The content of the invention
The technical problem to be solved is, there is provided a kind of wireless cloud computing system performance prediction method, can be with
By modular unit predict cloud computing system performance, for the feasibility of system, parameter setting the problems such as reference is provided.
For solving above-mentioned technical problem, the present invention provides a kind of wireless cloud computing system performance prediction method, including as follows
Step:
(1) confirm that workflow, to the mapping relations of AP, calculates the average task amount of each AP internal task, calculate
The average amount of transmission of data;
(2) average completion time of each AP internal task is calculated, the average transmission time of transmission data between AP is calculated;
(3) by the task link layering of workflow, certain the task link place number of plies is defined as first to be appointed to an office business and reaches the path
The most short jumping figure of rear-guard task;
(4) for the main congestion of user in system is when the communication process, between the jam situation of AP links and link
The transmission time of total data is relevant, it is assumed that have k bar AP link congestions, and the order of congestion AP links is always transmitted according to data in link
Time arranges from high to low, calculates each congestion AP links data queued bag number, obtains in this case graph model;
(5) the congestion data bag number calculated by step (4), calculates the communication queue time delay of every AP link, calculates each
Longest path of the graph model first to be appointed to an office business to tail task, the graph model for finally selecting longest path value maximum is used as the user in system
Main congestion is in communication process and system reaches the forecast model under maximum processing capability, obtains makespan;
(6) for the main congestion of user in system is when the calculating process, it is assumed that in wireless cloud computing system, group is more
The number of users that can be accommodated is R, and factor alpha is incorporated herein, its unit for/people, then represent Queued tasks or packet with R* α
Number, it is assumed that only one of which AP congestion, and the number of tasks of congestion is R* α, other AP congestions number of tasks are 1;
(7) congestion number of tasks before AP is obtained by step (6), calculates the computing resource queuing delay of each AP, supplement figure
The computing resource stand-by period of task node in model, the graph model for finally selecting longest path value maximum is used as the use in system
Householder wants congestion in calculating process and system reaches forecast model under maximum processing capability, obtains makespan;
(8) makespan obtained by step (5) and step (7), takes the minimum of a value of makespan* λ and R, and wherein λ is
Arrival rate, the makespan that the minimum of a value is obtained as system maximum galleryful input step (5) and step (7) continue
Iteration, until stable, then obtains the makespan under different arrival rate.
Preferably, in step (1), confirm that workflow, to the mapping relations of AP, calculates the average of each AP internal task and appoints
Business amount, the data volume of all tasks that will be in AP outside virtual task be added after divided by task quantity;Calculate transmission data between AP
Average amount, that is, divided by task number of links after the task link transmitted data amount sum included in seeking certain AP link.
Preferably, the average completion time of AP internal tasks in step (2), is calculated, i.e., inside the AP for being obtained with step (1)
Processing speed of the average task amount divided by AP;The average transmission time of the transmission data between AP is calculated, i.e., is obtained with step (1)
Between AP, average data bag data amount is divided by the total bandwidth transmitted between corresponding A P.
Preferably, in step (4), it is assumed that be the main congestion of user in system in the situation of communication process, AP links
Between jam situation and link, the transmission time of total data is relevant, first assumes there is an AP link congestion, and the congestion path is sum
According to the most long path of transmission time, then graph model in this case can be obtained;It is next assumed that have two AP link congestions, together
Reason is gone down always, it is assumed that have k bar AP link congestions, the order of congestion AP links according to data overall transmission time in link by height to
Low arrangement, obtains graph model in this case, for every kind of graph model is by the delamination in step (3), obtains each link
Middle communication queue number-of-packet, it is assumed that have Ni paths, can obtain Ni graph model.
Preferably, in step (5), it is assumed that be the main congestion of user in system in the situation of communication process, by congestion
Number-of-packet, calculates the communication queue time delay of every AP link, and AP link communications queuing delay is assigned to correspondence in graph model
The communication resource between task waits time delay, calculates each graph model first to be appointed to an office business to the longest path of tail task, finally selects longest path
The maximum graph model of footpath value as in system the main congestion of user is in communication process and system is reached under maximum processing capability
Forecast model, obtain makespan.
Preferably, in step (6), it is assumed that be the main congestion of user in system in the situation of calculating process, only one of which
Before AP congestions, and the AP, the number of tasks of congestion is R × α, and other AP congestions number of tasks are 1, then if Nap AP, can obtain
To Nap graph model.
Preferably, in step (7), it is assumed that be the main congestion of user in system in the situation of calculating process, by step
(6) the congestion number of tasks of computing resource is waited before obtaining AP, calculates the computing resource queuing delay of each AP, by the calculating before AP
Queuing delay is assigned to the computing resource in graph model before each task and waits time delay, selects longest path value maximum by last
Graph model as the main congestion of user in system is in calculating process and system reaches the prediction mould under maximum processing capability
Type, obtains makespan.
Preferably, in step (8), the makespan that above-mentioned steps (5) and step (7) are obtained takes makespan × λ and R
Minimum of a value, wherein λ are arrival rate, are obtained the minimum of a value as system maximum galleryful input step (5) and step (7)
Makespan, continues iteration, until stable, then obtains the makespan under different arrival rate.
Beneficial effects of the present invention are:Performance prediction for huge cloud computing system provides a kind of simple prediction block
Frame, modular method can be by the jam situation of access point, the congestion feelings of channel in the estimated cloud computing system of initial parameters
Condition and the average process time in systems of each user.For the feasibility of system, the parameter setting of system the problems such as provide
With reference to substantially increasing the efficiency of cloud computing system design work.
Specific embodiment
As shown in Fig. 1,3,4,5 and 6, a kind of wireless cloud computing system performance prediction method specifically includes following steps:
(1) confirm that workflow, to the mapping relations of AP, calculates the average task amount of each AP internal task, it is assumed that APnReflect
Being mapped to for task has p, respectively t1,t2,...,tp, their task amount is I1,I2,…,Ip, then average task amount such as formula
(1) shown in, it is assumed that APnTo APmPacket amount number have q, the transmitted data amount included in each packet is respectively
Transferdata1,Transferdata2,…,Transferdataq, then calculate the average amount of transmission packet between AP
As shown in formula (2).
(2)APnShown in the average completion time of internal task such as formula (3), wherein, PapFor the disposal ability of an AP,
Assume there is N inside an APvmIndividual VM (virtual machine), then the disposal ability of each VM is Pvm=Pap/Nvm。APnArrive
APmTransmission data average transmission time such as formula (4) shown in, wherein NchFor APnTo APmLink in radio channels, B0
For the bandwidth of these channels.
(3) by the task link layering of workflow, certain the task link place number of plies is defined as first to be appointed to an office business and reaches the path
The most short jumping figure of rear-guard task, as shown in figure 1, task 1 is set to ground floor to the task link of task 2, task 2 to task 8
Task link be the second layer, task link of the identical layer as same type.
(4) the main congestion of user being assumed to be in system is in the situation of communication process, the jam situation of AP links and link
Between total data transmission time it is relevant, first assume there is an AP link congestion, the congestion path is that total data transmission time is most long
Path, the number of congestion data bag is R × α, then can obtain graph model in this case.It is next assumed that there is two AP links
Congestion, is the most long AP links with vice-minister of total data transmission time, then can equally obtain graph model in this case, in the same manner
Go down always, it is assumed that have k bar AP link congestions, the order of congestion AP links according to data overall transmission time in link from high to low
Arrangement, calculates each congestion AP links data queued bag number, it is assumed that task number of links is respectively N below1,N2,…,Nk.It is all this
A little task links have Q types (type is illustrated in step 3), and the number-of-packet for obtaining each AP links congestion is (R
×α/Q)×N1, (R × α/Q) × N2 ..., (R × α/Q) × NkIn this case graph model is obtained then.Finally, it is assumed that have NlBar road
Footpath, can obtain NlIndividual graph model.
(5) the communication queue time delay of every AP link by congestion data bag number, is calculated, by AP link communications queuing delay
Be assigned to the communication resource in graph model between correspondence task and wait time delay, each graph model first to be appointed to an office business is calculated to the longest path of tail task
Footpath, the graph model for finally selecting longest path value maximum as in system the main congestion of user is in communication process and system reaches
Forecast model under maximum processing capability, obtains makespan.
(6) the main congestion of user being assumed to be in system in the situation of calculating process, by dividing to a large amount of simulation results
The heart, it is assumed that before the congestion of only one of which AP, and the AP number of tasks of congestion be R × α, other AP congestions number of tasks be 1, then if
NapIndividual AP, then can obtain NapIndividual graph model.
(7) obtain before AP, waiting the congestion number of tasks of computing resource, the computing resource for calculating each AP to queue up by step 6
The queuing delay that calculates before AP is assigned in graph model the wait time delay of the computing resource before each task, by last by time delay
The graph model for selecting longest path value maximum as in system the main congestion of user is in calculating process and system reaches maximum
Forecast model under reason ability, obtains makespan.
(8) bring the makespan obtained in above procedure into formula (5) and formula (6) respectively, solve, obtain difference and arrive
In makespan up under rate, wherein formula (1), f represents step 3,4,5 or step 6,7 process, only during R for become
Measure, use RuserRepresent.
Makespan=f (Ruser×α) (5)
Ruser=min { λ × makespan, R } (6)
Below for above-mentioned steps for 10 points of workflows example, 10 points of workflows as shown in Fig. 2 this be one figure
Model.Design parameter, it is Pap=100000MIPS that we arrange the processing speed of an AP, and each AP has 10 VM, i.e. Nvm
=10, then processing speed P of single VMvm=Pap/Nvm=10000MIPS, in data transmission procedure, bandwidth B0=10^
6bit/sm, Nch=1.
As shown in table 1, task amount unit is MI (million instruction set) to the corresponding task amount I of its task number.
Table 1
Task number |
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
Task amount |
0 |
1339000 |
1383000 |
1336000 |
1378000 |
1037000 |
1059000 |
1088000 |
1755000 |
0 |
The data volume transmitted between its task is as shown in table 2.Data volume unit is bit.Wherein in table, data volume 1 represents task
Between have before and after drive relation, but transmission data volume be 0, it is 1 to take transmission quantity here very little.
Table 2
Task number |
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
1 |
0 |
1 |
1 |
1 |
1 |
0 |
0 |
0 |
0 |
0 |
2 |
0 |
0 |
0 |
0 |
0 |
8334624 |
0 |
8873456 |
0 |
0 |
3 |
0 |
0 |
0 |
0 |
0 |
0 |
8517569 |
0 |
0 |
0 |
4 |
0 |
0 |
0 |
0 |
0 |
7511132 |
8340796 |
8324756 |
0 |
0 |
5 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
7853276 |
0 |
0 |
6 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
8326168 |
0 |
7 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
8861129 |
0 |
8 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
8315294 |
0 |
9 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
1 |
10 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
In fig. 2, each task node has corresponding weights, and it is data in table 1 divided by P to be worthvm, during each side correspondence weights are table 2
Data are divided by B0。
Task allocation result is as shown in table 3, wherein task 1, and task 10 is virtual task, and which is not involved in average amount
Calculate.
Table 3
AP labels |
1 |
2 |
3 |
4 |
Comprising task |
1,5,6,9 |
3,7 |
2 |
4,8,10 |
The task link that AP links are included is as shown in table 4:
Table 4
Bring formula into and obtain the average task amount of each AP internal task as shown in table 5, unit is MI.Transmission data between AP
The average amount of bag is as shown in table 6:
Table 5
AP labels |
1 |
2 |
3 |
4 |
Task amount |
1390000 |
1221000 |
1339000 |
1212000 |
Table 6
AP labels |
1 |
2 |
3 |
4 |
1 |
/ |
/ |
/ |
7853276 |
2 |
8861129 |
/ |
/ |
/ |
3 |
8334624 |
/ |
/ |
8873456 |
4 |
7913213 |
8340796 |
/ |
/ |
Bring formula into and obtain the average handling time of each AP internal task as shown in table 7, unit is MI.Number is transmitted between AP
Average transmission time according to bag is as shown in table 8:
Table 7
AP labels |
1 |
2 |
3 |
4 |
Task amount |
13.9 |
12.21 |
13.39 |
12.12 |
Table 8
Task link layering result is as shown in table 9:
Table 9
Task number |
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
1 |
/ |
1 |
1 |
1 |
1 |
/ |
/ |
/ |
/ |
/ |
2 |
/ |
/ |
/ |
/ |
/ |
2 |
/ |
2 |
/ |
/ |
3 |
/ |
/ |
/ |
/ |
/ |
/ |
2 |
/ |
/ |
/ |
4 |
/ |
/ |
/ |
/ |
/ |
2 |
2 |
2 |
/ |
/ |
5 |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
2 |
/ |
/ |
6 |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
3 |
/ |
7 |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
3 |
/ |
8 |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
3 |
/ |
9 |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
4 |
10 |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
/ |
By in table, during the time delay that corresponding communication queue and calculating are queued up adds Fig. 1,6 corresponding graph models are obtained, led to
Cross be calculated 6 kinds of graph models longest path it is as shown in table 10:
Table 10
Artwork model |
1 |
2 |
3 |
4 |
5 |
6 |
Longest path value |
2020.911 |
2905.377 |
2905.377 |
2905.377 |
2905.377 |
2905.377 |
Then in order, select second graph model as the main congestion of user in system in the prediction of communication process
Graph model.
In the same manner, for user's congestion is when the calculating process.For longest path such as 11 institute of table for obtaining 4 kinds of graph models
Show.The first graph model is then chosen as the main congestion of user in system in the forecast model of calculating process, figure is finally obtained
Model is as shown in Figure 4.
Table 11
Artwork model |
1 |
2 |
3 |
4 |
Longest path value |
6026.155 |
5335.322 |
3156.195 |
5296.029 |
Solution formula (1) and formula (2) obtain the performance curve of wireless cloud computing system, and the performance refers to average each user
Process time in systems, as shown in Figure 5.
Compute-intensive applications can be decomposed into a series of set of tasks, and the logical relation formed between a kind of task is claimed
For workflow, by the mapping of workflow to AP (Wireless Access Point), the communication resource and computing resource can be made
Distributed, completed the process of workflow.When substantial amounts of user reaches, that is, there is substantial amounts of workflow in cloud computing system
Middle parallel processing, for each workflow of each user, could set up a graph model, i.e., workflow each
Before business before link, computation delay and communication delay can be produced, afterwards by seeking first to be appointed to an office business (task 1 in Fig. 2) in graph model
To the longest path of tail task (task 10 in Fig. 2), process time makespan for obtaining the user that can be approximate, this figure
Model is that the prediction of the performance to whole system provides the foundation.Secondly, it is assumed that can at most accommodate in wireless cloud computing system
Number of users be R.When arrival rate of customers reaches certain value, the number of users that system is accommodated will remain in R, and now system
Reach " saturation state " of maximum processing capability, the system processing time of each user will increase no longer as arrival rate increases
Greatly, then systematic function under this kind of " saturation state " needs to be predicted.Again, the product of time delay is analyzed under " saturation state "
It is raw, during task is pending or data are to be sent, computing resource or the communication resource may be waited, is at this moment just generated
Time delay, by the mapping relations of task to AP, can obtain in AP average data between the process time of average each task and AP
Transmission time, then now the prediction of the stand-by period in graph model can be changed into the prediction of queuing number of tasks before AP and
Between AP, the prediction of link data queued bag number, is incorporated herein factor alpha, and its unit is individual/people, then represent Queued tasks with R × α
Or the number of packet, in this process α take 1.Finally, the main congestion of user model being divided in system is in communication process
Considered in two kinds of situations of calculating process respectively with the main congestion of user in system, for the main congestion of user in system is in logical
The situation of letter process, it is believed that no computing resource queuing delay, the biography of total data between the jam situation and AP links between AP links
The defeated time is relevant, first assumes there is an AP link congestion, and congestion AP links are the most long path of total data transmission time, then may be used
To obtain graph model in this case.It is next assumed that there is two paths congestions, congestion path be total data transmission time it is most long and
The path of vice-minister, goes down in the same manner always, it is assumed that have NiBar AP links, each AP links congestion data bag number is by task link point
The mode of Layer assignment is determined.N can be obtainediIndividual graph model, the graph model for finally selecting longest path value maximum is used as being
Forecast model of the main congestion of user in system under communication process " saturation state ".For the main congestion of user in system in
The situation of calculating process, it is believed that task does not have communication queue time delay, it is assumed that only one of which AP congestion, and the number of tasks of congestion is R
× α, other AP congestions number of tasks are 1, then if Nap AP, can obtain Nap graph model, finally select longest path
The maximum graph model of value as the main congestion of user in system in calculating process and system reach it is pre- under maximum processing capability
Model is surveyed, the makespan obtained by above procedure can be obtained under each arrival rate by the solution for seeking one group of function
makespan。
Although the present invention is illustrated with regard to preferred embodiment and has been described, it is understood by those skilled in the art that
Without departing from scope defined by the claims of the present invention, variations and modifications can be carried out to the present invention.