CN105915467B - A kind of data center network flow equalization method and device that software-oriented defines - Google Patents
A kind of data center network flow equalization method and device that software-oriented defines Download PDFInfo
- Publication number
- CN105915467B CN105915467B CN201610326176.5A CN201610326176A CN105915467B CN 105915467 B CN105915467 B CN 105915467B CN 201610326176 A CN201610326176 A CN 201610326176A CN 105915467 B CN105915467 B CN 105915467B
- Authority
- CN
- China
- Prior art keywords
- data packet
- flow table
- forwarding
- data
- stream
- 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
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L47/00—Traffic control in data switching networks
- H04L47/10—Flow control; Congestion control
- H04L47/12—Avoiding congestion; Recovering from congestion
- H04L47/125—Avoiding congestion; Recovering from congestion by balancing the load, e.g. traffic engineering
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a kind of data center network flow equalization methods and device that software-oriented defines, method includes: the data packet that controller desampler is sent, the feature field of data packet is matched with the field in flow table, and according to preset movement priority by the action record of data packet in the behavior aggregate of flow table;If the fields match success in the feature field and flow table of data packet executes respective operations to data packet according to the sequence of movement in behavior aggregate;If respective action is pretreatment forwarding, judge whether the slot fields of data packet match with the transmission time slot value in flow table, if matching, data packet is forwarded, if mismatching, data pack buffer to next time slot is forwarded.This invention simplifies deployment, enhance the scalability of network, solve the problems, such as the issuable data packet disorder of data flow packet loss caused by strategy process implementation process and receiving end.
Description
Technical field
The present invention relates to data center network flow techniques fields, and in particular to a kind of data center that software-oriented defines
Network flow equalization methods and device.
Background technique
The traffic engineering of data center network is an important directions of current network research, since data center network is returned
It is all to belong to single operator, convenient for the unified plan of traffic engineering, naturally meets software defined network (Software Defined
Network, SDN) required for centralized control requirement.Meanwhile usually there is data center network symmetrical topological structure to make
There are a large amount of redundant links between node pair, can improve single path routing using multi-path transmission and make since " routing " is concentrated
At congestion problems, while can accomplish being switched fast for faulty link, enhance reliability, and can aggregated link bandwidth, sufficiently benefit
The main method that flow equalization is current data center network traffic engineering is carried out with these redundant links.
In current research work, one is the flow equalization strategies based on data flow granularity, using static road of equal value
Increase the step of random selection core node in the uplink by (Equal-Cost Multipath Routing, ECMP) mechanism
Suddenly, but when occurring big stream in link, it is flat to will cause link circuit resource unfair distribution.By using a kind of heuristic routing
Algorithm (Global First Fit, GFF), which introduces link state monitoring mechanism, can overcome the shortcomings of ECMP, but can not
The meeting avoided generates packet loss in the transition process flowed greatly and may cause receiving end data packet disorder;Another kind is based on number
According to the flow equalization strategy of packet granularity, Fastpass also controls the transmission of data source other than designing new communication protocol
Agreement is modified, while also needing to increase the caching of receiving end to handle the data packet of receiving end random ordering, implement compared with
More to transport protocol or architecture change for complexity, each of which increases the complexity of deployment, scalability is not high.
Therefore, the significant challenge faced in current research work is how to design a kind of data center network flow equalization side
Method, can simplify deployment, enhancing scalability and in strategy process implementation process caused by data flow packet loss and
The issuable data packet disorder problem in receiving end.
Summary of the invention
Since current data center network flow equalization method disposes cumbersome, poor expandability, and can not solve in plan
The problem of data flow packet loss caused by slightly in method implementation process and the issuable data packet disorder in receiving end, the present invention mentions
A kind of data center network flow equalization method and device that software-oriented defines out.
In a first aspect, the present invention proposes a kind of data center network flow equalization method that software-oriented defines, comprising:
The data packet that controller desampler is sent carries out the field in the feature field and flow table of the data packet
Matching, and according to preset movement priority by the action record of the data packet in the behavior aggregate of flow table;
If the fields match success in the feature field of the data packet and the flow table, according to the movement in behavior aggregate
Sequentially, respective operations are executed to the data packet;
If the respective action is pretreatment forwarding, the slot fields for judging the data packet and the biography in the flow table
Whether defeated slot values match, if matching, the data packet is forwarded, if mismatching, by the data pack buffer to next
A time slot is forwarded, the flow of the data center network with equilibrium towards software definition;
Wherein, the movement of the data packet includes processing, pretreatment forwarding, forwarding and abandons.
Preferably, if the fields match of flow table fails in the feature field of the data packet and interchanger, according to default
Static flow table in behavior aggregate to the data packet execute respective operations.
Preferably, the data packet that the controller desampler is sent, by the feature field and flow table of the data packet
In field matched, and according to it is preset movement priority by the action record of the data packet flow table behavior aggregate
In before, further includes:
If single stream link bandwidth occupancy is more than bandwidth threshold in the currently transmitted period, it is determined that current stream is big stream,
And the big stream is added in big adfluxion conjunction;
If the link load that the big stream passes through is more than load threshold, from the marginal layer direct-connected with the source of the big stream
The smallest two paths of time delay are selected in the preset path of interchanger, for cutting in two paths when data packet forwarding
It changes.
Preferably, current to flow if single stream link bandwidth occupancy is more than bandwidth threshold in the currently transmitted period
To flow greatly, and by the big stream be added to big adfluxion close in after, further includes:
If current big stream is more than the load threshold in the link load that a upper sending cycle is passed through, from it is current greatly
The smallest two paths of time delay are selected in the direct-connected preset path of edge layer switch of the source of stream, when for data packet forwarding
Switch in two paths.
Preferably, the link load passed through when the big stream is more than load threshold, then from the source with the big stream
The smallest two paths of time delay are selected in the direct-connected preset path of edge layer switch, described two when forwarding for data packet
After switching in paths, further includes:
Transmission time slot value is calculated according to the propagation delay time in path, and the flow table is updated according to the transmission time slot value.
Second aspect, the present invention also propose a kind of data center network flow equalization device that software-oriented defines, comprising:
Fields match module, for the data packet that desampler is sent, by the feature field and flow table of the data packet
In field matched, and according to it is preset movement priority by the action record of the data packet flow table behavior aggregate
In;
First action executing module, for when the fields match success in the feature field and the flow table of the data packet
When, according to the sequence of movement in behavior aggregate, respective operations are executed to the data packet;
Forwarding module is pre-processed, for judging the time slot of the data packet when the respective action is pretreatment forwarding
Whether field matches with the transmission time slot value in the flow table,, will if mismatching if matching, the data packet is forwarded
The data pack buffer to next time slot is forwarded, the flow of the data center network with equilibrium towards software definition;
Wherein, the movement of the data packet includes processing, pretreatment forwarding, forwarding and abandons.
Preferably, further includes:
Second action executing module is lost for the fields match when flow table in the feature field of the data packet and interchanger
When losing, respective operations are executed to the data packet according to the behavior aggregate in preset static flow table.
Preferably, further includes:
Big stream judgment module, is used for when single stream link bandwidth occupancy is more than bandwidth threshold in the currently transmitted period,
It determines that currently stream is big stream, and the big stream is added in big adfluxion conjunction;
First path selecting module, for when it is described it is big stream pass through link load be more than load threshold when, from it is described
The smallest two paths of time delay are selected in the direct-connected preset path of edge layer switch of the source flowed greatly, for data packet forwarding
When switch in two paths.
Preferably, further includes:
Second path selection module is more than described for ought currently flow the link load passed through in a upper sending cycle greatly
When load threshold, selection time delay is two the smallest from the edge layer switch preset path direct-connected with the source that currently flows greatly
Path, for switching in two paths when data packet forwarding.
Preferably, further includes:
Slot values update module, for calculating transmission time slot value according to the propagation delay time in path, and when according to the transmission
Gap value updates the flow table.
As shown from the above technical solution, the present invention passes through controller by the action record of data packet in behavior aggregate, and root
Respective operations are executed according to the sequence of movement in behavior aggregate, simplify deployment, while enhancing the scalability of network;Pass through judgement
Whether the slot fields of data packet match with the transmission time slot value in flow table, if matching, the data packet is forwarded, if not
Match, is then forwarded the data pack buffer to next time slot, number caused by solving in strategy process implementation process
According to stream packet loss and the issuable data packet disorder problem in receiving end.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
Other attached drawings are obtained according to these figures.
Fig. 1 is the data center network flow equalization method that a kind of software-oriented that one embodiment of the invention provides defines
Flow diagram;
Fig. 2 is the system construction drawing for the data center network that a kind of software-oriented that one embodiment of the invention provides defines;
Fig. 3 be another embodiment of the present invention provides a kind of data center network flow equalization method that defines of software-oriented
Flow chart;
Fig. 4 be another embodiment of the present invention provides a kind of data center network flow equalization method that defines of software-oriented
Triggering and cancel flow chart;
Fig. 5 is the data center network flow equalization device that a kind of software-oriented that one embodiment of the invention provides defines
Structural schematic diagram.
Specific embodiment
With reference to the accompanying drawing, the specific embodiment of invention is further described.Following embodiment is only used for more clear
Illustrate to Chu technical solution of the present invention, and not intended to limit the protection scope of the present invention.
Fat-tree network topology is used first against embodiment, the interchanger in topology is needed using under controller
It sends out static flow table to be configured, workflow in order to more clearly describe the present invention is as follows to the concept definition being related to:
O-D (Origin-Destination) is right: referring to two that the source of certain stream and purpose end main frame are directly connected to
Edge layer switch, O-D in experiment is to being belonging respectively to different POD (Performance Optimization
Datacenter)。
Intermediary switch: refer to convergence layer and core layer switch.
Interchanger path: refer to the edge passed through during O-D pairs of interchanger one end to the other end when some data packet
Path composed by interchanger and intermediary switch.
Static path: referring to for each O-D pairs, all a specified in advance fixed transmission path.When FLAT not by
When triggering, the data flow between O-D will be along the path transmission, wherein FLAT indicates software-oriented definition provided by the invention
Data center network flow equalization method.
The path FLAT: referring to for each O-D pairs, and all k/2 item specified in advance (parameter that k is Fat-tree topology) is solid
Fixed, disjoint interchanger path.When FLAT is when some O-D is to being above triggered, only selected from this k/2 fixed route
The transmitting path of current time slots.For some O-D pairs, static path is included in the path FLAT.
The setting of the flow table in static path and the path FLAT: for intermediary switch, all FLAT routing informations all will be with
The form of static flow entry is stored in flow table.For edge switch, then the corresponding stream of all static paths is only stored in advance
Table.When FLAT is not triggered, data packet can be completed to forward along preset static path, and after FLAT starting, then by controlling
Device processed issues the higher FLAT flow table realizing route switching of priority to edge switch.
Secondly, also relating to the detection process to path and link data, relevant concept in the course of work of FLAT
And main points are explained as follows:
Global path delay detection: in order to assess each FLAT transmission path before first time slot when FLAT is triggered
Superiority and inferiority and calculate transmission time slot value, it would be desirable to 4 (transmission path quantity can be set in advance) between each O-D pairs are pre-
If sending detection packet on path carrys out to acquisite approachs time delay value.This detection is known as global path detection, these detection coatings are known as
Global path detection packet.It should be noted that this process will be continued for, whether there is path triggering FLAT strategy.Detection
The transmission interval of packet has to suitably, not strong if too long obtained data age, too short, will increase data link and
Time delay of the interchanger to controller communication link.
The FLAT path delay of time detects: flowing the latency measurement that corresponding all paths FLAT carry out greatly to certain, purpose and complete
Office's detection is similar.Since this expense is relatively small, and need the timeliness to big stream detection stronger, so this detection is than complete
The time interval of office's detection wants short.
Link load detection: FLAT entry condition is that big stream causes certain link load more than threshold value;When FLAT starts
Afterwards, it needs to select optimal path according to the loading condition in path, it is therefore desirable to all on all paths FLAT flowed greatly
Link load measures, this is known as single-pathway link detection (abbreviation link detection).It does not need to issue spy on the link
Survey packet, it is only necessary to periodically to controller upwards transmission port statistical data, but only come into force on specified path.When usually it is with FLAT
Prolonging detection has the same period.
FLAT detection cycle: referring to the period of FLAT delay detection and link detection, it also determines that FLAT obtains number simultaneously
According to, period shortest time of decision and judgement, it is hereinafter also called the period for short.
Fig. 1 shows the data center network flow equalization side that a kind of software-oriented of one embodiment of the invention offer defines
The flow diagram of method, comprising:
The data packet that S1, controller desampler are sent, by the field in the feature field and flow table of the data packet
Matched, and according to preset movement priority by the action record of the data packet in the behavior aggregate of flow table;
If the fields match success in the feature field of S2, the data packet and the flow table, according in behavior aggregate
Sequence of movement executes respective operations to the data packet;
If S3, the respective action are pretreatment forwarding, judge in slot fields and the flow table of the data packet
Transmission time slot value whether match, if matching, the data packet is forwarded, if mismatch, extremely by the data pack buffer
Next time slot is forwarded, the flow of the data center network with equilibrium towards software definition;
Wherein, the movement of the data packet includes processing, pretreatment forwarding, forwarding and abandons.
The present embodiment passes through controller by the action record of data packet in behavior aggregate, and suitable according to the movement in behavior aggregate
Sequence executes respective operations, simplifies deployment, while enhancing the scalability of network;By judge the slot fields of data packet with
Whether the transmission time slot value in flow table matches, if matching, the data packet is forwarded, if mismatching, by the data packet
Caching is forwarded to next time slot, solves data flow packet loss and receiving end caused by strategy process implementation process
Issuable data packet disorder problem.
Further, after S3, further includes:
If the fields match failure of flow table in S4, the feature field of the data packet and interchanger, according to preset quiet
Behavior aggregate in state flow table executes respective operations to the data packet.
By the way that static flow table is arranged, the fields match of flow table is lost in the feature field and interchanger to realize the data packet
Routing when losing, between each pair of edge switch.
Specifically, before S1, further includes:
If S01, single stream link bandwidth occupancy is more than bandwidth threshold in the currently transmitted period, it is determined that current stream is big
Stream, and the big stream is added in big adfluxion conjunction;
If the link load that S02, the big stream pass through is more than load threshold, from the side direct-connected with the source of the big stream
The smallest two paths of time delay are selected in the preset path of edge layer interchanger, when forwarding for data packet in two paths
Switching.
By judging that the link load that big stream passes through is more than load threshold and executes corresponding operating, appearance tradition can be avoided
Data flow packet loss and data packet disorder situation caused by link load is excessive in method.
Further, after S01, further includes:
If the link load that passes through in a upper sending cycle of S012, current big stream more than the load threshold, from
The source currently flowed greatly selects the smallest two paths of time delay in the direct-connected preset path of edge layer switch, for data packet
Switch in two paths when forwarding.
In order to avoid current big stream is more than the load threshold in the link load that a upper sending cycle is passed through, and it is current
The link load that period passes through is less than the load threshold, thus caused by route oscillation problem and reduce implementation effect
Problem.
Further, after S02, further includes:
S03, transmission time slot value is calculated according to the propagation delay time in path, and the stream is updated according to the transmission time slot value
Table.
By updating the transmission time slot value in flow table, so that transmission time slot value is more in line with the current operating condition of network,
Further Balanced network load.
It is first in order to which the data center network flow equalization method that software-oriented provided in this embodiment defines is described in more detail
The data center network first defined to software-oriented provided in this embodiment is introduced:
The data center network that one software-oriented defines includes: server, access-layer switch, core layer switch with
And controller, wherein controller is connected with all interchangers, and interchanger supports OpenFlow (OF) agreement;
As shown in Fig. 2, controller as application layer platform, integrates four modular assemblies to realize entire controller platform
Function include: topology information collection module, routing information statistical module, path calculation module, flow table update module;
Topology information collection module finds network topology by collecting more new information to neighbouring interchanger.Interchanger can
The information of update includes finding the link-state message of neighbours' message and each neighbours interconnection, passes through the interaction with exchanger information
It determines current network topology figure, and provides information to path calculation module.Communication protocol between controller platform and interchanger
OF agreement is all used, each interchanger runs OF agency by agreement software, is responsible for the message that received data packet forwarding updates and lays equal stress on
Newly write the forwarding table based on exchange hardware;
Routing information statistical module realizes two functions in such a way that active inquiry and passive reception combine: first is that logical
It crosses and inquires correlation OF interchanger periodically to collect the traffic statistics of each port;Second is that by the acquisition of structural exploration packet and more
The Delay of new transmission path, information above will be all saved in data statistic.Wherein, traffic statistics include control
After device receives FlowRemoved message, the byte number that each port of interchanger sends and receives is counted according to flow table counter
Amount, the calculating for link residual capacity provide foundation;
Path calculation module runs core of the invention algorithm FLAT, the net provided first by topology information collection module
Network topology calculates available path between any node pair, it is contemplated that the computing cost and exchange hardware condition of controller
Limitation (for example, TCAM capacity and cost), can to selection multi-path transmission quantity be limited (for example, be no more than three
Item);Secondly, the data provided according to routing information statistical module can carry out residual capacity calculating and biography to the link of selection
The calculating of defeated time slot, and link most lightly loaded is selected to carry out forwarding data flow;Finally, the result of path computing is converted to road
Interchanger is issued to by forward rule;
The installation for the routing forwarding rule progress flow table that flow table update module issues driving OF interchanger foundation controller,
Update and delete operation.
It is random that data center network provided in this embodiment not will cause receiving end data packet during data stream scheduling
The problem of sequence;Controller platform obtains the Delay of flow information and transmission path in such a way that master passively combines measurement,
It ensure that the accuracy of algorithm;It will not be to generally being used in current data center network while meeting fine granularity flow equalization
TCP communication agreement do any change, be easy to dispose, have scalability.
Below with reference to Fig. 2-4, method provided in this embodiment is further elaborated.
Fig. 2 is that the present invention is based on the prototype systems that FLAT algorithm implements flow equalization method.As shown in Fig. 2, controller is made
For application layer platform, integrated four modular assemblies are respectively as follows: topology information collection module, routing information statistical module, path
Computing module and flow table update module realize the function of entire controller platform jointly;Wherein, path calculation module runs this reality
Apply the core methed FLAT of example.
The workflow of control platform is as follows: firstly, topology information collection module to neighbouring interchanger by collecting more
New information finds network topology, and FLAT is then available between any node pair to calculate by network topology that the module provides
Path;In next step, link of the data provided according to routing information statistical module to selection is carried out residual capacity calculating by FLAT
And the calculating of transmission time slot, and link most lightly loaded is selected to carry out forwarding data flow;Finally, FLAT is by the knot of path computing
Fruit is converted to routing forwarding rule and is issued to interchanger realization data packet forwarding.
Fig. 3 is FLAT workflow, specifically includes the following steps:
S201, data packet reach Openflow interchanger;
S202, interchanger are according in the certain feature fields (such as source IP, destination IP, MAC Address) and flow table of data packet
Entry matched.If it fails to match, S203 is executed;If successful match, jumps and execute S204;
It will appear packet-in event under S203, default situations, when matching unsuccessful.But in experimental data central site network
In, all interchangers are all prefixed static flow table with lower priority, and Lai Shixian does not trigger each pair of edge switch when FLAT
Between one routing.So if it fails to match, then illustrates static flow table failure, mistake should be returned to, and under trial again
Static flow table is sent out, execution S202 is jumped later and reattempts to matching flow table;
S204, in Openflow1.3 standard, a data packet, can be suitable according to scheduled priority when matching flow table
All entry movement is all recorded in behavior aggregate by sequence, is executed in order after having matched.Usually forwarding, discarding are by as one
The final movement of a data packet and execute, therefore other movements other than forwarding and discarding can all be come by what both be come
It completes.And FLAT can regard a kind of forwarding behavior being modified as, therefore can also come other movement after;When having executed
After other movements, remaining movement should be FLAT, forwarding or discarding;
If S205, current action are FLAT, S206 is executed;S207 is executed if it is not, then jumping;
S206, judge current time slots, whether which, which needs, caches.If you do not need to caching, then execute S207;If necessary
Caching then jumps and executes S208;
Forwarding behavior (coming from S206) under S207, execution FLAT strategy, or execute and act it from the non-FLAT of S205
Outer forwarding abandons operation, after the completion of the above movement, jumps and executes S209;
The current packet of S208, caching arrives queue, is forwarded after current transmission time slot further according to FLAT strategy;
S209, the data packet are disposed, and terminate process;If there is next data packet reaches, jumps and execute S202.
Fig. 4 be FLAT triggering with cancel work flow diagram, specifically includes the following steps:
The sending cycle of S301, the global detection packet of setting, link detection packet;
S302, start to carry out global detection, while recording measurement result;
S303, judge whether there is big miscarriage life in network;Judge that the foundation flowed greatly is source end host stream link bandwidth
Occupancy is more than given threshold (usually 10%).If there is life of miscarrying greatly, then it is added to big adfluxion and closes in F and execute S304;
Otherwise, it jumps and executes S313;
S304, link load detection is carried out to the corresponding link of all big streams in set, on the one hand can finds out and flows through greatly
Whether some link crossed is more than threshold value;On the other hand information is provided for the path computing after FLAT triggering;
Whether S305, the current big stream of judgement in a upper detection cycle executed FLAT, if it is not, executing S306;
Otherwise it jumps and executes S307, being the reason of whether executing FLAT in a upper period is judged to current big stream: if the big stream
Fail to trigger FLAT in this period, then FLAT may be triggered again to generate route oscillation by returning to original route, reduce
The implementation effect of FLAT strategy, therefore, the big stream for executing FLAT to a upper period continue to execute, until it be no longer big stream or
Person terminates transmission;
If some link load that S306, big stream pass through is more than threshold value, S307 is executed;Otherwise next is just checked
Big stream, and jump and execute S312;
S307, FLAT algorithm triggers and according to current measurement result, the direct-connected edge layer switch of Cong great Liu source is pre-
In 4 transmission paths set, transmission path of the smallest 2 paths of time delay as FLAT is selected.In the validity period of FLAT flow table
Interior, the data packet currently flowed greatly will switch on this two paths according to forwarding strategy.It should be noted that the of operation FLAT
One time slot, interchanger need first to cache first flow (data packet that mono- time slot of FLAT is transmitted) of the stream, prevent from counting
Receiving end, which is reached, according to packet generates random ordering;
S308, FLAT calculate transmission time slot value according to the measurement result of current record.The calculating of slot values briefly describes
As follows: when the time delay difference between 4 preset transmission paths is less than 100 μ s, taking slot values is 100 μ s;When time delay difference
In 100-500 μ s, taking slot values is 500 μ s;When time delay difference is greater than 500 μ s, using actual difference as slot values;
It is preferential that the priority setting that S309, controller issue FLAT flow table is higher than static flow table preset in interchanger
Grade, at the same in flow table encapsulated delivery slot values, the life span of forwarding port and flow table corresponding to transmission path (TTL,
Time To Live), ttl value is link detection periodic quantity;
FLAT flow table issuance is flowed the direct-connected edge layer switch of source to big by S310, controller.Due to each OD
To optional 4 transmission paths, and the interchanger being issued to by way of static flow table in all transmission paths, because
This, data flow according to FLAT flow table between mulitpath switching according to packet can centainly arrive at the destination location, and FLAT
Flow table need to only be issued to edge layer switch, and design in this way can greatly reduce the communication between interchanger and controller
Expense;
If S311, current big stream still have in the next period, it will need the timeliness stronger path delay of time
Information calculates new FLAT time slot size, it is therefore desirable in this period measure all FLAT path delays of time again;
S312, start to check next big stream, jump and execute S305;If all big stream inspecteds finish in set F,
Execute S313;
S313, current period have been fully completed the detection and decision flowed greatly, it is also possible to which current period does not detect
At this moment big stream (coming from S303) is just terminating program if receiving program determination instruction;If continued to run, execute
S314;
S314, it after waiting current period, is again introduced into next cycle and jumps execution S303;
By a large amount of emulation experiment, find method provided in this embodiment compared with other several multi-path transmission methods
(for example, ECMP, GFF), handling capacity improves 13%~45%, while receiving end occurs almost without out-of-order data packet, with
It is equivalent using single path laser propagation effect, but handling capacity by comparison but can maximum be increased to 83%.From experimental result it may be said that
Bright the present embodiment has reached expected purpose.
Fig. 5 shows the knot for the data center network flow equalization device that a kind of software-oriented provided in this embodiment defines
Structure schematic diagram, comprising:
Fields match module 11, for the data packet that desampler is sent, by the feature field and stream of the data packet
Field in table is matched, and according to it is preset movement priority by the action record of the data packet flow table movement
It concentrates;
First action executing module 12, for when the data packet feature field and the flow table in fields match at
When function, according to the sequence of movement in behavior aggregate, respective operations are executed to the data packet;
Pre-process forwarding module 13, for when the respective action be pretreatment forwarding when, judge the data packet when
Whether gap field matches with the transmission time slot value in the flow table, if matching, the data packet is forwarded, if mismatching,
The data pack buffer to next time slot is forwarded, the flow of the data center network with equilibrium towards software definition;
Wherein, the movement of the data packet includes processing, pretreatment forwarding, forwarding and abandons.
The present embodiment passes through controller by the action record of data packet in behavior aggregate, and suitable according to the movement in behavior aggregate
Sequence executes respective operations, simplifies deployment, while enhancing the scalability of network;By judge the slot fields of data packet with
Whether the transmission time slot value in flow table matches, if matching, the data packet is forwarded, if mismatching, by the data packet
Caching is forwarded to next time slot, solves data flow packet loss and receiving end caused by strategy process implementation process
Issuable data packet disorder problem.
Optionally, described device further include:
Second action executing module is lost for the fields match when flow table in the feature field of the data packet and interchanger
When losing, respective operations are executed to the data packet according to the behavior aggregate in preset static flow table.
Further, described device further include:
Big stream judgment module, is used for when single stream link bandwidth occupancy is more than bandwidth threshold in the currently transmitted period,
It determines that currently stream is big stream, and the big stream is added in big adfluxion conjunction;
First path selecting module, for when it is described it is big stream pass through link load be more than load threshold when, from it is described
The smallest two paths of time delay are selected in the direct-connected preset path of edge layer switch of the source flowed greatly, for data packet forwarding
When switch in two paths.
Further, described device further include:
Second path selection module is more than described for ought currently flow the link load passed through in a upper sending cycle greatly
When load threshold, selection time delay is two the smallest from the edge layer switch preset path direct-connected with the source that currently flows greatly
Path, for switching in two paths when data packet forwarding.
Further, described device further include:
Slot values update module, for calculating transmission time slot value according to the propagation delay time in path, and when according to the transmission
Gap value updates the flow table.
The data center network flow equalization device that software-oriented described in the present embodiment defines can be used for executing above-mentioned
Embodiment of the method, principle is similar with technical effect, and details are not described herein again.
In specification of the invention, numerous specific details are set forth.It is to be appreciated, however, that the embodiment of the present invention can be with
It practices without these specific details.In some instances, well known method, structure and skill is not been shown in detail
Art, so as not to obscure the understanding of this specification.
Claims (6)
1. a kind of data center network flow equalization method that software-oriented defines characterized by comprising
If single stream link bandwidth occupancy is more than bandwidth threshold in the currently transmitted period, it is determined that current stream is big stream, and will
The big stream is added in big adfluxion conjunction;
If the link load that passes through in a upper sending cycle of current big stream more than load threshold, from the source that currently flows greatly
The smallest two paths of time delay are selected in the direct-connected preset path of edge layer switch, described two when forwarding for data packet
Switch in paths;
If the link load that the big stream passes through is more than load threshold, exchanged from the marginal layer direct-connected with the source of the big stream
The smallest two paths of time delay are selected in the preset path of machine, for switching in two paths when data packet forwarding;
The data packet that controller desampler is sent, by the field progress in the feature field and flow table of the data packet
Match, and according to preset movement priority by the action record of the data packet in the behavior aggregate of flow table;
If the fields match success in the feature field of the data packet and the flow table, suitable according to the movement in behavior aggregate
Sequence executes respective operations to the data packet;
If the respective action be pretreatment forwarding, judge the data packet slot fields and the flow table in transmission when
Whether gap value matches, if matching, the data packet is forwarded, if mismatch, by the data pack buffer to it is next when
Gap is forwarded, the flow of the data center network with equilibrium towards software definition;
Wherein, the movement of the data packet includes processing, pretreatment forwarding, forwarding and abandons.
2. the method according to claim 1, wherein if flow table in the feature field and interchanger of the data packet
Fields match failure, then respective operations are executed to the data packet according to the behavior aggregate in preset static flow table.
3. according to the method described in claim 2, it is characterized in that, the link load passed through when the big stream is more than load
Threshold value then selects the smallest two roads of time delay from the layer switch preset path in edge direct-connected with the source of the big stream
Diameter, after switching in two paths when data packet forwarding, further includes:
Transmission time slot value is calculated according to the propagation delay time in path, and the flow table is updated according to the transmission time slot value.
4. a kind of data center network flow equalization device that software-oriented defines characterized by comprising
Big stream judgment module, if being more than bandwidth threshold for single stream link bandwidth occupancy in the currently transmitted period, it is determined that
Currently stream is big stream, and the big stream is added in big adfluxion conjunction;
Second path selection module, if being more than load threshold for the current big link load passed through in a upper sending cycle that flows
Value then selects the smallest two paths of time delay from the layer switch preset path in edge direct-connected with the source that currently flows greatly,
For switching in two paths when data packet forwarding;
First path selecting module, if for it is described it is big stream pass through link load be more than load threshold, from the big stream
The direct-connected preset path of edge layer switch of source in select the smallest two paths of time delay, for when data packet forwarding
Switch in two paths;
Fields match module will be in the feature field and flow table of the data packet for the data packet that desampler is sent
Field is matched, and according to preset movement priority by the action record of the data packet in the behavior aggregate of flow table;
First action executing module, for when the fields match success in the feature field of the data packet and the flow table,
According to the sequence of movement in behavior aggregate, respective operations are executed to the data packet;
Forwarding module is pre-processed, for judging the slot fields of the data packet when the respective action is pretreatment forwarding
Whether matched with the transmission time slot value in the flow table, it, will be described if mismatching if matching, the data packet is forwarded
Data pack buffer to next time slot is forwarded, the flow of the data center network with equilibrium towards software definition;
Wherein, the movement of the data packet includes processing, pretreatment forwarding, forwarding and abandons.
5. device according to claim 4, which is characterized in that further include:
Second action executing module fails for the fields match when flow table in the feature field of the data packet and interchanger
When, respective operations are executed to the data packet according to the behavior aggregate in preset static flow table.
6. device according to claim 5, which is characterized in that further include:
Slot values update module, for calculating transmission time slot value according to the propagation delay time in path, and according to the transmission time slot value
Update the flow table.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610326176.5A CN105915467B (en) | 2016-05-17 | 2016-05-17 | A kind of data center network flow equalization method and device that software-oriented defines |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610326176.5A CN105915467B (en) | 2016-05-17 | 2016-05-17 | A kind of data center network flow equalization method and device that software-oriented defines |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105915467A CN105915467A (en) | 2016-08-31 |
CN105915467B true CN105915467B (en) | 2019-06-18 |
Family
ID=56749133
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610326176.5A Active CN105915467B (en) | 2016-05-17 | 2016-05-17 | A kind of data center network flow equalization method and device that software-oriented defines |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105915467B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112506648A (en) * | 2020-11-20 | 2021-03-16 | 鹏城实验室 | Traffic stateless migration method of virtual network function instance and electronic equipment |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110351196B (en) * | 2018-04-02 | 2020-09-08 | 华中科技大学 | Load balancing method and system based on accurate congestion feedback in cloud data center |
CN109561026B (en) * | 2018-12-10 | 2021-09-28 | 福州大学 | Burst flow processing method based on software defined network data plane with state |
EP3716537A1 (en) * | 2019-03-25 | 2020-09-30 | Siemens Aktiengesellschaft | Method for data communication, network nodes, computer program and computer readable medium |
CN110086870A (en) * | 2019-04-28 | 2019-08-02 | 北京工业大学 | The parallel method for reliable transmission of the multipath of data-oriented central site network |
CN111654446B (en) * | 2020-05-26 | 2022-11-29 | 北京城建智控科技股份有限公司 | Big data flow control method and device and flow server |
CN113794655A (en) * | 2020-07-17 | 2021-12-14 | 北京京东尚科信息技术有限公司 | Flow monitoring method and device, computer storage medium and electronic equipment |
CN114448899A (en) * | 2022-01-20 | 2022-05-06 | 天津大学 | Method for balancing network load of data center |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103354522A (en) * | 2013-06-28 | 2013-10-16 | 华为技术有限公司 | Method and device for searching multistage flow table |
US9009349B2 (en) * | 2013-02-08 | 2015-04-14 | Dell Products, Lp | System and method for dataplane extensibility in a flow-based switching device |
CN105099920A (en) * | 2014-04-30 | 2015-11-25 | 杭州华三通信技术有限公司 | Method and device for setting SDN flow entry |
CN105227481A (en) * | 2015-09-02 | 2016-01-06 | 重庆邮电大学 | The SDN congestion control method for routing of cost minimization is dispatched based on path cost and stream |
-
2016
- 2016-05-17 CN CN201610326176.5A patent/CN105915467B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9009349B2 (en) * | 2013-02-08 | 2015-04-14 | Dell Products, Lp | System and method for dataplane extensibility in a flow-based switching device |
CN103354522A (en) * | 2013-06-28 | 2013-10-16 | 华为技术有限公司 | Method and device for searching multistage flow table |
CN105099920A (en) * | 2014-04-30 | 2015-11-25 | 杭州华三通信技术有限公司 | Method and device for setting SDN flow entry |
CN105227481A (en) * | 2015-09-02 | 2016-01-06 | 重庆邮电大学 | The SDN congestion control method for routing of cost minimization is dispatched based on path cost and stream |
Non-Patent Citations (2)
Title |
---|
《数据中心网络多路径路由算法》;杨洋 等;《清华大学学报(自然科学版)》;20160315;第56卷(第3期);全文 |
《软件定义的数据中心网络中一种基于时隙传输的流量均衡方案设计》;杨洋 等;《中国科技论文在线》;20160201;第2页第1.1节至第7页第3.3节 |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112506648A (en) * | 2020-11-20 | 2021-03-16 | 鹏城实验室 | Traffic stateless migration method of virtual network function instance and electronic equipment |
Also Published As
Publication number | Publication date |
---|---|
CN105915467A (en) | 2016-08-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105915467B (en) | A kind of data center network flow equalization method and device that software-oriented defines | |
CN106027416B (en) | A kind of data center network traffic scheduling method and system based on space-time combination | |
Wang et al. | Adaptive path isolation for elephant and mice flows by exploiting path diversity in datacenters | |
CN106790104B (en) | IP communication and FC-AE-1553 communication means between multi-protocols emerging system, node | |
CN106063195B (en) | Control equipment discovery in network with independent control equipment and forwarding device | |
CN103346922B (en) | The controller of determination network state based on SDN and determine method | |
US20150200850A1 (en) | Communication system, control device, communication method, and program | |
CN110166366A (en) | Method for controlling network congestion, device and system | |
EP2652923B1 (en) | Communication path control system, path control device, communication path control method, and path control program | |
CN110166367A (en) | The method and device of flow is controlled in a kind of packet network | |
EP2637363A1 (en) | Communication system, control device, method for controlling packet transfer path, and program | |
CN112714023B (en) | TSN bandwidth reservation system, bandwidth reservation value calculation method and device | |
KR20040093492A (en) | Methods and apparatus for in order delivery of fibre channel frames | |
Zarifis et al. | DIBS: Just-in-time congestion mitigation for data centers | |
CN104734957A (en) | Service transmission method and device in software defined network (SDN) | |
CN105308920B (en) | Multicast Flows reorder scheme | |
CN107396396A (en) | Support the data transfer management method of multi-source multipath | |
CN105357124B (en) | A kind of MapReduce bandwidth optimization methods | |
CN102098596B (en) | Route establishing method and device in optical network | |
CN102546406A (en) | Network-on-chip routing centralized control system and device and adaptive routing control method | |
US9729449B2 (en) | Method and apparatus for enhanced routing within a shortest path based routed network containing local and long distance links | |
CN105207944B (en) | Transfer control method based on FAST TCP | |
CN114500354B (en) | Switch control method, device, control equipment and storage medium | |
Li et al. | Using indirect routing to recover from network traffic scheduling estimation error | |
Huang et al. | An enhanced scheduling framework for elephant flows in SDN-based data center networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |