power into full account, can be applied to the specially shaped mobile social network, is quite simple to implement, and can achieve a good load balancing effect when d is equal to 2.

Description

A kind of task load balance method for mobile social networking

Technical field

The method that the present invention relates to task reallocation in a kind of mobile social networking, its use can effectively be increased work efficiency, and saves the equipment energy.

Background technology

In recent years, smart mobile phone and relevant application program utilization rate significantly rise.The appearance of mass-rent and gunz perception allow more sensor information holder unconscious, cooperate of one's own accord.In such mobile computing environment, deeper cooperation is necessary.Mobile subscriber, when sharing and redistribute the task of they self varying number, also will consider the height of dump energy and computing power.For example, the user of battery electric quantity deficiency is unloaded to their task with it the user in communication range with higher battery electric quantity, and certainly, task also should be distributed to the node with high arithmetic capability.

Yet research work in the past, owing to paying close attention to multiple users share and mass-rent Data Collection, causes realizing load balancing between user.In this case, some user's queue length may be very high, and this will inevitably cause exhausting and postponing for a long time of the energy.Basic reason is, the efficiency of data sharing is all absorbed in these work, rather than the load balance between distributing user.

Traditional load balancing scheme can not directly apply to mobile social networking.Because in mobile social networking, it is all distributed that information and task are distributed.The features such as centralized scheme is infeasible, and mobile social networking distinctive duration of contact of Duan, task queue is dynamic also make some existing distributed algorithms guarantee that the balanced this respect of user load lost efficacy.

Summary of the invention

Technical matters: the present invention proposes a kind of task load balance method for mobile social networking, task redistribution method can be effectively goes task immigration to carry out to equipment more suitably, has also saved the equipment energy when increasing work efficiency.

Technical scheme: a kind of task load balance method for mobile social networking of the present invention is: the user of mobile social networking is when allocating task, need select least-loaded person in two or more users for each, and than the i.e. i.e. each method of selecting arbitrarily a user from user of least-loaded person and Random assignment in each user of selection of optimum allocation, not only need not know global information but also can reach well the load balance between user, said method comprising the steps of:

1), in each time slot, each user judges that other users are whether in communication range;

2) all users in communication range, select arbitrarily several users;

3) more select user's load, found out load the lowest in user;

4) assign the task to load the lowest;

5) if there is no user in communication range, wait for the distribution of next time slot;

6) in a time slot, all users distribute task, start the distribution of next time slot;

7) finish.

In step 1) in,

N user carries out random walk in limited area, the communication radius that wherein r is each mobile subscriber, when two nodes are all in mutual communication range, task can be reallocated, judge whether in communication range according to being:

(X _i-X _j) ²+(Y _i-Y _j) ²＜r ²

(X wherein _i, Y _i) and (X _j, Y _j) be respectively the position coordinates of user i and user j, if above formula is set up, show that user i and user j, in communication range, can carry out the reallocation of message exchange and task, if above formula is false, mark user i can not communicate by letter with user j.

Load balance between user, refers to being uniformly distributed of task, reduces to greatest extent the difference of queue length between each user, can be provided by following formula:

$\min \underset{i\∈U}{\mathrm{\Σ}}|Q_i-E[Q_i\left]\right|$

Wherein, U={1,2 ... n} is user's set, and i ∈ U represents that user i belongs to set U, Q _ithe queue length that represents user i, E[] be the mean value of stochastic variable, min is the minimum value of getting expression formula.

Beneficial effect: compared with prior art, its remarkable advantage is in the present invention: it is the task load balance under pure distributed environment, takes energy level and computing power into consideration, can be applied in the mobile social networking of isomery.Implement very simply, and when d=2, can reach good load balance effect.

Accompanying drawing explanation

Fig. 1 is the basic scene of task reallocation.

Fig. 2-Fig. 5 is the simulation analysis of communication radius on this method impact.

Fig. 6-Fig. 8 is the simulation analysis of task weight on this method impact.

Fig. 9 is the simulation analysis of user's computing power on this method impact.

Figure 10 is the simulation analysis of energy level on this method impact.

Figure 11 is the simulation analysis on this method impact that arranges of selecting parameter d.

Wherein Fig. 2-Figure 11 is cumulative distribution function curve diagram.

Embodiment

A kind of task load balance method for mobile social networking of the present invention is: the user of mobile social networking is when allocating task, need select least-loaded person in two or more users for each, and than the i.e. i.e. each method of selecting arbitrarily a user from user of least-loaded person and Random assignment in each user of selection of optimum allocation, not only need not know global information but also can reach well the load balance between user, said method comprising the steps of:

1), in each time slot, each user judges that other users are whether in communication range;

2) all users in communication range, select arbitrarily several users;

3) more select user's load, found out load the lowest in user;

4) assign the task to load the lowest;

5) if there is no user in communication range, wait for the distribution of next time slot;

6) in a time slot, all users distribute task, start the distribution of next time slot;

7) finish.

In step 1) in,

N user carries out random walk in limited area, the communication radius that wherein r is each mobile subscriber, when two nodes are all in mutual communication range, task can be reallocated, judge whether in communication range according to being:

(X _i-X _j) ²+(Y _i-Y _j) ²＜r ²

(X wherein _i, Y _i) and (X _j, Y _j) be respectively the position coordinates of user i and user j, if above formula is set up, show that user i and user j, in communication range, can carry out the reallocation of message exchange and task, if above formula is false, mark user i can not communicate by letter with user j.

Load balance between user, refers to being uniformly distributed of task, reduces to greatest extent the difference of queue length between each user, can be provided by following formula:

$\min \underset{i\∈U}{\mathrm{\Σ}}|Q_i-E[Q_i\left]\right|$

Wherein, U={1,2 ... n} is user's set, and i ∈ U represents that user i belongs to set U, Q _ithe queue length that represents user i, E[] be the mean value of stochastic variable, min is the minimum value of getting expression formula.

Evaluating the minimum value of the difference of user's average queue length and maximum queue length, is the index of evaluating average case and worst case gap:

In allocative decision, each task is endowed different weights, evaluates and can be expressed as:

$\min \mathrm{ma}{x}_{i,j\∈U}\underset{k=1}{\overset{\left|\right|Q_i\left|\right|}{\mathrm{\Σ}}}w\left(q_{\mathrm{ik}}\right)-E_{i\∈U}\left[W\left(Q_i\right)\right]$

Wherein, || Q _i|| represent the number of task in user i, q _ikrepresent k task in user i queue, w (q _ik) represent the weight of k task in user i;

represent to consider the actual loading of user i in task weight situation; Minmax represents to get the minimum value in some set maximal values;

When user node is endowed different computing powers, be expressed as follows formula:

$\min \mathrm{ma}{x}_{i,j\∈U}\underset{k=1}{\overset{\left|\right|Q_i\left|\right|}{\mathrm{\Σ}}}\frac{w\left(q_{\mathrm{ik}}\right)}{c_i}-E_{i\∈U}\left[\hat{W}\left(Q_i\right)\right]$

Wherein, c _ithe computing power that represents user,

represent to consider the actual loading of user i in computing power situation.

Below in conjunction with accompanying drawing and example, the invention will be further described.

1. in order to illustrate the method for task reallocation, we do the explanation as Fig. 1 to basic scene: in this figure, we can more clearly see the process of task reallocation.When user's 1 allocating task, user is considered to the center of communication range.For example, always have 5 users in user 1 communication range.1 of user needs wherein two of random chooses (user 2 and user 3), and task is consigned to the less user of load 2.Certainly, user 1 also can assign the task to different users according to different indexs such as dump energy, user's computing powers.

2. when distributing time slot to arrive for one, we generate user's position (X first at random _i, Y _i), check the distance of they and user j, record the user in user j communication range.Select several users in communication range, task is consigned to wherein least-loaded person, rather than simple Random assignment.If unfortunate, there is no user in communication radius, when a time slot arrives instantly, continue to distribute.After determining the user s that selects payment task, make the queue length of s add 1, we use c _sthe queue length that represents user s, so c _s=c _s+ 1.Because task has different weights, each user's processing power is also different, and actual load is obviously different from queue length, and we use w _irepresent the load of user i reality, so the load w of user s _sthere is w _s=w _sthe ability of+task weight/user j, continues above-mentioned steps, until distribute, finishes.

3. simulation analysis

(1) emulation setting and explanation

In this emulation, we simulate following mobile social networking scene.For each time slot i, n user arbitrarily walks the square area at 100 meters * 100 meters, produces at random user's position (X _i, Y _i), { 0 < X < 100,0 < Y < 100,1≤i≤n}.The user of take in the communication range that radius is r, can contact with the other side, exchange essential information is also carried out redistributing of task.At identical time slot, each user distributes the task (each time slot has n task) of oneself separately.For each task, distributor is considered to the center of circle of communication range.We calculate he and other all users' distance, find and distribute candidate.In candidate, random choose is wherein two, and relatively queue length, assigns the task to the shorter user of task queue length.More practically, task weight, user's computing power, energy consumptions etc. take into account, and we have also carried out simulating, verifying to various situation in experiment.Design parameter arranges in Table 1.

(2) impact of communication range

In first experiment, we compare the impact of communication radius r on our method performance.Here, we select 100 users and 30 time slot samplings, d=2.Do not consider task weight and user capability.As shown in Figure 2, when communication range is set to 5 meters, our method and Random assignment do not have much difference.Main cause is, communication radius r value is too little causes can not find enough candidates.Along with the increase of communication radius r, can find that our method is obviously better than Random assignment.Let us is looked at Fig. 3, and when communication radius is set to 10 meters, Random assignment makes maximum queue length reach 38 and the shortest queue length is 17, and the difference between minimum and maximum queue length is 21.By contrast, this method has narrowed down to 4 by difference effectively, and each user's queue length is also more or less the same with mean value.In addition, as shown in Figure 4, during r=30, our method is also obviously better than simple randomization distribution.

For further proof, we have compared the performance of institute's extracting method of the present invention and optimal situation (directly task being given to the minimum user of communication range internal burden).As shown in Figure 5, the curve during due to d=2 almost overlaps with optimal situation, and we not without reason believe that our method performance is outstanding.In our method, there will not be and have indivedual nodes too busy, and there is idle situation in some node.

(3) impact of task weight

In second experiment, our impact of evaluation tasks weight on this method performance.In this experiment, each task has it self weight.As everyone knows, different tasks has different difficulty, and we give the weight that difficult task is larger.So last selecteed user does not just add queue length, but given task weight.If we have an opportunity to understand everyone present load, the efficiency that our effectively raising task is distributed.As Fig. 6, shown in 7,8, task weight is respectively that { 1-500} is uniformly distributed for 1-5,1-50.Under identical experimental configuration condition, the difference between maximum load and minimum load is significantly dwindled.We find, this patent institute extracting method is almost best.In sum, our method has realized better performance, rather than simple randomization distributes.

(4) impact of user's computing power

In social networks, user's computing power is different.Go-getter should do more work, and we consider this point, quantizes the ability of different user, and capabilities setting is become to 1-5 rank.In other words,, if a people's ability is 4, processing speed can be fast 4 times of level 1 than basic horizontal.In experiment, for clearly showing the impact of user's computing power, our fixed communication radius, does not consider task weight.When people is selected while carrying out Processing tasks, its queue length c adds 1, and load w increases by 1/ user j computing power.When everyone computing power is known, we can be directly according to load allocating task, and as shown in Figure 9, solid line is the method that we propose, and dotted line represents Random assignment method.Obviously, our method is more far better than Random assignment method, because user load is slightly unsteady centered by equalization point comparatively ideally, illustrates that the counterbalance effect of task load is fine.

(5) impact that energy consumes

Execute the task and will expend mobile device energy.Dividing timing, we must consider user's energy level.Obviously, we can not give the user assignment task that there is no enough energy certainly, although this user may least-loaded.In other words, first we will keep the equilibrium of user's energy level, and then check Subscriber Queue length.Select the high user of energy in 2 candidates during our allocating task, 1 point can decline at every turn.As shown in figure 10, our method (solid line) is also better than Random assignment method (dotted line) greatly.

(6) impact that d parameter arranges

We have completed after all working above, the impact that the d parameter that begins one's study arranges.Due to the impact of randomness, if we increase selection quantity d in theory, performance can be better.On this basis, we arrange respectively d for { 2,3,4}, comparative result is shown in Figure 11.We are not difficult to find, although the performance outline of d=4 is better than d=2, this also means that we need to understand the information of more users, increased and expended again in the middle of invisible, so we think in mobile social networking, d=2 is enough.

The setting of table 1 simulation parameter

Claims (3)

1. the task load balance method for mobile social networking, it is characterized in that, user in mobile social networking is when allocating task, need select least-loaded person in two or more users for each, and than the i.e. i.e. each method of selecting arbitrarily a user from user of least-loaded person and Random assignment in each user of selection of optimum allocation, not only need not know global information but also can reach well the load balance between user, said method comprising the steps of:

1), in each time slot, each user judges that other users are whether in communication range;

2) all users in communication range, select arbitrarily several users;

3) more select user's load, found out load the lowest in user;

4) assign the task to load the lowest;

5) if there is no user in communication range, wait for the distribution of next time slot;

6) in a time slot, all users distribute task, start the distribution of next time slot;

7) finish.

2. a kind of task load balance method for mobile social networking according to claim 1, is characterized in that: in step 1) in,

N user carries out random walk in limited area, the communication radius that wherein r is each mobile subscriber, when two nodes are all in mutual communication range, task can be reallocated, judge whether in communication range according to being:

(X _i-X _j) ²+(Y _i-Y _j) ²＜r ²

(X wherein _i, Y _i) and (X _j, Y _j) be respectively the position coordinates of user i and user j, if above formula is set up, show that user i and user j, in communication range, can carry out the reallocation of message exchange and task, if above formula is false, mark user i can not communicate by letter with user j.

3. a kind of task load balance method for mobile social networking according to claim 1, is characterized in that:

Load balance between user, refers to being uniformly distributed of task, reduces to greatest extent the difference of queue length between each user, can be provided by following formula:

$\min \underset{i\&Element;U}{\mathrm{\&Sigma;}}|Q_i-E[Q_i\left]\right|$

Wherein, U={1,2 ... n} is user's set, and i ∈ U represents that user i belongs to set U, Q _ithe queue length that represents user i, E[] be the mean value of stochastic variable, min is the minimum value of getting expression formula.

Images