CN104852831B - A kind of Forecasting Methodology of hierarchical network RTT - Google Patents
A kind of Forecasting Methodology of hierarchical network RTT Download PDFInfo
- Publication number
- CN104852831B CN104852831B CN201510293578.5A CN201510293578A CN104852831B CN 104852831 B CN104852831 B CN 104852831B CN 201510293578 A CN201510293578 A CN 201510293578A CN 104852831 B CN104852831 B CN 104852831B
- Authority
- CN
- China
- Prior art keywords
- network
- rtt
- network rtt
- forecasting methodology
- distance
- 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.)
- Active
Links
Abstract
The Forecasting Methodology of a kind of hierarchical network RTT, wherein according to real network daily record, utilizes information entropy theory to catch the factor affecting network RTT；Determine measurement network RTT fluctuation situation index；For in-plant main frame pair, build the network attribute database in regional extent by the operation that measures actively, record the time delay of each, select the most close network path to recombinate the time delay of each jumping in network；For remote main frame pair, one machine learning model of dynamic training can be carried out by analyzing the relation of network delay and physical distance, thus estimates network delay.The method may be implemented in without in advance and efficiently estimating network round-trip time delay in the case of main frame contacts on network.
Description
Technical field
The present invention relates to computer network, the Forecasting Methodology of a kind of hierarchical network RTT.
Background technology
Network RTT i.e. user issues a request to receive the time interval of server service, is current online real time service
(real-time video is live, extensive game on line) is weighed the important indicator of Consumer's Experience.The size of RTT decides terminal use
Persistently use the length of this service, directly affect the income of service provider.Look-ahead network RTT thus select optimum
Server to service for user, can at utmost promote the experience of terminal use, thus have vast potential for future development.
In order to predict network RTT, existing method is broadly divided into: (1) calculates two joints by virtual coordinate space
Distance between point, and estimate that round-trip delay, this method are difficult to capture the dynamic of network delay accordingly；(2) structure is passed through
Build a network map collection, utilize the plyability of network path, reconfigure according to existing path segments delay data, estimate
Calculate the round-trip delay of whole network path, but the establishment of atlas needs to expend substantial amounts of active measurement and passive measurement, very
Difficulty guarantees the autgmentability of system.
Summary of the invention
Present invention is primarily targeted at for the deficiencies in the prior art, in order to realize without main frame in advance and on network
In the case of contact, efficiently estimation network round-trip time delay, proposes the Forecasting Methodology of a kind of hierarchical network RTT.
For achieving the above object, the present invention is by the following technical solutions:
The Forecasting Methodology of a kind of hierarchical network RTT, comprises the following steps:
A1: according to the network log being provided previously by, and by analyzing network delay i.e. network RTT and affecting network RTT's
The mutual information gain of factor X:Select to exceed predetermined on network RTT impact
The factor of degree, wherein, formula Represent the comentropy of its variable Y,
yiIt is one of the n kind value of variable Y, P (Y=yi) represent variable Y value be yiProbability；H (RTT | X) represent that network RTT is relative
Conditional information entropy in influence factor X；
A2: pass through formula:The maximum of network RTT in max (RTT) the expression scheduled time
Value, min (RTT) represents the minimum of a value of network RTT in the scheduled time, and α represents the degree of fluctuation of network RTT in the scheduled time, uses α
Weigh the different user main frame network RTT degree of fluctuation in the given time to server, according to degree of fluctuation with use householder
Machine determines distance cut-point to the corresponding relation between the distance of server, divides in-plant main frame pair according to distance cut-point
With remote main frame pair；
A3: for in-plant main frame pair, by measuring operation on one's own initiative, analyze and record each jumping in network
Time delay, build the network attribute database in predetermined areas；For needing in described predetermined areas to predict network
The path of RTT, by the time delay of each jumping corresponding at described database lookup, and splices, determines respective paths
Network RTT；
A4: for remote main frame pair, by the relation of known network RTT Yu respective physical distance, according to pre-timing
Between the cycle dynamically train decision tree, affect network RTT's determined by intermediate node correspondence step A1 of described decision tree
Factor, the leaf node of described decision tree is corresponding RTT classification, can be assigned to which kind of network RTT by prediction network RTT
In carry out network RTT estimation, the average RTT value of the network RTT value of the final prediction network RTT classification by being distributed.
Further:
In step A2, utilize described mutual information gain as index, by the distance maximum with the mutual information gain of network RTT
It is chosen as described distance cut-point.
In step A3, if some route segment does not has record in database in path, then select and it according to similitude
The record of immediate route segment replaces.
In step A3, use ping instrument to obtain the test point time delay to intermediate router, thus calculate middle any
One time delay jumped:
In step A3, by being updated at predetermined intervals safeguarding about every in the middle of network topology and network
One database jumping time delay.
In step A4, each hour or one decision tree of a few hours training.
In step A4, the described factor affecting network RTT includes that subscriber's main station is to the distance between server, server institute
The Internet Service Provider's type used, and the time of request service.
In step A4, select with analysis mode based on comentropy in decision tree the optimum branching point of correlation distance with
And the position that each factor is in decision tree.
In step A4, K-means algorithm is used to carry out needing the categorizing selection of the network RTT of prediction.
Beneficial effects of the present invention:
The Forecasting Methodology of the hierarchical network RTT of the present invention, selects the cut-point of mixed model by maximizing comentropy, from
And make two modules after division all reach the degree of accuracy of higher prediction network delay, before ensureing system overall accuracy
Put minimizing and measure expense.Present invention achieves hierarchical network RTT Forecasting Methodology, assess for user experience quality, network congestion
Detection, the application such as server selection opens road.
Accompanying drawing explanation
Fig. 1 is the process chart of hierarchical network RTT Forecasting Methodology embodiment of the present invention；
Fig. 2 be an embodiment of the present invention closely in the case of use network RTT predict schematic diagram；
The decision tree schematic diagram that Fig. 3 is trained in the case of remote by an embodiment of the present invention.
Detailed description of the invention
Hereinafter embodiments of the present invention are elaborated.It is emphasized that what the description below was merely exemplary,
Rather than in order to limit the scope of the present invention and application thereof.
Refering to Fig. 1, the hierarchical network RTT Forecasting Methodology of the present invention is according to techniques below thought: first, according to a large amount of actual
Network log, utilizes information entropy theory to catch the factor affecting network RTT；It is then determined that weigh network RTT fluctuation situation index,
Different Forecasting Methodologies is taked according to the fluctuation of RTT under different distance yardstick situation；For in-plant main frame pair, it is contemplated that
The characteristic of network internal, measures operation by some ping of carrying out, traceroute actively etc. and builds the net in regional extent
Network attribute database, the time delay that analytic statistics is each, select the most close network path to recombinate the prolonging of each jumping in network
Time, it is achieved estimate the time delay of whole network path；For remote main frame pair, can by by User IP information MAP to user
Geographical position and operator, and analysis mining such as network delay and the operator of user and main frame between physical distance
Relation, carry out one machine learning model of dynamic training, thus estimate network delay.
In the particular embodiment, can operate by following mode.
Step A1: according to formula:Comentropy H of variable Y
(Y) the uncertain of variable Y, y are weighediIt is one of n kind value that variable Y is desirable, P (Y=yi) represent variable Y value be yi's
Probability.Mutual information gain according to RTT Yu possible factorSelect RTT shadow
Ring the spacing of bigger factor, such as user and server, Internet Service Provider's type that server is used, request
The time etc. of service.Distance between user and server is to be obtained by the disclosed Geo-IP database of inquiry.
Step A2: according toWeigh RTT that user experiences in certain time (such as one
My god) in degree of fluctuation.The maximum of network RTT in max (RTT) the expression scheduled time, in min (RTT) represents the scheduled time
The minimum of a value of network RTT, α represents the degree of fluctuation of network RTT in the scheduled time.We have found that the distance ratio of user and server
Relatively big (>about 120km) when, degree of fluctuation smaller (0≤α<4.5), apart from smaller (about≤120km) time
Wait degree of fluctuation bigger (0≤α < 65).In order to be accurately determined cut-point, utilize mutual information gain as index, by pressing
When dividing according to different distance, the cut-point of the mutual information gain maximum of chosen distance and RTT is as the segmentation taking distinct methods
Point.Such as, all distances are calculated more than 10km (20km ... 110km and 130km ... in the case of), distance increases with the information of RTT
Benefit is less than calculating all mutual information gains more than distance in the case of 120km Yu RTT.Therefore according to 120km as cut-point,
It is respectively adopted different predicting strategies.Therefore under different distance scales, we use different predicting strategies, i.e. for ripple
The dynamic fine-grained processing method of bigger consideration, i.e. below step A3, walk for below the employing that fluctuation situation is less
The processing method of rapid A4.
Step A3: for in-plant situation, our fine granularity ground considers network internal situation, such as network topology and
The time delay of each jumping of intermediate path.Build a network topology database by traceroute, use ping instrument to obtain and survey
Examination node is to the time delay of intermediate router, thus calculates the time delay of middle any one jumpingFinal maintenance one is about each jumping time delay in the middle of network topology and network
Database, is updated in certain time interval.For arbitrarily needing to predict the path of RTT, by database lookup pair
The time delay of each jumping answered, and splice.If in path, some interlude does not has record in database, then by similar
The immediate replacement of Sexual behavior mode.Example as shown in Figure 2, by traceroute, we can obtain the path of src to des and are
Src-> R1-> R2-> R3-> R4-> R5-> des, by test point server periodically (per minute) ping R1, R2, R3, R4,
R5 acquisition server is to the network RTT of intermediate router, as shown in phantom in Figure 2, According to these information, we can calculate
Go out the delay of each jumping in network path, as illustrated in solid line in figure 2, So by accumulation, we finally safeguard one about network topology (such as src-> R1-> R2-> R3-> R4-
> R5-> des) and the middle each jumping of network is (such as) database of time delay, it is updated in certain time interval,
Wherein ping measurement per minute is once, owing to network topology is the most stable, so traceroute updates once for mono-day.For appointing
Meaning needs to predict the path of RTT, by the time delay of each jumping corresponding at database lookup, and splices, such as WhereinWithValue can be obtained by historical record, without historical record, then select recently according to similitude
Host record, such as select and the historical record of src i in same subnetAsValue.Same, as
Really certain router in path does not has record in database, then same procedure utilizes similar record to replace.
Step A4: can be by the factor affecting RTT obtained in A1, each hour (can also be longer or shorter
Suitable period) train a decision tree, it was predicted which kind of RTT RTT can be assigned in.Classification for RTT can use K-
Means algorithm realizes.Decision tree intermediate node corresponds to affect the factor of RTT, for apart from this continuous variable branch point
Select select optimum branch point also based on information entropy theory, leaf node is corresponding RTT classification.The network of final prediction
RTT value is the average RTT value of the RTT classification of prediction.
Such as, by the factor affecting RTT of acquisition in step A1, (user and the spacing of server, server is made
Internet Service Provider's type, request service time), each hour training one decision tree.Due to decision tree
Whole target is classification, so we need to cluster this continuous variable of RTT.For RTT classification we use K-
Means algorithm realizes, and we take K=5 to implement middle K-means algorithm, and according to the otherness between RTT value, K-means calculates
All of RTT is divided into K class (C1, C2, C3, C4, C5) by method.In view of time factor, we each hour by user with
Distance between server, server used service provider type sets up decision tree as input, it was predicted which RTT can be assigned to
In one class RTT.Decision tree intermediate node corresponds to affect the factor of RTT, and (spacing of user and server, server is used
Internet Service Provider's type), for being also based on information entropy theory apart from the selection of this continuous variable branch point, choosing
Selecting the branch point of optimum, leaf node is corresponding RTT classification, as shown in Figure 3.When the historical record utilizing hour is set up
As shown in Figure 3 after decision tree, if one RTT between main frame and server of prediction, such as distance is d (d < d1), ISP
For telecommunications, then can predict that RTT falls C1 according to Fig. 3 left branch；If distance is d (d < d2), ISP is that UNICOM is then according to Fig. 3
Left branch can predict that RTT falls C3；If distance is d (d2 < d < d1), ISP is that UNICOM then can predict according to Fig. 3 left branch
RTT falls C4；If distance is d (d < d1), according to Fig. 3 left branch, ISP then can predict that RTT falls C5 for mobile；If away from
From for d (d > d1), ISP is according to Fig. 3 right branch, telecommunications then can predict that RTT falls C2；If distance is d (d > d1), ISP is
According to Fig. 3 right branch, UNICOM then can predict that RTT falls C3；If distance is d (d > d1), ISP then parts on the right side according to Fig. 3 for mobile
Prop up and can predict that RTT falls C4；The network RTT value finally predicted is the average of the RTT classification (C1, C2, C3, C4, C5) of prediction
RTT value.
The present invention proposes network RTT hybrid predicting framework, selects the cut-point of mixed model by maximizing comentropy, from
And make two modules after division all reach the degree of accuracy of higher prediction network delay, it is achieved without in advance and leading on network
In the case of machine contact, efficiently estimation network round-trip time delay, reduces on the premise of ensureing system overall accuracy and measures expense.
Above content is to combine concrete/further description the most made for the present invention, it is impossible to recognize
Determine the present invention be embodied as be confined to these explanations.For general technical staff of the technical field of the invention,
Without departing from the inventive concept of the premise, these embodiments having described that can also be made some replacements or modification by it,
And these substitute or variant all should be considered as belonging to protection scope of the present invention.
Claims (9)
1. the Forecasting Methodology of a hierarchical network RTT, it is characterised in that comprise the following steps:
A1: according to the network log being provided previously by, and by factor X analyzing network delay i.e. network RTT with affect network RTT
Mutual information gain:Select network RTT impact is exceeded predetermined extent
Factor, wherein, formula Represent the comentropy of its variable Y, yiIt is to become
One of n kind value of amount Y, P (Y=yi) represent variable Y value be yiProbability；H (RTT | X) represent that network RTT is relative to impact
The conditional information entropy of factor X；
A2: pass through formula:The maximum of network RTT in max (RTT) the expression scheduled time,
Min (RTT) represents the minimum of a value of network RTT in the scheduled time, and α represents the degree of fluctuation of network RTT in the scheduled time, weighs with α
Amount different user main frame is to the network RTT degree of fluctuation in the given time of server, according to degree of fluctuation and subscriber's main station
Corresponding relation between the distance of server determine distance cut-point, according to distance cut-point divide in-plant main frame to
Main frame pair at a distance；
A3: for in-plant main frame pair, by measuring operation on one's own initiative, analyze and record prolonging of each jumping in network
Time, build the network attribute database in predetermined areas；For needing in described predetermined areas to predict network RTT's
Path, by the time delay of each jumping corresponding at described database lookup, and splices, determines the network of respective paths
RTT；
A4: for remote main frame pair, by the relation of known network RTT Yu respective physical distance, to schedule week
Phase dynamically trains decision tree, affects the factor of network RTT determined by intermediate node correspondence step A1 of described decision tree,
The leaf node of described decision tree is corresponding RTT classification, can be assigned to which kind of network RTT by prediction network RTT
Carry out network RTT estimation, the average RTT value of the network RTT value of the final prediction network RTT classification by being distributed.
2. the Forecasting Methodology of hierarchical network RTT as claimed in claim 1, it is characterised in that in step A2, utilize described mutual trust
The distance maximum with the mutual information gain of network RTT, as index, is chosen as described distance cut-point by breath gain.
3. the Forecasting Methodology of hierarchical network RTT as claimed in claim 1, it is characterised in that in step A3, if in path certain
A little route segments do not have record in database, then select the record of route segment immediate with it to replace according to similitude.
4. the Forecasting Methodology of hierarchical network RTT as claimed in claim 1, it is characterised in that in step A3, use network measure
Ping instrument obtains the test point time delay to intermediate router, thus calculates the time delay of middle any one jumping:WhereinBe by ping instrument measure from test point S
To router Ri+1Network delay,Be by ping instrument measure from test point S to router RiNetwork
Time delay,It is the router R finally calculatediTo router Ri+1Network delay.
5. the Forecasting Methodology of hierarchical network RTT as claimed in claim 1, it is characterised in that in step A3, by according to predetermined
Time interval be updated safeguarding about the database of each jumping time delay in the middle of network topology and network.
6. the Forecasting Methodology of the hierarchical network RTT as described in any one of claim 1 to 5, it is characterised in that in step A4, often
One hour or one decision tree of a few hours training.
7. the Forecasting Methodology of the hierarchical network RTT as described in any one of claim 1 to 5, it is characterised in that in step A4, institute
State affect the factor of network RTT include subscriber's main station to the distance between server, the network service that server is used provides
Business's type, and the time of request service.
8. the Forecasting Methodology of the hierarchical network RTT as described in any one of claim 1 to 5, it is characterised in that in step A4,
Decision tree select the optimum branching point of correlation distance and each factor decision tree with analysis mode based on comentropy
In position.
9. the Forecasting Methodology of the hierarchical network RTT as described in any one of claim 1 to 5, it is characterised in that in step A4, adopt
Carry out needing the categorizing selection of the network RTT of prediction with K-means algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510293578.5A CN104852831B (en) | 2015-06-01 | 2015-06-01 | A kind of Forecasting Methodology of hierarchical network RTT |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510293578.5A CN104852831B (en) | 2015-06-01 | 2015-06-01 | A kind of Forecasting Methodology of hierarchical network RTT |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104852831A CN104852831A (en) | 2015-08-19 |
| CN104852831B true CN104852831B (en) | 2016-08-24 |
Family
ID=53852198
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510293578.5A Active CN104852831B (en) | 2015-06-01 | 2015-06-01 | A kind of Forecasting Methodology of hierarchical network RTT |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104852831B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106506275B (en) * | 2016-11-09 | 2019-08-20 | 中国科学院计算技术研究所 | A kind of method and device for predicting switching node destination port propagation delay time |
| CN108243234B (en) * | 2016-12-27 | 2020-04-24 | 腾讯科技（深圳）有限公司 | Access scheduling method and device |
| CN109444866A (en) * | 2018-11-15 | 2019-03-08 | 荆门博谦信息科技有限公司 | Distance measuring method and user terminal |
| CN111327688B (en) * | 2020-01-21 | 2022-03-25 | 北京奇艺世纪科技有限公司 | Resource scheduling method, server and readable storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102740331A (en) * | 2012-05-23 | 2012-10-17 | 浙江省通信产业服务有限公司 | Complex wireless communication network operation quality evaluation method based on fuzzy analysis |
| CN104080091A (en) * | 2014-07-22 | 2014-10-01 | 重庆邮电大学 | Family base station frequency spectrum allocation method based on load prediction grouping in layered heterogenous network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9191297B2 (en) * | 2013-03-28 | 2015-11-17 | Hcl Technologies Limited | Providing feedback to media senders over real time transport protocol (RTP) |
-
2015
- 2015-06-01 CN CN201510293578.5A patent/CN104852831B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102740331A (en) * | 2012-05-23 | 2012-10-17 | 浙江省通信产业服务有限公司 | Complex wireless communication network operation quality evaluation method based on fuzzy analysis |
| CN104080091A (en) * | 2014-07-22 | 2014-10-01 | 重庆邮电大学 | Family base station frequency spectrum allocation method based on load prediction grouping in layered heterogenous network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104852831A (en) | 2015-08-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11070453B2 (en) | Providing network traffic endpoint recommendation based on network traffic data analysis | |
| CN109921941B (en) | Network service quality evaluation and optimization method, device, medium and electronic equipment | |
| CN112561191B (en) | Prediction model training method, prediction device, prediction apparatus, prediction program, and program | |
| CN104852831B (en) | A kind of Forecasting Methodology of hierarchical network RTT | |
| JP2011138487A (en) | Method and system for traffic prediction based on space-time relation | |
| US20190239095A1 (en) | Automated intelligent self-organizing network for optimizing network performance | |
| Çelebi et al. | On use of big data for enhancing network coverage analysis | |
| CN103068035A (en) | Wireless network location method, device and system | |
| CN107026750B (en) | User Internet QoE evaluation method and device | |
| JP6053027B2 (en) | Apparatus, program, and method for searching for user who takes similar movement route | |
| KR20130052014A (en) | Terminal quantity estimation device and terminal quantity estimation method | |
| Caceres et al. | Inferring origin–destination trip matrices from aggregate volumes on groups of links: a case study using volumes inferred from mobile phone data | |
| US11290975B2 (en) | Method and apparatus for obtaining emission probability, method and apparatus for obtaining transition probability, and sequence positioning method and apparatus | |
| JP6049186B2 (en) | Apparatus, program, and method for estimating staying place of user having portable terminal | |
| CN109697637A (en) | Object type determines method, apparatus, electronic equipment and computer storage medium | |
| Skorin-Kapov et al. | Energy efficient and quality-driven continuous sensor management for mobile IoT applications | |
| Zygouras et al. | Towards Detection of Faulty Traffic Sensors in Real-Time. | |
| Monserrat et al. | Towards user-centric operation in 5G networks | |
| CN106993296B (en) | Terminal performance evaluation method and device | |
| Jensen et al. | Location intelligence application in digital data activity dimensioning in smart cities | |
| Midoglu et al. | Large scale “speedtest” experimentation in Mobile Broadband Networks | |
| CN105848104B (en) | Flow of personnel state monitoring method and device based on region | |
| Huang et al. | The impact of the Internet on global industry: New evidence of Internet measurement | |
| Khan et al. | Inforank: Information-centric autonomous identification of popular smart vehicles | |
| Rikic et al. | Cellular network bandwidth prediction in consumer applications |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |