CN109729017A - A kind of load-balancing method based on DPI prediction - Google Patents
A kind of load-balancing method based on DPI prediction Download PDFInfo
- Publication number
- CN109729017A CN109729017A CN201910196102.8A CN201910196102A CN109729017A CN 109729017 A CN109729017 A CN 109729017A CN 201910196102 A CN201910196102 A CN 201910196102A CN 109729017 A CN109729017 A CN 109729017A
- Authority
- CN
- China
- Prior art keywords
- flow
- node
- load
- data
- dpi
- 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.)
- Granted
Links
Abstract
The invention belongs to network technology research fields, more particularly to a kind of load-balancing method based on DPI prediction, the following steps are included: input flow rate is divided into a rule stream according to five-tuple and sequence number, analyze agreement by DPI technology and determines which kind of the flow belongs to using data using feature；It determines that flow belongs to the feature of which kind of application, the feature database model correctly applied is selected to be matched；Inquiry of the cryptographic Hash to historical record library content is calculated by traffic flow information, present traffic situation is estimated, the discreet value of uninterrupted is obtained；The present invention obtains subordinate's nodal information using dynamic monitoring system, carries out real-time control to server state.It is pre-allocated by using discreet value and node real time information convection current, alleviates the complexity of load balancing, and calculating is transferred in load balancing prediction, realize the optimization to load balancing entirety.
Description
Technical field
The invention belongs to network technology research fields, and in particular to a kind of load-balancing method based on DPI prediction.
Background technique
The length of the counterbalance effect of load balancing and processing time in network node largely decide entire
The performance of network system.The numerous studies made in terms of load-balancing technique with people, and propose many various
Method.Such as the load balancing based on DNS proxy, this method by replacing response server although it is contemplated that reached
The effect of load balancing, but the state of each server is not accounted for, reply can not be made rapidly by fortuitous event occur, and
The factor of request of data itself is not accounted for；Load balancing based on protocol stack is then the seven-layer structure to analyze data packet
As according to classifying to flow, this method connects equipment in the line, obtains all flows for passing through route, uses
Data link layer of the matching technique to data packet, network layer, transportation level, application layer etc. is parsed, then according to configured
Distribution policy is forwarded flow.This method analyzes and determines data content, so that distribution better effect, but by
Then cascaded structure, data processing time influence whether overall network performance.
Patent document about load-balancing method is more, respectively there is its advantage and disadvantage.Such as application number 201710968835.X
In " a kind of application layer traffic load-balancing method based on DPI " disclosed in patent document, it is connected on client and server
Between, and service traffics are parsed using modes such as light splitting or mirror images, and export dynamic priority as load balancing
One of according to, optimal business service end is selected in conjunction with static priority, but light splitting or mirror image are excessive to the time of service resolution,
And the flow of part in series needs dynamic priority to influence entirety so waiting relationship occur in two modules as parameter
Performance.In addition the content that is parsed using DPI technology is simultaneously indefinite.
Summary of the invention
The purpose of the present invention is to provide a kind of load-balancing methods with more preferable performance.
A kind of load-balancing method based on DPI prediction, comprising the following steps:
(1) input flow rate is divided into a rule stream according to five-tuple and sequence number, by DPI technology analyze agreement and
Determine which kind of the flow belongs to using data using feature；
(2) it determines that flow belongs to the feature of which kind of application, the feature database model correctly applied is selected to be matched；
(3) inquiry of the cryptographic Hash to historical record library content is calculated by traffic flow information, present traffic situation is carried out
It estimates, obtains the discreet value of uninterrupted；
(4) real-time monitoring load-balancing device subordinate's node of dynamic node is used, and node current memory is occupied into feelings
Condition is fed back；
(5) feedback result is analyzed, is pre-allocated using the discreet value that step (3) obtains, selects suitable distribution node；
(6) data flow is sent, by flow actual size, resource name, the information updates such as source address apply feature to it
In library；
(7) step 1) -6 is repeated), until flow, which is all sent, to be terminated.
It is described that input flow rate is divided into a rule stream according to five-tuple and sequence number, by DPI technology analyze agreement with
And determine which kind of the flow belongs to using data using feature, comprising:
Receive the flow for being input to load-balancing device, and according to five-tuple source IP, destination IP, source port, destination port
And agreement flows to distinguish, and the not cocurrent flow of the same port ip is divided using sequence number, uses five-tuple and sequence number as Kazakhstan
Uncommon key value, record is stored among network layer header and transportation level head, and the stream data definition for meeting characteristic model R is
Predictable data e, characteristic model R:
R={ r1, r2..., rn-1, rn}
Wherein, rnFor a certain characteristic element.
The determining flow belongs to the feature of which kind of application, and the feature database model correctly applied is selected to be matched, comprising:
When predictable data e meets the feature for applying A by multimode matching, determine that G (e, A) has association, i.e.,
Wherein aiFor one of the discharge model of application A, that is, meet feature i applies A flow, and e belongs to one using A flow
Kind.
It is described that inquiry of the cryptographic Hash to historical record library content is calculated by traffic flow information, present traffic situation is carried out
It estimates, obtains the discreet value of uninterrupted, comprising:
By using aiAs characteristic model, the key message obtained in data e includes resource name name, resource characteristic
L={ l1, l2..., ln-1, ln, source resource title sname, source resource Ipsip, resource type type, resource size
Size, resource remarks note, resource acquisition time t calculate the inquiry of cryptographic Hash content in the LOG of historical record library, obtain same
Class resource information simultaneously carries out Predict analysis to present traffic situation, obtains the discreet value Y of uninterrupted:
Wherein, sizeiIt is characterized the corresponding resource size of i.
Real-time monitoring load-balancing device subordinate's node using dynamic node, and node current memory is occupied into feelings
Condition is fed back, comprising:
Using dynamic node real-time monitoring load-balancing device subordinate's node j, each node j gather around there are two types of flow handle mould
Formula, one kind be recombination classes, need flow in all data collect one by one recombinate after carry out data processing again；Another kind is forwarding
Class, with the formal layout data flow of assembly line；It is currently interior that real-time monitoring module carries out real-time monitoring acquisition node j to each node
Occupancy situation σ, connection number M are deposited, flow tupe τ, Task scheduling pattern ρ etc. contents feed back to load balance process module.
The analysis feedback result is pre-allocated using the discreet value that step (3) obtains, and selects suitable distribution section
Point, comprising:
It first has to analyze load node state,
If flow tupe τ is recombination classes, load node EMS memory occupation is discreet value Y；
If flow tupe τ is forwarding class, the EMS memory occupation of load balancing node is than the discreet value Y of bulk flow
It is small:
The memory that node j is able to bear are as follows:
Wherein, tcFor the processing time of single data packet, M is connection number, tdFor timesharing interval, α and β are adjusting parameter, tw
For waiting time, tkFor interior external memory copy time, tsIt is data from data sending terminal to the data receiving terminal time experienced.
The beneficial effects of the present invention are:
The present invention is analyzed using DPI, and the flow of input is carried out to the division of characteristic matching, and it is whole to extract key message convection current
It is estimated, obtains discreet value.And subordinate's nodal information is obtained using dynamic monitoring system, handle in real time is carried out to server state
Control.It is pre-allocated by using discreet value and node real time information convection current, alleviates the complexity of load balancing, and will
Calculating is transferred in load balancing prediction, realizes the optimization to load balancing entirety.
Detailed description of the invention
Fig. 1 is the load-balancing method flow chart based on DPI prediction；
Fig. 2 is physical topological structure；
Fig. 3 is the load balancing module figure based on DPI prediction.
Specific embodiment
The present invention is described further with reference to the accompanying drawing.
The length of the counterbalance effect of load balancing and processing time in network node largely decide entire
The performance of network system.The numerous studies made in terms of load-balancing technique with people, and propose many various
Method.Such as the load balancing based on DNS proxy, this method by replacing response server although it is contemplated that reached
The effect of load balancing, but the state of each server is not accounted for, reply can not be made rapidly by fortuitous event occur, and
The factor of request of data itself is not accounted for；Load balancing based on protocol stack is then the seven-layer structure to analyze data packet
As according to classifying to flow, this method connects equipment in the line, obtains all flows for passing through route, uses
Data link layer of the matching technique to data packet, network layer, transportation level, application layer etc. is parsed, then according to configured
Distribution policy is forwarded flow.This method analyzes and determines data content, so that distribution better effect, but by
Then cascaded structure, data processing time influence whether overall network performance.
Patent document about load-balancing method is more, respectively there is its advantage and disadvantage.Such as application number 201710968835.X
In " a kind of application layer traffic load-balancing method based on DPI " disclosed in patent document, it is connected on client and server
Between, and service traffics are parsed using modes such as light splitting or mirror images, and export dynamic priority as load balancing
One of according to, optimal business service end is selected in conjunction with static priority, but light splitting or mirror image are excessive to the time of service resolution,
And the flow of part in series needs dynamic priority to influence entirety so waiting relationship occur in two modules as parameter
Performance.In addition the content that is parsed using DPI technology is simultaneously indefinite.
The object of the present invention is achieved like this:
1) input flow rate is divided into a rule stream according to five-tuple and sequence number, by DPI technology analyze agreement and
Determine which kind of the flow belongs to using data using feature；
2) it determines that flow belongs to the feature of which kind of application, the feature database model correctly applied is selected to be matched；
3) inquiry of the cryptographic Hash to historical record library content is calculated by traffic flow information, present traffic situation is carried out pre-
Estimate, obtains the discreet value of uninterrupted；
4) real-time monitoring load-balancing device subordinate's node of dynamic node is used, and by node current memory occupancy situation
It is fed back；
5) feedback result is analyzed, is pre-allocated using the discreet value that step 3) obtains, selects suitable distribution node；
6) data flow is sent, by flow actual size, resource name, the information updates such as source address apply feature database to it
In；
7) step 1) -6 is repeated), until flow, which is all sent, to be terminated.
Each application all carries the data of itself, and the content of data itself will not be scattered, but have certain spy
Sign rule, it is possible to data content be detected and is distinguish using packet check technology.By to using the number of plies
According to the acquisition of feature, feature string can establish, and then all contents being matched to are saved and establish feature database, with feature
Library constantly improve, and the prediction based on feature database also can be more and more accurate, and the effect of load balancing also can gradually improve.And simultaneously
Using Dynamic Load-Balancing Strategy, so that system algorithm has the control of real-time to server node, server can be directed to
It is adjusted in real time, makes load balancing effect more preferably stable.
The present invention is analyzed using DPI, and the flow of input is carried out to the division of characteristic matching, and it is whole to extract key message convection current
It is estimated, obtains discreet value.And subordinate's nodal information is obtained using dynamic monitoring system, handle in real time is carried out to server state
Control.It is pre-allocated by using discreet value and node real time information convection current, alleviates the complexity of load balancing, and will
Calculating is transferred in load balancing prediction, realizes the optimization to load balancing entirety.
It illustrates with reference to the accompanying drawing and the present invention is described in detail:
1) receive the flow for being input to load-balancing device, and according to five-tuple source IP, destination IP, source port, destination
Mouthful and agreement come distinguish stream, the not cocurrent flow of the same port ip is divided using sequence number, use five-tuple and sequence number as
The key value of Hash, saves record, and the above content is stored among network layer header and transportation level head, can use
Structural body pointer obtains.And application layer data feature is obtained using deep message analytic technique, meet the data flow of characteristic model R
It is defined as predictable data e.
R={ r1, r2..., rn-1, rn}
Before carrying out characteristic matching, need to carry out signature analysis to application layer data, in order to establish characteristic model R,
First collect widely apply A communication data, to data content carry out clustering, using DBSCAN Density Clustering to data into
Row analysis.Such as data e1Possess data segment p1p2p3p4p5, data e2Possess p1p6p7p4p5, data e3Possess p1p8p9p4p5, then
By p1, p4p5As feature string.
2) when predictable data e meets the feature for applying A by multimode matching, it is possible to determine that there is association in G (e, A),
I.e.
Wherein aiFor one of the discharge model of application A, that is, meet feature i applies A flow, and e belongs to one using A flow
Kind；
3) by using aiAs characteristic model, the key message in data e is obtained.Resource name name, resource characteristic L
={ l1, l2..., ln-1, ln, source resource title sname, source resource IPsip, resource type type, resource size size,
Resource remarks note, resource acquisition time t etc..The inquiry for calculating cryptographic Hash content in the LOG of historical record library, obtains similar money
Source information simultaneously carries out Predict analysis to present traffic situation, obtains the discreet value Y of uninterrupted；
4) gather around that there are two types of at flow using the real-time monitoring load-balancing device subordinate's node j of dynamic node, each node j
Reason mode, one kind be recombination classes, need flow in all data collect one by one recombinate after carry out data processing again；Another kind is to turn
Class is sent out, with the formal layout data flow of assembly line.It is current that real-time monitoring module carries out real-time monitoring acquisition node j to each node
EMS memory occupation situation σ, connection number M, flow tupe τ, Task scheduling pattern ρ etc. contents feed back to load balance process mould
Block；
6) feedback result is analyzed, is pre-allocated using the discreet value that step 4) obtains.Predistribution is executed, is first had to negative
It carries node state to be analyzed, when flow tupe τ is recombination classes, load node EMS memory occupation is discreet value Y；Flow processing
When mode τ is forwarding class, the EMS memory occupation of load balancing node is smaller than the discreet value Y of bulk flow:
When data flow e is divided into e1e2e3……en-1enIt is sent to load balancing node, data sending terminal sends data
e1, receiving terminal receives data e1When, the time passes through ts.Receiving end has handled e1When receive data ei, then receiving 1 and i
Between time be single data packet processing time tc.So practical committed memory of load node is about
The case where being single thread list CPU above, it is contemplated that multithreading situation can divide according to the form different from of Task scheduling pattern ρ
When the case where dispatching, connection number M and timesharing interval tdSome times can be occupied respectively, also will continue in this time data flow e
It sends, occupies more spaces；Under short task priority strategy, waiting time tw, but data can be relayed in waiting time
External memory is set, then re-defining interior external memory copy time is tk.Then the memory that node j is able to bear is estimated are as follows:
Wherein α and β is adjusting parameter, is adjusted according to factors such as actual task scheduling method ρ.Memory at most occupies in advance
The case where valuation Y, i.e. waiting time are too long or are directly entered external memory storage, switching to the size occupied after memory is Y.
Predistribution judgement is carried out to actual node j EMS memory occupation situation σ for data flow e, is selected currently using small top heap
The smallest node j of EMS memory occupation, if flow e is sent to node j, if the problem of memory excess occur, be, reselect
Otherwise node selects the node；
7) data flow e is sent, by flow actual size, resource name, the information such as source address calculate cryptographic Hash, and update
Into the historical record library LOG of its application, inquired for flow next time；
8) step 1) -7 is repeated), until flow, which is all sent, to be terminated.
The above is a kind of implementation of algorithm proposed by the present invention, but in certain steps, can suitably be changed, with suitable
Answer the demand of concrete condition.For example, the mode that other clusterings can be used is established when step 1) carries out feature string simultaneous
Feature string and feature database, each application model is different, and classification is different, using angle difference, it may be said that obtained using other modes
Take feature.Multimode matching algorithm can specify clearly effective multimode matching algorithm according to data characteristic in step 2).Step 6)
In by load node each information analysis and obtain node to flow can bearing value, compression calculation amount also can be used
Mode directly carries out fitting analysis by pre- appraisal Y.
Method of the present invention greatly has compressed calculation amount, load-balancing algorithm stage, by selecting in the smallest
It deposits and occupies the time complexity that node only needs (1) O；Dynamic node monitoring modular and feedback processing modules belong to parallel mould
Block does not cause time delay to whole system；Predicted portions parse message content using DPI technology, and use multimode
Matching technique matching characteristic, the minimum O (m) of time complexity, m are the length of pattern string, are then calculated historical record,
Predicted value is obtained, calculating section time complexity is the cumulative of N number of record, and data can be carried out with the meter of average value during preservation
It calculates, searches by the way that the information architectures hash data structure such as resource name source, time complexity is O (1), i.e. overall time complexity
For O (n).
The present invention passes through DPI technology to the analysis of data as the beginning, devises to predict the load balancing as main body
Algorithm structure, has compressed the calculation amount of Load Balancing Model itself, so it will be argued that the method and a kind of " base that this patent proposes
In the application layer traffic load-balancing method of DPI " there is difference substantially, although having used same technology, emphasis
It is different.Present invention incorporates multimode matching and Clustering Analysis Technologies, the use of DPI technology are only that cannot reach better effects
's.Different according to the content of clustering, the application layer data structure of carrying is different, and match pattern and analytic process may
It is adjusted according to the actual situation, has reached suitable effect.
Claims (6)
1. a kind of load-balancing method based on DPI prediction, which comprises the following steps:
(1) input flow rate is divided into a rule stream according to five-tuple and sequence number, agreement and application is analyzed by DPI technology
Feature determines which kind of the flow belongs to using data；
(2) it determines that flow belongs to the feature of which kind of application, the feature database model correctly applied is selected to be matched；
(3) inquiry of the cryptographic Hash to historical record library content is calculated by traffic flow information, present traffic situation is estimated,
Obtain the discreet value of uninterrupted；
(4) use dynamic node real-time monitoring load-balancing device subordinate's node, and by node current memory occupancy situation into
Row feedback；
(5) feedback result is analyzed, is pre-allocated using the discreet value that step (3) obtains, selects suitable distribution node；
(6) data flow is sent, by flow actual size, resource name, the information updates such as source address are to it using in feature database；
(7) step 1) -6 is repeated), until flow, which is all sent, to be terminated.
2. a kind of load-balancing method based on DPI prediction according to claim 1, which is characterized in that described to input
Flow is divided into a rule stream according to five-tuple and sequence number, analyzes agreement by DPI technology and determines the stream using feature
Which kind of amount belongs to using data, comprising:
Receive to be input to the flow of load-balancing device, and according to five-tuple source IP, destination IP, source port, destination port and
Agreement flows to distinguish, and the not cocurrent flow of the same port ip is divided using sequence number, uses five-tuple and sequence number as Hash
Record is stored among network layer header and transportation level head by key value, meet characteristic model R stream data definition be can be pre-
Measured data e, characteristic model R:
R={ r1, r2..., rn-1, rn}
Wherein, rnFor a certain characteristic element.
3. a kind of load-balancing method based on DPI prediction according to claim 1, which is characterized in that the determining stream
Amount belongs to the feature of which kind of application, and the feature database model correctly applied is selected to be matched, comprising:
When predictable data e meets the feature for applying A by multimode matching, determine that G (e, A) has association, i.e.,
Wherein aiFor one of the discharge model of application A, that is, meet feature i applies A flow, and e belongs to one kind using A flow.
4. a kind of load-balancing method based on DPI prediction according to claim 1, which is characterized in that described to pass through number
Inquiry of the cryptographic Hash to historical record library content is calculated according to stream information, present traffic situation is estimated, uninterrupted is obtained
Discreet value, comprising:
By using aiAs characteristic model, the key message obtained in data e includes resource name name, resource characteristic L=
{l1, l2..., ln-1, ln, source resource title sname, source resource Ipsip, resource type type, resource size size, money
Source remarks note, resource acquisition time t calculate the inquiry of cryptographic Hash content in the LOG of historical record library, obtain similar resource letter
It ceases and Predict analysis is carried out to present traffic situation, obtain the discreet value Y of uninterrupted:
Wherein, sizeiIt is characterized the corresponding resource size of i.
5. a kind of load-balancing method based on DPI prediction according to claim 1, which is characterized in that described using dynamic
Real-time monitoring load-balancing device subordinate's node of state node, and node current memory occupancy situation is fed back, comprising:
It is gathered around using dynamic node real-time monitoring load-balancing device subordinate's node j, each node j there are two types of flow tupe,
One kind be recombination classes, need flow in all data collect one by one recombinate after carry out data processing again；Another kind is forwarding class, with
The formal layout data flow of assembly line；Real-time monitoring module carries out real-time monitoring acquisition node j current memory to each node and accounts for
With situation σ, connection number M, flow tupe τ, Task scheduling pattern ρ etc. contents feed back to load balance process module.
6. a kind of load-balancing method based on DPI prediction according to claim 1, which is characterized in that the analysis is anti-
Feedback selects suitable distribution node as a result, pre-allocated using the discreet value that step (3) obtains, comprising:
It first has to analyze load node state,
If flow tupe τ is recombination classes, load node EMS memory occupation is discreet value Y；
If flow tupe τ is forwarding class, the EMS memory occupation of load balancing node is smaller than the discreet value Y of bulk flow:
The memory that node j is able to bear are as follows:
Wherein, tcFor the processing time of single data packet, M is connection number, taFor timesharing interval, α and β are adjusting parameter, twFor etc.
To time, tkFor interior external memory copy time, tsIt is data from data sending terminal to the data receiving terminal time experienced.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910196102.8A CN109729017B (en) | 2019-03-14 | 2019-03-14 | Load balancing method based on DPI prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910196102.8A CN109729017B (en) | 2019-03-14 | 2019-03-14 | Load balancing method based on DPI prediction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109729017A true CN109729017A (en) | 2019-05-07 |
| CN109729017B CN109729017B (en) | 2023-02-14 |
Family
ID=66302373
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910196102.8A Active CN109729017B (en) | 2019-03-14 | 2019-03-14 | Load balancing method based on DPI prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109729017B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110601991A (en) * | 2019-09-16 | 2019-12-20 | 赛尔网络有限公司 | Flow packet-by-packet distribution method and device, electronic equipment and storage medium |
| CN110995769A (en) * | 2020-02-27 | 2020-04-10 | 上海飞旗网络技术股份有限公司 | Deep data packet detection method and device and readable storage medium |
| CN111880926A (en) * | 2020-06-30 | 2020-11-03 | 苏州浪潮智能科技有限公司 | Load balancing method and device and computer storage medium |
| CN112491738A (en) * | 2020-12-08 | 2021-03-12 | 深圳海智创科技有限公司 | Real-time data stream balanced distribution method and device |
| CN114138463A (en) * | 2021-11-04 | 2022-03-04 | 中国电力科学研究院有限公司 | Deep neural network-based spot system application layer load balance prediction method |
| US20240022514A1 (en) * | 2022-07-18 | 2024-01-18 | Microsoft Technology Licensing, Llc | Dynamic load balancing based on flow characteristics |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101645806A (en) * | 2009-09-04 | 2010-02-10 | 东南大学 | Network flow classifying system and network flow classifying method combining DPI and DFI |
| CN101945407A (en) * | 2010-10-22 | 2011-01-12 | 东南大学 | Load balancing method for content monitoring of mobile service |
| CN101958843A (en) * | 2010-11-01 | 2011-01-26 | 南京邮电大学 | Intelligent routing selection method based on flow analysis and node trust degree |
| CN105847078A (en) * | 2016-03-17 | 2016-08-10 | 哈尔滨工程大学 | HTTP (Hyper Text Transport Protocol) traffic refined identification method based on DPI (Data Processing Installation) self-study mechanism |
| CN106209506A (en) * | 2016-06-30 | 2016-12-07 | 瑞斯康达科技发展股份有限公司 | A kind of virtualization deep-packet detection flow analysis method and system |
| CN107864189A (en) * | 2017-10-18 | 2018-03-30 | 南京邮电大学 | A kind of application layer traffic load-balancing method based on DPI |
| CN108632159A (en) * | 2017-03-16 | 2018-10-09 | 哈尔滨英赛克信息技术有限公司 | A kind of network service traffic load-balancing method based on prediction |
-
2019
- 2019-03-14 CN CN201910196102.8A patent/CN109729017B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101645806A (en) * | 2009-09-04 | 2010-02-10 | 东南大学 | Network flow classifying system and network flow classifying method combining DPI and DFI |
| CN101945407A (en) * | 2010-10-22 | 2011-01-12 | 东南大学 | Load balancing method for content monitoring of mobile service |
| CN101958843A (en) * | 2010-11-01 | 2011-01-26 | 南京邮电大学 | Intelligent routing selection method based on flow analysis and node trust degree |
| CN105847078A (en) * | 2016-03-17 | 2016-08-10 | 哈尔滨工程大学 | HTTP (Hyper Text Transport Protocol) traffic refined identification method based on DPI (Data Processing Installation) self-study mechanism |
| CN106209506A (en) * | 2016-06-30 | 2016-12-07 | 瑞斯康达科技发展股份有限公司 | A kind of virtualization deep-packet detection flow analysis method and system |
| CN108632159A (en) * | 2017-03-16 | 2018-10-09 | 哈尔滨英赛克信息技术有限公司 | A kind of network service traffic load-balancing method based on prediction |
| CN107864189A (en) * | 2017-10-18 | 2018-03-30 | 南京邮电大学 | A kind of application layer traffic load-balancing method based on DPI |
Non-Patent Citations (1)
| Title |
|---|
| 张宁: ""高性能网络业务分流技术研究"", 《硕士电子期刊》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110601991A (en) * | 2019-09-16 | 2019-12-20 | 赛尔网络有限公司 | Flow packet-by-packet distribution method and device, electronic equipment and storage medium |
| CN110995769A (en) * | 2020-02-27 | 2020-04-10 | 上海飞旗网络技术股份有限公司 | Deep data packet detection method and device and readable storage medium |
| CN111880926A (en) * | 2020-06-30 | 2020-11-03 | 苏州浪潮智能科技有限公司 | Load balancing method and device and computer storage medium |
| CN111880926B (en) * | 2020-06-30 | 2022-07-01 | 苏州浪潮智能科技有限公司 | Load balancing method and device and computer storage medium |
| CN112491738A (en) * | 2020-12-08 | 2021-03-12 | 深圳海智创科技有限公司 | Real-time data stream balanced distribution method and device |
| CN114138463A (en) * | 2021-11-04 | 2022-03-04 | 中国电力科学研究院有限公司 | Deep neural network-based spot system application layer load balance prediction method |
| CN114138463B (en) * | 2021-11-04 | 2024-03-26 | 中国电力科学研究院有限公司 | Method for predicting load balance of spot system application layer based on deep neural network |
| US20240022514A1 (en) * | 2022-07-18 | 2024-01-18 | Microsoft Technology Licensing, Llc | Dynamic load balancing based on flow characteristics |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109729017B (en) | 2023-02-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109729017A (en) | A kind of load-balancing method based on DPI prediction | |
| US6981029B1 (en) | System and method for processing a request for information in a network | |
| CN108599994A (en) | A kind of SDN slice building methods based on flow cluster | |
| CN106533806B (en) | Method for providing cross-layer QoS based on application awareness in multi-tenant SDN network | |
| CN104102700A (en) | Categorizing method oriented to Internet unbalanced application flow | |
| CN104767690B (en) | A kind of flow scheduling device and method | |
| Larsen et al. | Classification of dynamic vehicle routing systems | |
| CN108733821A (en) | A kind of distribution of monitor video sectional drawing and methods of exhibiting and system | |
| CN105656964B (en) | The implementation method and device of data-pushing | |
| CN107124303A (en) | The service chaining optimization method of low transmission time delay | |
| CN110324260A (en) | A kind of network function virtualization intelligent dispatching method based on flow identification | |
| CN106453143A (en) | Bandwidth setting method, device and system | |
| CN106254273A (en) | The dispatching method of a kind of data center network stream and system | |
| CN107846341A (en) | Method, relevant apparatus and the system of schedules message | |
| Cowdrey et al. | Applying queueing theory for the optimization of a banking model | |
| Sachan et al. | SDN control-enabled and time-quantum-driven max-pressure approach for intersection management in smart city | |
| CN109087219B (en) | Tourism cloud computing system and method and terminal | |
| CN102271078A (en) | Service quality guarantee oriented load balancing method | |
| CN109495565A (en) | High concurrent service request processing method and equipment based on distributed ubiquitous computation | |
| CN106341343B (en) | A kind of automation services degradated system and method | |
| CN110365749B (en) | Message pushing method, message pushing system and storage medium | |
| CN108141377A (en) | Network flow early stage classifies | |
| Mehdian et al. | Join-the-shortest-queue scheduling with delay | |
| CN111611332B (en) | Optimization method and system for route transfer | |
| CN107302571A (en) | Information centre's network route and buffer memory management method based on drosophila algorithm |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |