CN108549719A - A kind of adaptive cache method based on cluster in mobile edge calculations network - Google Patents
A kind of adaptive cache method based on cluster in mobile edge calculations network Download PDFInfo
- Publication number
- CN108549719A CN108549719A CN201810369050.5A CN201810369050A CN108549719A CN 108549719 A CN108549719 A CN 108549719A CN 201810369050 A CN201810369050 A CN 201810369050A CN 108549719 A CN108549719 A CN 108549719A
- Authority
- CN
- China
- Prior art keywords
- user
- request
- content
- hypercube
- popularity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/56—Provisioning of proxy services
- H04L67/568—Storing data temporarily at an intermediate stage, e.g. caching
Abstract
The invention discloses a kind of adaptive cache methods based on cluster in mobile edge calculations network, include the following steps:1) user's request is sent on MEC servers;2) feedback user is asked;3) feature vector is generated;4) the prediction popularity of user's request content is obtained;5) prediction popularity judges whether to need to cache user's request content to MEC servers, and is operated accordingly according to judging result;6) MEC servers optimize feature space by on-line study;7) when next user asks to reach MEC servers, step 2) is repeated to step 6), it completes until completing all user's requests, wherein, after completing U user's request, then all request contents of cache user are carried out with the prediction of popularity, and update ordered information tables Q, this method can be by carrying out Popularity prediction to the content in mobile edge calculations network, and determines that cache contents, the hit rate of caching are higher with this.
Description
Technical field
The present invention relates to a kind of adaptive cache methods in mobile edge calculations network, and in particular to a kind of mobile edge
Calculate the adaptive cache method based on cluster in network.
Background technology
Demand due to mobile cellular network to large scale multimedia business is continuously increased, and the flow load of network shows
Index increased situation, huge challenge is proposed to network capacity and backhaul link.According to the nearest a report of Cisco, move
Dynamic data traffic increased 4000 times in past 10 years.5,500,000,000 are up to the year two thousand twenty mobile subscriber's quantity, accounts for global people
The 70% of mouth.Widely used, the mobile rapid expansion of coverage area of mobile device and rapidly riseing for mobile content demand, will
So that the amplification of mobile subscriber reaches as many as twice of population in the world amplification in the coming five years.To the year two thousand twenty, Intelligent mobile equipment
With connection quantity it is expected that international mobile equipment will be accounted for and couples the 72% of total amount, 36% compared to 2015 increases rapidly.
Smart machine expect the year two thousand twenty by will produce 98% mobile data flow, mobile video flow will be in application program among these
In possess highest amplification, requirement of the user for high-quality video can bring great load pressure to mobile network.Simultaneously
The sign that global mobile data flow speedup does not slow down.To the year two thousand twenty, global mobile data flow is up to monthly 30.6EB,
And it was only 3.7EB in 2015;The year two thousand twenty whole world mobile data flow is up to 366.8EB, and is only 44.2EB in 2015.
When the year two thousand twenty, before the annual mobile data flow of 366.8EB is equivalent to 10 years, i.e., the mobile flow in the whole world caused by 2010
120 times.For next generation mobile communication network, how to cope with such challenge becomes the common target of academia.
Huge burden can be brought to Internet resources by a large amount of initial data of network transmission.In some cases, exist
Its source nearby handles data and will only have valuable data is sent to remote data center then more efficiently by network.It moves
Dynamic edge calculations (Mobile Edge Computing, MEC) are exactly an effective method.Mobile edge calculations (MEC) are bases
In the framework of 5G evolution, and by a kind of technology of Mobile Access Network and Internet service depth integration.On the one hand MEC can improve
User experience saves bandwidth resources, on the other hand by the way that computing capability is sunk to mobile fringe node.It is carried by deployment
The edge device (such as gateway, base station and end user device) of computing function furthers at a distance from user, by cloud computing and cloud
After storage zooms in network edge, a service environment for having high-performance, low latency and high bandwidth can be createed, accelerates net
The distribution and download of every content, service and alllication, allow consumer to enjoy higher quality web experience in network.
Mobile network has switched to the existing pattern based on data transmission from traditional pattern based on voice communication,
And as can be seen that the data transmission of non real-time nature accounts for a very big part from the report of many network flows.In addition to this, big
In the mobile Internet data of amount, a small part prevalence data occupies great flow load, this comes to backhaul Netowrk tape
Greatly burden.And with the development of big data analysis technology, the prediction of hot content becomes possibility.Telecom operators can be with
By collecting day regular data come rational management and control mobile network.And many information from social networks etc. can also help us
Understand the daily mobility of user, social characteristic, preference characteristics etc., this data analysis based on social networks etc. can be helped
Us are helped to classify user crowd.Cloud computing and the development of edge calculations are but also data analysis in real time is answered
With.The price reduction of fast storing medium is but also the cost of arrangement caching substantially reduces.So being cached in advance at the network idle moment
Hot content to mobile network's edge device is possibly realized, and this method is not only avoided that the remote transmission of content, in reduction
Improves the experience of user the time required to holding transmission, and also cross-domain flow rate can be transferred in the internal network of operator, reduces operation
The expense of quotient.
Existing research is proved to cache most popular content (popularity highest) under the conditions of single edge device can be with
Obtain highest cache hit rate, but the popularity of content and uncertain and popularity can be over time under normal conditions
It changes, popularity is difficult to determine.
Invention content
It is an object of the invention to overcome the above-mentioned prior art, base in a kind of mobile edge calculations network is provided
In the adaptive cache method of cluster, this method considers the problems of content size, by moving in edge calculations network
Hold and carry out Popularity prediction, and determines that cache contents, the hit rate of caching are higher with this.
In order to achieve the above objectives, the adaptive cache method based on cluster in mobile edge calculations network of the present invention
Include the following steps:
1) base station connects MEC servers, when user accesses internet content by mobile network, then asks to send out by user
It send to MEC servers;
2) when the content of user's request is on MEC servers, then the service user on MEC servers is utilized, when
The content of user's request then obtains the content of user's request not when on MEC servers by core net;
3) the history access record daily record for obtaining user's request content, gives birth to according to the history access information of user's request content
At feature vector;
4) in the maps feature vectors to feature space for generating step 3), the prediction for obtaining user's request content is popular
Degree;
5) judge whether to need to cache user's request content to MEC to take according to the prediction popularity of user's request content
It is engaged on device, when needing to cache user's request content to MEC servers, then judges whether user's request content meets caching
Condition then caches user's request content to ordered information on MEC servers when user's request content meets caching condition
In table Q;
6) MEC servers optimize feature space by on-line study;
7) when next user asks to reach MEC servers, step 2) is repeated to step 6), is completed until completing all
Until user asks, wherein after completing U user's request, then carried out to all request contents of cache user popular
The prediction of degree, and ordered information tables Q is updated, complete the adaptive cache based on cluster in mobile edge calculations network.
By the kth of user time access request be integrated into one group of ordered information tables Req (k)=<C(k),S(k),V(k),X
(k),I(k),T(k),P(k)>, wherein C (k) is the content of user's request, and S (k) is the content size of user's request, and V (k) is
Feature vector, X (k) ∈ { 0,1 } indicate whether user's request content hits, wherein 1 indicates hit, and 0 indicates miss, I
(k) ∈ { 0,1 } indicates whether MEC servers cache the content in the user asks, and T (k) is that user asks the time occurred
Stamp, P (k) are the popularity that the user asks corresponding content.
The concrete operations of step 2) are:
When user asks Req (k) to reach MEC servers, when being cached with user request content C (k) in MEC servers
When, then enable I (k)=0, X (k)=1;When there is no cache user request content C (k) in MEC servers, then X (k)=0 is enabled;
Step 3) specifically includes following steps:
3a) when first time obtaining user request content C (k) by the base station, then by initialising subscriber request content C
(k) history access record empties, and after base station obtains user's request content, is then accessed in the history of user request content C (k)
The time stamp T (k) of current accessed is recorded in record;
All time stamp datas of user's request content 3b) are obtained, and by all timestamps in chronological order with TPFor
Interval is counted and is recorded, D group data before obtaining, and feature vector V (k)=(V of 1 × D is built according to preceding D groups data1,V2,
L,VD), wherein VDFor [T (k)-DTP,T(k)-(D-1)·TP] quantity of timestamp in section;
Feature vector is not present in initialization feature space in feature space after initialization, feature space is 1 grade super vertical at this time
Cube Ω1,1,1, with increasing for user's request, the feature vector in feature space gradually increases, according to all feature vectors
Distribution situation, feature space can be split into multiple unduplicated hypercubes, wherein each feature vector finally belongs to ID and is
The n grade hypercubes Ω of (i, j)i,j,n, hypercube Ωi,j,nIn there are variable En(Ωi,j,n) and PM(Ωi,j,n), wherein En
(Ωi,j,n) it is hypercube Ωi,j,nThe number of middle feature vector, PM(Ωi,j,n) it is hypercube Ωi,j,nMiddle feature vector
Popularity summation, hypercube Ωi,j,nPopularity be
Step 4) specifically includes following steps:
4a) set i=1, j=1, n=1;
4b) check hypercube Ωi,j,nIn whether there is cluster centre;
4c) as hypercube Ωi,j,nIn when cluster centre is not present, then feature vector V (k) ∈ Ωi,j,n;Work as hypercube
Body Ωi,j,nIn there are when cluster centre, then according to Yp,q,n+1=min | | V (k)-αp,q,n+1||,αp,q,n+1∈Ωi,j,n, find out with
Feature vector V (k) is apart from nearest cluster centre αp,q,n+1, obtain cluster centre αp,q,n+1Corresponding hypercube Ωp,q,n+1;
4d) enable i ← p, j ← q, n ← n+1, repeat step 3b) to step 3c), until finding super vertical belonging to feature vector
Until cube, i.e. V (k) ∈ Ωi,j,n, the prediction popularity P (k) of Req (k) institute request content C (k) is:
P (k)=PSpace(Ωi,j,n)·S(k)·R(C(k))
Feature vector V (k) the ∈ Ω of Req (k)i,j,n, R (C (k)) be user request content C (k) Popularity prediction miss
The default value of poor correction factor, R (C (k)) is 1.
Step 5) specifically includes following steps:
5a) there are the partial information that an ordered information tables Q is used for memory buffers content, the orderly letters in MEC servers
Breath table Q is:
Wherein, CQ(i) the user's request content cached for MEC servers, PQ(i) it is corresponding prediction popularity, SQ
(i) it is corresponding user's request content size, Q1,Q2,L,Qi,L,QMBy the prediction big minispread of popularity, element Q1Indicate prediction
Popularity is minimum,Wherein, SrestIndicate the residual caching capacity of MEC servers, Srest>=0, Smax
Indicate the total capacity of MEC servers;
5b) work as SrestUser request content C (k) is then stored in MEC servers, and enables I (k)=1 by >=S (k), simultaneously
Update SrestValue, i.e. Srest←Srest- S (k) then utilizes user that Req (k) is asked to rebuild ordered information tables<CQ(1),L,CQ
(M)>, and end step 5);
Work as Srest<S (k) then enables Ssum=0, Psum=0, i=1, wherein SsumIndicate the size of multiple replaceable contents
With PsumIt indicates the prediction popularity weighted sum of multiple replaceable contents, then goes to step 5c);
5c) enable Ssum←Ssum+SQ(i),SQ(i) ∈ Q (i), Psum←λ·Psum+PQ(i),PQ(i) ∈ Q (i), i ← i+1;
Wherein, λ is cost coefficient;
5d) go to step 5b), until Ssum>=S (k) and Psum≤ P (k) then enables I (k)=1, Srest←Srest-Ssum, and
Req (k) is asked to rebuild ordered information tables using user<CQ(1),L,CQ(M)>, when going to step 5b) number be more than or equal to it is pre-
If value, and still not satisfy Ssum≥S(k)、PsumWhen≤P (k), then user request content C (k) is unsatisfactory for caching condition, MEC clothes
Be engaged in device not cache user request content C (k), and enables I (k)=0.
Step 6) specifically includes following steps:
6a) hypercube Ωi,j,nThe feature vector accommodated is limited, hypercube Ωi,j,nSegmentation threshold beWherein, l1And l2For threshold coefficient.
6b) user asks the feature vector V (k) of req (k) to enter hypercube Ωi,j,n, and pass through time TcAfter observe
Its true popularity PLAfter (C (k)), wherein PL(C (k))=Nc(C (k)) S (k), Nc(C (k)) is in time TcMiddle content
The number of C (k) requests, S (k) are the content size of request, then update hypercube Ωi,j,nVariable En(Ωi,j,n)=En
(Ωi,j,n)+1, PM(Ωi,j,n)=PM(Ωi,j,n)+PL(C (k)), while updating Popularity prediction correction factor
6c) work as En(Ωi,j,nWhen) >=ε, then hypercube Ωi,j,nUsing the method for K mean cluster into line splitting, hypercube
Body Ωi,j,nCorresponding feature vector belongs to the new hypercube obtained after division, and new hypercube inherits former hypercube
Variable En(Ωi,j,n) and PM(Ωi,j,n)。
The invention has the advantages that:
The adaptive cache method based on cluster is when specific operation in mobile edge calculations network of the present invention, when
When user asks to reach MEC servers, if the content caching of user's request directly invokes MEC servers in MEC servers
Service user, if MEC server no users request content when, MEC servers will obtain content from core net,
The prediction for carrying out user's request content popularity according to the history access information of user's request content simultaneously, is asked further according to user
The prediction popularity of content judges whether to need to cache user's request content to MEC servers, and according to the knot of judgement
Fruit is operated accordingly, and the hit rate of caching is higher, simple, convenient, and through experiment, the present invention can be in MEC servers
On to greatest extent utilize cache contents, it is possible to reduce backhaul link pressure avoids content from transmitting at a distance, reduce content transmission
Time improves user experience, reduces operator's expense, increases operator's profitable offering newly.
Description of the drawings
Fig. 1 is the illustraton of model of system;
Fig. 2 is the flow chart of Popularity prediction;
Fig. 3 a are in buffer memory capacity Smax=400, the lower cache hit in λ=1 is than the change curve with α;
Fig. 3 b are in buffer memory capacity Smax=400, the lower caching cost in λ=1 is than the change curve with α;
Fig. 4 a are in α=0.64, the lower cache hit in λ=1 than the change curve with spatial cache;
Fig. 4 b are in α=0.64, the lower caching cost in λ=1 than the change curve with spatial cache;
Fig. 5 a are in buffer memory capacity Smax=400, α=0.64, lower change curve of the cache hit than T at any time in λ=1;
Fig. 5 b are in buffer memory capacity Smax=400, α=0.64, lower caching change curve of the cost than T at any time in λ=1;
Fig. 6 a are in caching cost Smax=400, the lower cache hit in α=0.64 is than the change curve with λ;
Fig. 6 b are in caching cost Smax=400, the lower caching cost in α=0.64 is than the change curve with λ.
Specific implementation mode
The present invention is described in further detail below in conjunction with the accompanying drawings:
With reference to figure 1, in mobile edge calculations network of the present invention the adaptive cache method based on cluster include with
Lower step:
1) base station connects MEC servers, when user accesses internet content by mobile network, then asks to send out by user
It send to MEC servers;
2) when the content of user's request is on MEC servers, then the service user on MEC servers is utilized, when
The content of user's request then obtains the content of user's request not when on MEC servers by core net;
3) the history access record daily record for obtaining user's request content, gives birth to according to the history access information of user's request content
At feature vector;
4) in the maps feature vectors to feature space for generating step 3), the prediction for obtaining user's request content is popular
Degree;
5) judge whether to need to cache user's request content to MEC to take according to the prediction popularity of user's request content
It is engaged on device, when needing to cache user's request content to MEC servers, then judges whether user's request content meets caching
Condition then caches user's request content to ordered information on MEC servers when user's request content meets caching condition
In table Q;
6) MEC servers optimize feature space by on-line study;
7) when next user asks to reach MEC servers, step 2) is repeated to step 6), is completed until completing all
Until user asks, wherein after completing U user's request, then carried out to all request contents of cache user popular
The prediction of degree, and ordered information tables Q is updated, complete the adaptive cache based on cluster in mobile edge calculations network.
By the kth of user time access request be integrated into one group of ordered information tables Req (k)=<C(k),S(k),V(k),X
(k),I(k),T(k),P(k)>, wherein C (k) is the content of user's request, and S (k) is the content size of user's request, and V (k) is
Feature vector, X (k) ∈ { 0,1 } indicate whether user's request content hits, wherein 1 indicates hit, and 0 indicates miss, I
(k) ∈ { 0,1 } indicates whether MEC servers cache the content in the user asks, and T (k) is that user asks the time occurred
Stamp, P (k) are the popularity that the user asks corresponding content.
The concrete operations of step 2) are:
When user asks Req (k) to reach MEC servers, when being cached with user request content C (k) in MEC servers
When, then enable I (k)=0, X (k)=1;When there is no cache user request content C (k) in MEC servers, then X (k)=0 is enabled;
Step 3) specifically includes following steps:
3a) when first time obtaining user request content C (k) by the base station, then by initialising subscriber request content C
(k) history access record empties, and after base station obtains user's request content, is then accessed in the history of user request content C (k)
The time stamp T (k) of current accessed is recorded in record;
All time stamp datas of user's request content 3b) are obtained, and by all timestamps in chronological order with TPFor
Interval is counted and is recorded, D group data before obtaining, and feature vector V (k)=(V of 1 × D is built according to preceding D groups data1,V2,
L,VD), wherein VDFor [T (k)-DTP,T(k)-(D-1)·TP] quantity of timestamp in section;
Feature vector is not present in initialization feature space in feature space after initialization, feature space is 1 grade super vertical at this time
Cube Ω1,1,1, with increasing for user's request, the feature vector in feature space gradually increases, according to all feature vectors
Distribution situation, feature space can be split into multiple unduplicated hypercubes, wherein each feature vector finally belongs to ID and is
The n grade hypercubes Ω of (i, j)i,j,n, hypercube Ωi,j,nIn there are variable En(Ωi,j,n) and PM(Ωi,j,n), wherein En
(Ωi,j,n) it is hypercube Ωi,j,nThe number of middle feature vector, PM(Ωi,j,n) it is hypercube Ωi,j,nMiddle feature vector
Popularity summation, hypercube Ωi,j,nPopularity be
Step 4) specifically includes following steps:
4a) set i=1, j=1, n=1;
4b) check hypercube Ωi,j,nIn whether there is cluster centre;
4c) as hypercube Ωi,j,nIn when cluster centre is not present, then feature vector V (k) ∈ Ωi,j,n;Work as hypercube
Body Ωi,j,nIn there are when cluster centre, then according to Yp,q,n+1=min | | V (k)-αp,q,n+1||,αp,q,n+1∈Ωi,j,n, find out with
Feature vector V (k) is apart from nearest cluster centre αp,q,n+1, obtain cluster centre αp,q,n+1Corresponding hypercube Ωp,q,n+1;
4d) enable i ← p, j ← q, n ← n+1, repeat step 3b) to step 3c), until finding super vertical belonging to feature vector
Until cube, i.e. V (k) ∈ Ωi,j,n, the prediction popularity P (k) of Req (k) institute request content C (k) is:
P (k)=PSpace(Ωi,j,n)×S(k)·R(C(k))
Feature vector V (k) the ∈ Ω of Req (k)i,j,n, R (C (k)) is the Popularity prediction error of user request content C (k)
The default value of correction factor, R (C (k)) is 1.
Step 5) specifically includes following steps:
5a) there are the partial information that an ordered information tables Q is used for memory buffers content, the orderly letters in MEC servers
Breath table Q is:
Wherein, CQ(i) the user's request content cached for MEC servers, PQ(i) it is corresponding prediction popularity, SQ
(i) it is corresponding user's request content size, Q1,Q2,L,Qi,L,QMBy the prediction big minispread of popularity, element Q1Indicate prediction
Popularity is minimum,Wherein, SrestIndicate the residual caching capacity of MEC servers, Srest>=0, Smax
Indicate the total capacity of MEC servers;
5b) work as SrestUser request content C (k) is then stored in MEC servers, and enables I (k)=1 by >=S (k), simultaneously
Update SrestValue, i.e. Srest←Srest- S (k) then utilizes user that Req (k) is asked to rebuild ordered information tables<CQ(1),L,CQ
(M)>, and end step 5);
Work as Srest<S (k) then enables Ssum=0, Psum=0, i=1, wherein SsumIndicate the size of multiple replaceable contents
With PsumIt indicates the prediction popularity weighted sum of multiple replaceable contents, then goes to step 5c);
5c) enable Ssum←Ssum+SQ(i),SQ(i) ∈ Q (i), Psum←λ·Psum+PQ(i),PQ(i) ∈ Q (i), i ← i+1;
Wherein, λ is cost coefficient;
5d) go to step 5b), until Ssum>=S (k) and Psum≤ P (k) then enables I (k)=1, Srest←Srest-Ssum, and
Req (k) is asked to rebuild ordered information tables using user<CQ(1),L,CQ(M)>, when going to step 5b) number be more than or equal to it is pre-
If value, and still not satisfy Ssum≥S(k)、PsumWhen≤P (k), then user request content C (k) is unsatisfactory for caching condition, MEC clothes
Be engaged in device not cache user request content C (k), and enables I (k)=0.
Step 6) specifically includes following steps:
6a) hypercube Ωi,j,nThe feature vector accommodated is limited, hypercube Ωi,j,nSegmentation threshold beWherein, l1And l2For threshold coefficient.
6b) user asks the feature vector V (k) of req (k) to enter hypercube Ωi,j,n, and pass through time TcAfter observe
Its true popularity PLAfter (C (k)), wherein PL(C (k))=Nc(C (k)) S (k), Nc(C (k)) is in time TcMiddle content
The number of C (k) requests, S (k) are the content size of request, then update hypercube Ωi,j,nVariable En(Ωi,j,n)=En
(Ωi,j,n)+1, PM(Ωi,j,n)=PM(Ωi,j,n)+PL(C (k)), while updating Popularity prediction correction factor
6c) work as En(Ωi,j,nWhen) >=ε, then hypercube Ωi,j,nUsing the method for K mean cluster into line splitting, hypercube
Body Ωi,j,nCorresponding feature vector belongs to the new hypercube obtained after division, and new hypercube inherits former hypercube
Variable En(Ωi,j,n) and PM(Ωi,j,n)。
K mean cluster purpose is to find cluster centre, and the method and optimization algorithm of K mean cluster are all suitable for this strategy,
Basic process is:
A) k initial point is generated at random as cluster centre;
B) data in data set are assigned to according to the distance apart from barycenter in each cluster;
C) data in each cluster are averaged, and using the average value as new cluster centre;
D) step a) to step c) is repeated averagely to divide according still further to cluster centre until all clusters no longer change
Hypercube.
Replication experiment
Using emulation data set come verification algorithm performance:Data set include 500 contents, size be randomly assigned for 1,
2,5,10,20 }, access frequency obeys Zipf-like distributions, and parameter alpha changes with data set, and total request number of times is N=
100000 times.
In all emulation data sets, for simulation popularity variation, extreme case is taken to carry out testing algorithm right pop degree
Prediction case;Data set is divided into two sections, data set popularity of content in time T=5 inverts;In leading portion
Between (T<5) the popularity highest for the content that ID is 1, in back segment time (T>5) popularity for the content that ID is 1 is minimum,
Performance standard has:
Cache cost ratio
Cache hit ratio
As shown in Fig. 2, if the popularity of content is unchanged, LFU is optimal close to theory.But in practice, content stream
Row degree can constantly change at any time, and LFU caches pollution problem due to existing, and hit rate is caused to reduce.
With reference to figure 3a to Fig. 5 b, under different parameters, performance caching cost ratio (CCR) and caching life that the present invention is obtained
The middle performance for being better than LFU, LRU, FIFO than (SCHR).For example, in spatial cache 400, cache hit ratio of the present invention relative to
LFU, LRU, FIFO improve 15%, 57% and 68% respectively;
As shown in Fig. 6 a and Fig. 6 b, the size of λ does not influence the performance of LFU and FIFO, and caching cost ratio (CCR) is fixed
Value;With the increase of λ, the CCR of SACA and LFU are continuously decreased;As λ=1.4, cache hit of the invention ratio reaches highest.
Claims (8)
1. a kind of adaptive cache method based on cluster in mobile edge calculations network, which is characterized in that include the following steps:
1) base station connects MEC servers, when user accesses internet content by mobile network, is then sent to user's request
On MEC servers;
2) when the content of user's request is on MEC servers, then the service user on MEC servers is utilized, user is worked as
The content of request then obtains the content of user's request not when on MEC servers by core net;
3) the history access record daily record for obtaining user's request content generates special according to the history access information of user's request content
Sign vector;
4) in the maps feature vectors to feature space for generating step 3), the prediction popularity of user's request content is obtained;
5) judge whether to need to cache user's request content to MEC servers according to the prediction popularity of user's request content
On, when needing to cache user's request content to MEC servers, then judge whether user's request content meets cache bar
Part then caches user's request content to ordered information tables on MEC servers when user's request content meets caching condition
In Q;
6) MEC servers optimize feature space by on-line study;
7) when next user asks to reach MEC servers, step 2) is repeated to step 6), is completed until completing all users
Until request, wherein after completing U user's request, then carry out popularity to all request contents of cache user
Prediction, and ordered information tables Q is updated, complete the adaptive cache based on cluster in mobile edge calculations network.
2. the adaptive cache method based on cluster in mobile edge calculations network according to claim 1, feature exist
In, by the kth of user time access request be integrated into one group of ordered information tables Req (k)=<C(k),S(k),V(k),X(k),I
(k),T(k),P(k)>, wherein C (k) is the content of user's request, and S (k) is the content size of user's request, and V (k) is characterized
Vector, X (k) ∈ { 0,1 } indicate whether user's request content hits, wherein 1 indicates hit, and 0 indicates miss, I (k) ∈
{ 0,1 } indicate whether MEC servers cache the content in the user asks, and T (k) is that user asks the timestamp occurred, P
(k) popularity of corresponding content is asked for the user.
3. the adaptive cache method based on cluster in mobile edge calculations network according to claim 2, feature exist
In the concrete operations of step 2) are:
When user asks Req (k) to reach MEC servers, when being cached with user request content C (k) in MEC servers,
Then enable I (k)=0, X (k)=1;When there is no cache user request content C (k) in MEC servers, then X (k)=0 is enabled.
4. the adaptive cache method based on cluster in mobile edge calculations network according to claim 3, feature exist
In step 3) specifically includes following steps:
3a) when first time obtaining user request content C (k) by the base station, then by initialising subscriber request content C's (k)
History access record empties, after base station obtains user's request content, then in the history access record of user request content C (k)
The time stamp T (k) of middle record current accessed;
All time stamp datas of user's request content 3b) are obtained, and by all timestamps in chronological order with TPFor be spaced into
Row statistics and record, D group data before obtaining, and according to feature vector V (k)=(V of 1 × D of preceding D groups data structure1,V2,L,VD),
Wherein, VDFor [T (k)-DTP,T(k)-(D-1)·TP] quantity of timestamp in section.
5. the adaptive cache method based on cluster in mobile edge calculations network according to claim 4, feature exist
In feature vector is not present in initialization feature space in feature space after initialization, feature space is 1 grade of hypercube at this time
Ω1,1,1, with increasing for user's request, the feature vector in feature space gradually increases, according to the distribution of all feature vectors
Situation, feature space can be split into multiple unduplicated hypercubes, wherein it is (i, j) that each feature vector, which finally belongs to ID,
N grade hypercubes Ωi,j,n, hypercube Ωi,j,nIn there are variable En(Ωi,j,n) and PM(Ωi,j,n), wherein En
(Ωi,j,n) it is hypercube Ωi,j,nThe number of middle feature vector, PM(Ωi,j,n) it is hypercube Ωi,j,nMiddle feature vector
Popularity summation, hypercube Ωi,j,nPopularity be
6. the adaptive cache method based on cluster in mobile edge calculations network according to claim 5, feature exist
In step 4) specifically includes following steps:
4a) set i=1, j=1, n=1;
4b) check hypercube Ωi,j,nIn whether there is cluster centre;
4c) as hypercube Ωi,j,nIn when cluster centre is not present, then feature vector V (k) ∈ Ωi,j,n;Work as hypercube
Ωi,j,nIn there are when cluster centre, then according to Yp,q,n+1=min | | V (k)-αp,q,n+1||,αp,q,n+1∈Ωi,j,n, find out and spy
Vector V (k) is levied apart from nearest cluster centre αp,q,n+1, obtain cluster centre αp,q,n+1Corresponding hypercube Ωp,q,n+1;
4d) enable i ← p, j ← q, n ← n+1, repeat step 3b) to step 3c), until finding the hypercube belonging to feature vector
Until, i.e. V (k) ∈ Ωi,j,n, the prediction popularity P (k) of Req (k) institute request content C (k) is:
P (k)=PSpace(Ωi,j,n)·S(k)·R(C(k))
Feature vector V (k) the ∈ Ω of Req (k)i,j,n, R (C (k)) is the Popularity prediction error correction of user request content C (k)
The default value of coefficient, R (C (k)) is 1.
7. the adaptive cache method based on cluster in mobile edge calculations network according to claim 6, feature exist
In step 5) specifically includes following steps:
5a) there are the partial information that an ordered information tables Q is used for memory buffers content, ordered information tables Q in MEC servers
For:
Wherein, CQ(i) the user's request content cached for MEC servers, PQ(i) it is corresponding prediction popularity, SQ(i) it is
Corresponding user's request content size, Q1,Q2,L,Qi,L,QMBy the prediction big minispread of popularity, element Q1Indicate prediction popularity
Minimum,Wherein, SrestIndicate the residual caching capacity of MEC servers, Srest>=0, SmaxIndicate MEC
The total capacity of server;
5b) work as SrestUser request content C (k) is then stored in MEC servers, and enables I (k)=1 by >=S (k), updates simultaneously
SrestValue, i.e. Srest←Srest- S (k) then utilizes user that Req (k) is asked to rebuild ordered information tables<CQ(1),L,CQ(M)>,
And end step 5);
Work as Srest<S (k) then enables Ssum=0, Psum=0, i=1, wherein SsumIndicate the size and P of multiple replaceable contentssum
It indicates the prediction popularity weighted sum of multiple replaceable contents, then goes to step 5c);
5c) enable Ssum←Ssum+SQ(i),SQ(i) ∈ Q (i), Psum←λ·Psum+PQ(i),PQ(i) ∈ Q (i), i ← i+1;Wherein, λ
For cost coefficient;
5d) go to step 5b), until Ssum>=S (k) and Psum≤ P (k) then enables I (k)=1, Srest←Srest-Ssum, and utilize
User asks Req (k) to rebuild ordered information tables<CQ(1),L,CQ(M)>, when going to step 5b) number be more than or equal to preset value,
And still not satisfy Ssum≥S(k)、PsumWhen≤P (k), then user request content C (k) is unsatisfactory for caching condition, MEC servers
Not cache user request content C (k), and enable I (k)=0.
8. the adaptive cache method based on cluster in mobile edge calculations network according to claim 7, feature exist
In step 6) specifically includes following steps:
6a) hypercube Ωi,j,nThe feature vector accommodated is limited, hypercube Ωi,j,nSegmentation threshold be
Wherein, l1And l2For threshold coefficient;
6b) user asks the feature vector V (k) of req (k) to enter hypercube Ωi,j,n, and pass through time TcAfter observe that it is true
Real popularity PLAfter (C (k)), wherein PL(C (k))=Nc(C (k)) S (k), Nc(C (k)) is in time TcMiddle content C (k)
The number of request, S (k) are the content size of request, then update hypercube Ωi,j,nVariable En(Ωi,j,n)=En(Ωi,j,n)
+ 1, PM(Ωi,j,n)=PM(Ωi,j,n)+PL(C (k)), while updating Popularity prediction correction factor
6c) work as En(Ωi,j,nWhen) >=e, then hypercube Ωi,j,nUsing the method for K mean cluster into line splitting, hypercube
Ωi,j,nCorresponding feature vector belongs to the new hypercube obtained after division, and new hypercube inherits former hypercube
Variable En(Ωi,j,n) and PM(Ωi,j,n)。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810369050.5A CN108549719A (en) | 2018-04-23 | 2018-04-23 | A kind of adaptive cache method based on cluster in mobile edge calculations network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810369050.5A CN108549719A (en) | 2018-04-23 | 2018-04-23 | A kind of adaptive cache method based on cluster in mobile edge calculations network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108549719A true CN108549719A (en) | 2018-09-18 |
Family
ID=63512053
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810369050.5A Pending CN108549719A (en) | 2018-04-23 | 2018-04-23 | A kind of adaptive cache method based on cluster in mobile edge calculations network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108549719A (en) |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109561470A (en) * | 2018-11-26 | 2019-04-02 | 厦门大学 | A kind of collaboration method of network code and edge cache in chain type communication network |
CN109600780A (en) * | 2019-02-18 | 2019-04-09 | 南京邮电大学 | A kind of duplicate of the document caching method based on base station sub-clustering |
CN109788305A (en) * | 2018-12-10 | 2019-05-21 | 北京爱奇艺科技有限公司 | A kind of data cached method for refreshing and device |
CN109885088A (en) * | 2019-03-12 | 2019-06-14 | 西安交通大学 | Unmanned plane during flying track optimizing method in edge calculations network based on machine learning |
CN109995865A (en) * | 2019-04-03 | 2019-07-09 | 广东工业大学 | A kind of request responding method and device of the data information based on mobile edge calculations |
CN110247793A (en) * | 2019-05-29 | 2019-09-17 | 暨南大学 | A kind of application department arranging method in mobile edge cloud |
CN110377533A (en) * | 2019-07-10 | 2019-10-25 | 北京邮电大学 | It is a kind of for cloud computing and the caching replacement method and device of edge calculations |
US10623782B1 (en) * | 2018-11-02 | 2020-04-14 | University-Industry Cooperation Group Of Kyung-Hee University | Deep learning based caching system and method for self-driving car in multi-access edge computing |
CN111491331A (en) * | 2020-04-14 | 2020-08-04 | 重庆邮电大学 | Network perception self-adaptive caching method based on transfer learning in fog computing network |
CN111629052A (en) * | 2020-05-26 | 2020-09-04 | 中国联合网络通信集团有限公司 | Content caching method, node, equipment and storage medium based on MEC |
CN112468597A (en) * | 2020-12-11 | 2021-03-09 | 深圳市知小兵科技有限公司 | Data center resource allocation method and device based on artificial intelligence |
CN112740642A (en) * | 2020-08-31 | 2021-04-30 | 华为技术有限公司 | Communication method and multi-access edge computing server |
CN112733060A (en) * | 2021-01-13 | 2021-04-30 | 中南大学 | Cache replacement method and device based on session clustering prediction and computer equipment |
CN112822727A (en) * | 2021-01-29 | 2021-05-18 | 重庆邮电大学 | Self-adaptive edge content caching method based on mobility and popularity perception |
CN112953746A (en) * | 2019-12-11 | 2021-06-11 | 国家广播电视总局广播电视科学研究院 | Edge network node optimization method and device and edge node equipment |
CN113504986A (en) * | 2021-06-30 | 2021-10-15 | 广州大学 | Cache-based edge computing system unloading method, device, equipment and medium |
CN113542330A (en) * | 2020-04-21 | 2021-10-22 | 中移(上海)信息通信科技有限公司 | Method and system for acquiring mobile edge calculation data |
CN113687960A (en) * | 2021-08-12 | 2021-11-23 | 华东师范大学 | Edge calculation intelligent caching method based on deep reinforcement learning |
CN115002138A (en) * | 2022-03-04 | 2022-09-02 | 昆明理工大学 | Edge caching method based on terminal unmanned aerial vehicle cruise video data |
CN115051868A (en) * | 2022-06-22 | 2022-09-13 | 杭州电子科技大学 | Cache pollution attack detection and defense method under edge calculation |
CN115866051A (en) * | 2022-11-15 | 2023-03-28 | 重庆邮电大学 | Edge caching method based on content popularity |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105049326A (en) * | 2015-06-19 | 2015-11-11 | 清华大学深圳研究生院 | Social content caching method in edge network area |
CN106101223A (en) * | 2016-06-12 | 2016-11-09 | 北京邮电大学 | A kind of caching method mated with node rank based on content popularit |
CN106453495A (en) * | 2016-08-31 | 2017-02-22 | 北京邮电大学 | Information centric networking caching method based on content popularity prediction |
CN106550039A (en) * | 2016-11-08 | 2017-03-29 | 华中科技大学 | A kind of cross-layer cooperation content buffering method and system based on SD RAN |
CN106603646A (en) * | 2016-12-07 | 2017-04-26 | 北京邮电大学 | Information centric networking caching method based on user interests and preferences |
CN107171961A (en) * | 2017-04-28 | 2017-09-15 | 中国人民解放军信息工程大学 | Caching method and its device based on content popularit |
CN107592656A (en) * | 2017-08-17 | 2018-01-16 | 东南大学 | Caching method based on base station cluster |
CN107909108A (en) * | 2017-11-15 | 2018-04-13 | 东南大学 | Edge cache system and method based on content popularit prediction |
-
2018
- 2018-04-23 CN CN201810369050.5A patent/CN108549719A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105049326A (en) * | 2015-06-19 | 2015-11-11 | 清华大学深圳研究生院 | Social content caching method in edge network area |
CN106101223A (en) * | 2016-06-12 | 2016-11-09 | 北京邮电大学 | A kind of caching method mated with node rank based on content popularit |
CN106453495A (en) * | 2016-08-31 | 2017-02-22 | 北京邮电大学 | Information centric networking caching method based on content popularity prediction |
CN106550039A (en) * | 2016-11-08 | 2017-03-29 | 华中科技大学 | A kind of cross-layer cooperation content buffering method and system based on SD RAN |
CN106603646A (en) * | 2016-12-07 | 2017-04-26 | 北京邮电大学 | Information centric networking caching method based on user interests and preferences |
CN107171961A (en) * | 2017-04-28 | 2017-09-15 | 中国人民解放军信息工程大学 | Caching method and its device based on content popularit |
CN107592656A (en) * | 2017-08-17 | 2018-01-16 | 东南大学 | Caching method based on base station cluster |
CN107909108A (en) * | 2017-11-15 | 2018-04-13 | 东南大学 | Edge cache system and method based on content popularit prediction |
Non-Patent Citations (1)
Title |
---|
王莹等: "移动边缘网络缓存技术", 《北京邮电大学学报》 * |
Cited By (34)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10623782B1 (en) * | 2018-11-02 | 2020-04-14 | University-Industry Cooperation Group Of Kyung-Hee University | Deep learning based caching system and method for self-driving car in multi-access edge computing |
CN109561470A (en) * | 2018-11-26 | 2019-04-02 | 厦门大学 | A kind of collaboration method of network code and edge cache in chain type communication network |
CN109788305A (en) * | 2018-12-10 | 2019-05-21 | 北京爱奇艺科技有限公司 | A kind of data cached method for refreshing and device |
CN109788305B (en) * | 2018-12-10 | 2021-03-02 | 北京爱奇艺科技有限公司 | Cache data refreshing method and device |
CN109600780A (en) * | 2019-02-18 | 2019-04-09 | 南京邮电大学 | A kind of duplicate of the document caching method based on base station sub-clustering |
CN109600780B (en) * | 2019-02-18 | 2021-10-29 | 南京邮电大学 | File copy caching method based on base station clustering |
CN109885088A (en) * | 2019-03-12 | 2019-06-14 | 西安交通大学 | Unmanned plane during flying track optimizing method in edge calculations network based on machine learning |
CN109885088B (en) * | 2019-03-12 | 2021-01-19 | 西安交通大学 | Unmanned aerial vehicle flight trajectory optimization method based on machine learning in edge computing network |
CN109995865A (en) * | 2019-04-03 | 2019-07-09 | 广东工业大学 | A kind of request responding method and device of the data information based on mobile edge calculations |
CN110247793A (en) * | 2019-05-29 | 2019-09-17 | 暨南大学 | A kind of application department arranging method in mobile edge cloud |
CN110377533A (en) * | 2019-07-10 | 2019-10-25 | 北京邮电大学 | It is a kind of for cloud computing and the caching replacement method and device of edge calculations |
CN110377533B (en) * | 2019-07-10 | 2021-03-19 | 北京邮电大学 | Cache replacement method and device for cloud computing and edge computing |
CN112953746A (en) * | 2019-12-11 | 2021-06-11 | 国家广播电视总局广播电视科学研究院 | Edge network node optimization method and device and edge node equipment |
CN111491331B (en) * | 2020-04-14 | 2022-04-15 | 重庆邮电大学 | Network perception self-adaptive caching method based on transfer learning in fog computing network |
CN111491331A (en) * | 2020-04-14 | 2020-08-04 | 重庆邮电大学 | Network perception self-adaptive caching method based on transfer learning in fog computing network |
CN113542330B (en) * | 2020-04-21 | 2023-10-27 | 中移(上海)信息通信科技有限公司 | Mobile edge calculation data acquisition method and system |
CN113542330A (en) * | 2020-04-21 | 2021-10-22 | 中移(上海)信息通信科技有限公司 | Method and system for acquiring mobile edge calculation data |
CN111629052A (en) * | 2020-05-26 | 2020-09-04 | 中国联合网络通信集团有限公司 | Content caching method, node, equipment and storage medium based on MEC |
CN111629052B (en) * | 2020-05-26 | 2021-12-07 | 中国联合网络通信集团有限公司 | Content caching method, node, equipment and storage medium based on MEC |
CN112740642A (en) * | 2020-08-31 | 2021-04-30 | 华为技术有限公司 | Communication method and multi-access edge computing server |
CN112468597B (en) * | 2020-12-11 | 2021-05-28 | 深圳市知小兵科技有限公司 | Data center resource allocation method and device based on artificial intelligence |
CN112468597A (en) * | 2020-12-11 | 2021-03-09 | 深圳市知小兵科技有限公司 | Data center resource allocation method and device based on artificial intelligence |
CN112733060A (en) * | 2021-01-13 | 2021-04-30 | 中南大学 | Cache replacement method and device based on session clustering prediction and computer equipment |
CN112733060B (en) * | 2021-01-13 | 2023-12-01 | 中南大学 | Cache replacement method and device based on session cluster prediction and computer equipment |
CN112822727A (en) * | 2021-01-29 | 2021-05-18 | 重庆邮电大学 | Self-adaptive edge content caching method based on mobility and popularity perception |
CN112822727B (en) * | 2021-01-29 | 2022-07-01 | 重庆邮电大学 | Self-adaptive edge content caching method based on mobility and popularity perception |
CN113504986A (en) * | 2021-06-30 | 2021-10-15 | 广州大学 | Cache-based edge computing system unloading method, device, equipment and medium |
CN113687960B (en) * | 2021-08-12 | 2023-09-29 | 华东师范大学 | Edge computing intelligent caching method based on deep reinforcement learning |
CN113687960A (en) * | 2021-08-12 | 2021-11-23 | 华东师范大学 | Edge calculation intelligent caching method based on deep reinforcement learning |
CN115002138B (en) * | 2022-03-04 | 2023-07-04 | 昆明理工大学 | Edge caching method based on cruise video data of terminal unmanned aerial vehicle |
CN115002138A (en) * | 2022-03-04 | 2022-09-02 | 昆明理工大学 | Edge caching method based on terminal unmanned aerial vehicle cruise video data |
CN115051868A (en) * | 2022-06-22 | 2022-09-13 | 杭州电子科技大学 | Cache pollution attack detection and defense method under edge calculation |
CN115051868B (en) * | 2022-06-22 | 2024-03-22 | 杭州电子科技大学 | Cache pollution attack detection and defense method under edge calculation |
CN115866051A (en) * | 2022-11-15 | 2023-03-28 | 重庆邮电大学 | Edge caching method based on content popularity |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108549719A (en) | A kind of adaptive cache method based on cluster in mobile edge calculations network | |
CN103812849B (en) | A kind of local cache update method, system, client and server | |
CN104426979B (en) | A kind of distributed caching scheduling System and method for based on social networks | |
CN107483630A (en) | A kind of construction method for combining content distribution mechanism with CP based on the ISP of edge cache | |
CN107592656B (en) | Caching method based on base station clustering | |
Sinky et al. | Responsive content-centric delivery in large urban communication networks: A LinkNYC use-case | |
CN108881445B (en) | Cooperation caching method based on Gono game in fog calculation | |
CN108093435B (en) | Cellular downlink network energy efficiency optimization system and method based on cached popular content | |
Li et al. | Deep reinforcement learning for cooperative edge caching in future mobile networks | |
CN110062357A (en) | A kind of D2D ancillary equipment caching system and caching method based on intensified learning | |
CN107105043B (en) | Content-centric network caching method based on software defined network | |
CN106603646B (en) | A kind of information centre's network-caching method based on user interest preference | |
CN108366089A (en) | A kind of CCN caching methods based on content popularit and pitch point importance | |
CN108541025B (en) | Wireless heterogeneous network-oriented base station and D2D common caching method | |
CN103905538A (en) | Neighbor cooperation cache replacement method in content center network | |
CN110418367A (en) | A kind of 5G forward pass mixture of networks edge cache low time delay method | |
CN108900617A (en) | A kind of three layers of cooperative caching method of mist wireless access network | |
CN105939385A (en) | Request frequency based real-time data replacement method in NDN cache | |
CN106899692A (en) | A kind of content center network node data buffer replacing method and device | |
CN107734482A (en) | The content distribution method unloaded based on D2D and business | |
CN108156218A (en) | The double-deck distributed caching method of mist wireless access network | |
CN106973088B (en) | A kind of buffering updating method and network of the joint LRU and LFU based on shift in position | |
Li et al. | Edge caching for D2D enabled hierarchical wireless networks with deep reinforcement learning | |
CN112887943A (en) | Cache resource allocation method and system based on centrality | |
CN108156249B (en) | Network cache updating method based on approximate Markov chain |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180918 |
|
RJ01 | Rejection of invention patent application after publication |