CN103294521A - Method for reducing communication loads and energy consumption of data center - Google Patents

Method for reducing communication loads and energy consumption of data center Download PDF

Info

Publication number
CN103294521A
CN103294521A CN2013102116525A CN201310211652A CN103294521A CN 103294521 A CN103294521 A CN 103294521A CN 2013102116525 A CN2013102116525 A CN 2013102116525A CN 201310211652 A CN201310211652 A CN 201310211652A CN 103294521 A CN103294521 A CN 103294521A
Authority
CN
China
Prior art keywords
virtual machine
business
call duration
duration time
threshold value
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
Application number
CN2013102116525A
Other languages
Chinese (zh)
Other versions
CN103294521B (en
Inventor
杨挺
康春建
房朝晖
冯瑛敏
盆海波
吴成
向文平
袁博
林志贤
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin University
Original Assignee
Tianjin University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tianjin University filed Critical Tianjin University
Priority to CN201310211652.5A priority Critical patent/CN103294521B/en
Publication of CN103294521A publication Critical patent/CN103294521A/en
Application granted granted Critical
Publication of CN103294521B publication Critical patent/CN103294521B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The invention discloses a method for reducing communication loads and the energy consumption of a data center. The method includes enabling a dispatching center to compute an initial placement sequence of virtual machines according to an expenditure matrix and a service demand matrix; configuring service communication data to corresponding switchboards and links according to bandwidth constraint conditions and acquiring an initial service routing information table; judging whether an execution communication duration of a task carried by a current virtual machine exceeds a communication duration threshold value or not, and migrating the current virtual machine into a target physical machine logically closest to a target virtual machine if the execution communication duration of the task carried by the current virtual machine exceeds the communication duration threshold value; enabling the dispatching center to judge whether all the virtual machines carried with tasks with execution communication durations exceeding the communication duration threshold value are migrated or not, enabling the dispatching center to update the initial placement sequence of the virtual machines and the initial service routing information table if all the virtual machines carried with the tasks with the execution communication durations exceeding the communication duration threshold value are migrated, and closing the switchboards which are not used by the data center. The method has the advantages that network information dynamic perception and feedback regulation functions can be realized, services with the communication durations exceeding the duration threshold value are reasonably distributed, the usage amount of the switchboards is reduced, and the purpose of saving energy for the data center is achieved.

Description

A kind of method that reduces data center's traffic load and energy consumption
Technical field
The present invention relates to the Computer Applied Technology field, particularly a kind of method that reduces data center's traffic load and energy consumption.
Background technology
Under the background that have swept the globe in cloud computing, cloud computing industry development tide surges forward, the data center of cloud computing in recent years energy consumption becomes one of topic of academia and industrial community concern.Japan's Economy, Trade and Industry Ministery (METI) prediction, global IT energy consumption will be turned over 5 times in 2025, and will increase by 12 times to the year two thousand fifty.A large amount of energy consumptions makes picture Google, and the electricity charge in IT company every year that Microsoft and Facebook are such are just up to millions of dollars, and wherein the energy consumption of data center network equipment is also quite huge.Network equipment energy consumption in the data center occupies the 10%-20% of the whole energy consumptions of data center according to statistics.For example, spend up to 3,000,000,000 in the network equipment energy consumption of U.S. data center in 2006.
Compare with conventional data centers, the characteristic of cloud computing data center is mainly reflected in: (1) modular standard base facility.At the cloud service demand of data center, cloud data center carries out the modular arrangements design to facilities such as server, memory device, networks by industrial standard, makes it have adaptability and extensibility; (2) virtual resource and environment.Cloud data center extensively adopts Intel Virtualization Technology that physical resource assemble is formed a shared virtual resource pond, thereby more flexible and efficient, use resource at low cost; (3) but highly reliable automatic management. cloud computing data center should be the unattended telemanagement of 24X7, and realization equipment is to using unified management end to end.(4) expandability fast.Along with the variable diversity of big data explosion formula growth and user's request, cloud data center must come dynamic-configuration, order, supplies virtual resource according to service application demand and service quality, has the quick extended capability of the utilization of resources; (5) energy-conservation and saving space.Cloud computing data center will use energy-saving server, storage and the network equipment in a large number, and by advanced electric power system and heat dissipation technology, realize the seamless integrated of power supply, heat radiation and computational resource and management, solve the excessive refrigeration of conventional data centers and the problem of insufficient space.
In the prior art, for reducing network equipment energy consumption, many scientists have proposed multiple solution:
1) people such as Heller proposes network traffics are accumulated in the elasticity tree, closes the network equipment that is not used [1]Yet the method is not brought into play the advantage (virtual machine (vm) migration technology just) of the data center of cloud computing nowadays, does not take full advantage of the resource of cloud computing data center, and control mode is single.
2) people such as Meng has proposed a new problem at network equipment power saving---based on the virtual machine assignment problem (TVMPP) of flow [2]But its solution is not considered the link capacity constraint, may cause the virtual machine allocation scheme that contains congestion link is introduced data center.
3) people such as Shirayanagi has proposed a Honeguid solution [3]But the method need increase gap-bridging link between physical machine and switch, has increased data center's probability that breaks down.
Summary of the invention
The invention provides a kind of method that reduces data center's traffic load and energy consumption, by this method, realize reasonable distribution data center network flow, reduce traffic load, reduce the switch use amount, reach the energy-conservation target of data center, description sees below:
A kind of method that reduces data center's traffic load and energy consumption, described method comprises:
(1) sequence is tentatively placed by cost matrix and business demand matrix computations virtual machine by the dispatching center;
(2) according to the constraint condition of bandwidth, the service communication data configuration in corresponding switch and link, is obtained the initial service route information table;
(3) judge that current virtual machine carries out the call duration time of institute's bearer service and whether surpass the call duration time threshold value, if, with current virtual machine (vm) migration to purpose virtual machine in logic in the nearest target physical machine, execution in step (4) then; If not, execution in step (4);
(4) judge whether virtual machine that all carryings surpass call duration time threshold value business moves and finish the dispatching center, if, execution in step (5); If not, re-execute step (4);
(5) dispatching center upgrades described virtual machine and tentatively places sequence and described initial service route information table, closes the switch that is not used in the data center.
Tentatively place before the step of sequence by cost matrix and business demand matrix computations virtual machine described dispatching center, and described method also is included as:
Initialization data center, dispatching center obtains the computational fields internal information; The dispatching center determines described cost matrix between each physical machine by data center's topological structure, generates described business demand matrix automatically according to user's request.
The constraint condition of described bandwidth is specially:
Σ 1 ≤ i , j ≤ n R ij l ≤ α * C l
Wherein
Figure BDA00003276398600022
Be Business Stream from virtual machine i to the traffic rate of virtual machine j on link l, α is the constant that arranges according to RSVP, C lIt is the bandwidth of link l.
Describedly judge that current virtual machine carries out the call duration time of institute's bearer service and whether surpass before the step of call duration time threshold value, described method also comprises: the dispatching center determines described call duration time threshold value.
Described dispatching center determines that the process of described call duration time threshold value is specially:
The dispatching center adds up by the network operation historical data to the data central record, determines described call duration time threshold value.
In addition, described dispatching center determine described call duration time threshold value process can also for:
Switch detects each professional traffic rate and transfers to the dispatching center, the dispatching center determines the call duration time between each business, and obtaining the amount of communication data of business data flow by the business demand matrix, the dispatching center determines the call duration time threshold value according to the amount of communication data of the call duration time between each business and business data flow.
In addition, described dispatching center determine the call duration time threshold value process can also for:
Network information aware switch detects the call duration time between each professional traffic rate and definite each business, call duration time between each business is transferred to the dispatching center, the dispatching center obtains the amount of communication data of business data flow by the business demand matrix, and the dispatching center determines the call duration time threshold value according to the amount of communication data of the call duration time between each business and business data flow.
Described network information aware switch comprises: optical module or netting twine interface, power module, processor, exchange chip, control panel, readable flash memory, promoter region read-only memory and the dynamic RAM write also comprise: Information Statistics, analysis module.
" Information Statistics, analysis module " is for detection of the working time of each professional traffic rate and definite each business data flow, to transfer to working time of each business data flow processor, processor will transfer to working time of each business data flow the dispatching center by exchange chip and optical module or netting twine interface.
The beneficial effect of technical scheme provided by the invention is:
1) data center can monitor the more operational factor of network behind application the present invention, comprises the traffic rate between the business data flow, call duration time etc.Utilize the unnecessary computing power of switch, a part of dispatching center distribution of computation tasks can be carried out to switch, realize Distributed Calculation widely.
2) data center's energy consumption obviously reduces behind application the present invention.Because network operations information in the image data center can dynamically accurately be adjusted virtual machine and tentatively place sequence and initial service route information table, thereby close the switch that is not used in real time, reaches the energy-conservation purpose of data center.According to the U.S.'s 2006 annual data center power consumption data, calculate in conjunction with simulation result, use this method can make about 1.8 hundred million degree of the annual saves energy of U.S. data center.
3) data center network communication data packet delay can reduce about 1% behind application the present invention.Because this method is rationally moved the business of communication time overtime threshold value, thereby can make this part business data flow take still less switch and communication link, block up thereby reduced link, make the network service quality of data center to get a promotion.
4) data center network can be saved out more available bandwidth behind application the present invention.This method has dynamic sensing network information and feedback regulation function, the wasting of resources situation of having avoided the additive method static topology to cause.This method is carried out reasonable distribution with the business of communication time overtime threshold value, can take full advantage of data center's bottom bandwidth, take still less data center middle layer and top layer communication link bandwidth, make other professional communication paths have more selection, thereby save out more middle layer and top layer available bandwidth, further promote the network service quality of data center.
Description of drawings
Fig. 1 is the data center scheduling system reference architecture synoptic diagram of the present invention's suggestion;
Fig. 2 is a kind of method particular flow sheet that reduces data center's traffic load and energy consumption;
Fig. 3 is the synoptic diagram of virtual machine (vm) migration process described in the step 106;
Fig. 4 is network information aware switch block diagram;
Fig. 5 is the block diagram of Information Statistics analysis module;
Fig. 6 is the ethernet frame structure;
Fig. 7 a is average energy consumption synoptic diagram under each traffic rate;
Fig. 7 b is that the algorithm energy-saving effect compares synoptic diagram under each traffic rate;
Fig. 8 is packet average delay synoptic diagram;
Each layer link remaining bandwidth synoptic diagram of GRAPS algorithm and GRASP-JAVPRS algorithm when Fig. 9 a obeys (5,10,0,8) truncation normal distribution for the CBR traffic rate;
Each layer link remaining bandwidth synoptic diagram of SA algorithm and SA-JAVPRS algorithm when Fig. 9 b obeys (5,10,0,8) truncation normal distribution for the CBR traffic rate;
Each layer link remaining bandwidth synoptic diagram of GRAPS algorithm and GRASP-JAVPRS algorithm when Figure 10 a obeys (5,70,0,8) truncation normal distribution for the CBR traffic rate;
Each layer link remaining bandwidth synoptic diagram of SA algorithm and SA-JAVPRS algorithm when Figure 10 b obeys (5,70,0,8) truncation normal distribution for the CBR traffic rate.
Embodiment
For making the purpose, technical solutions and advantages of the present invention clearer, embodiment of the present invention is described further in detail below in conjunction with accompanying drawing.
For reasonable distribution data center network flow, reduce traffic load, reduce the switch use amount, reach the energy-conservation target of data center, the embodiment of the invention provides a kind of method that reduces data center's traffic load and energy consumption, the present invention is mainly by monitoring data central site network data stream, utilize virtual machine (VM) migrating technology, the virtual machine of carrying cloud computing business is redistributed in physical machine, thereby close untapped switch, reach data center's purpose of energy saving, wherein, the switch among embodiment 1 and the embodiment 2 is switch known in those skilled in the art, and the switch among the embodiment 3 is the designed network information aware switch of the present invention, the operation steps of embodiment 1 and embodiment 2 also can adopt the designed network information aware switch of the present invention during specific implementation, sees for details hereinafter to describe:
Embodiment 1
101: initialization data center, dispatching center, obtain the computational fields internal information;
Wherein, various information in the computational fields generally includes: data center's topological structure, physical machine number, switch number and each link bandwidth, physical machine is used for carrying the virtual machine (the corresponding virtual machine collection of each physical machine, virtual machine concentrate the number of virtual machine to be determined by the load-bearing capacity of physical machine) of calculating and communication task; It is unobstructed that switch and link bandwidth information are used for guaranteeing communicate by letter, and determine path for the planning of route, and the occurrence of those information is definite by the dispatching center.
Dispatching center described in the present invention is for finishing the large server etc. of scheduler task, and the model embodiment of the invention of above-mentioned physical machine and switch does not limit this, all can as long as can finish the device of above-mentioned functions.
102: the dispatching center determines cost matrix D between each physical machine to generate the business demand matrix F automatically according to user's request by data center's topological structure;
Wherein, each element D in the cost matrix D QpExpression, D QpFor from physical machine q to the physical machine p communication needed switch number.The cost matrix D has different values in different network structures.This method has defined the mathematic(al) representation of fat tree (Fat-tree) structure cost matrix D, in fat tree construction, and each element D of cost matrix QpBe the function of each switch ports themselves number k value, see formula 1.
D qp Fat - tree = 0 if | p - q | = 0 1 if 0 < | p - q | &le; k 2 - 1 3 if k 2 - 1 < | p - q | &le; ( k 2 ) 2 - 1 5 if ( k 2 ) 2 - 1 < | p - q | - - - ( 1 )
Wherein, p, the value of q and k is the positive integer more than or equal to 1, the dimension of cost matrix D determines that by the number of physical machine the physical machine number is the function of each switch ports themselves number, namely For example: when each switch ports themselves number was 4, the physical machine number was that the value of 16, q and p is 1 and 2 o'clock, because
Figure BDA00003276398600053
So D 12=1; The value of q and p is 1 and 4 o'clock, because
Figure BDA00003276398600054
So D 14=3.When each switch ports themselves number was 4, the physical machine number was 16, and the cost matrix D of fat tree topology is following formula:
0 1 3 3 5 5 5 5 5 5 5 5 5 5 5 5 1 0 3 3 5 5 5 5 5 5 5 5 5 5 5 5 3 3 0 1 5 5 5 5 5 5 5 5 5 5 5 5 3 3 1 0 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 0 1 3 3 5 5 5 5 5 5 5 5 5 5 5 5 1 0 3 3 5 5 5 5 5 5 5 5 5 5 5 5 3 3 0 1 5 5 5 5 5 5 5 5 5 5 5 5 3 3 1 0 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 0 1 3 3 5 5 5 5 5 5 5 5 5 5 5 5 1 0 3 3 5 5 5 5 5 5 5 5 5 5 5 5 3 3 0 1 5 5 5 5 5 5 5 5 5 5 5 5 3 3 1 0 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 0 1 3 3 5 5 5 5 5 5 5 5 5 5 5 5 1 0 3 3 5 5 5 5 5 5 5 5 5 5 5 5 3 3 0 1 5 5 5 5 5 5 5 5 5 5 5 5 3 3 1 0
Wherein, each element F in the business demand matrix F IjExpression, F IjExpression from virtual machine i to the service communication amount the virtual machine j.Do not consider the Energy Saving Control strategy of physical machine in the present invention, thereby after step 101 obtains available physical machine numbering, the various types of communication business integration gone into virtual machine concentrate that a virtual machine collection S represents the set of a plurality of virtual machines, i.e. S={s 1, s 2..., s n.
Because a virtual machine collection is distributed in the physical machine, reduce network node quantity problem and can be summarized as quadratic assignment problem (QAP), mathematical formulae is seen formula 2.
min ( &Sigma; i , j = 1 n &Sigma; q , p = 1 n f ij d qp x iq x jp )
Obey:
&Sigma; i = 1 n x iq = 1 1≤q≤n
&Sigma; q = 1 n x iq = 1 1≤i≤n
(2)
&Sigma; j = 1 n x jp = 1 1≤p≤n
&Sigma; p = 1 n x jp = 1 1≤j≤n
x iq∈{0,1} 1≤i,q≤n
x jp∈{0,1} 1≤j,p≤n
N represents available physical machine quantity, f in formula 2 IjService communication amount between expression virtual machine i and the virtual machine j is an element of business demand matrix F; d QpBeing the cost between physical machine q and the physical machine p, is an element of cost matrix D; x Iq{ 0,1} represents x to ∈ IjValue is 0 or 1.x IqGet 1 expression virtual machine i and be placed among the physical machine q, get 0 expression virtual machine i and be not placed among the physical machine q.Equally, x JpGet 1 expression virtual machine j and be placed among the physical machine p, get 0 expression virtual machine j and be not placed among the physical machine p;
Figure BDA00003276398600076
Among 1≤q≤n, be expressed as virtual machine collection of a physical machine intelligence carrying, for example: when q=1
Figure BDA00003276398600077
When q=2 When q=3
Figure BDA00003276398600079
Figure BDA000032763986000710
Among 1≤i≤n, be expressed as a virtual machine collection and can only be placed in the physical machine, for example: when i=1 &Sigma; q = 1 n x 1 q = 1 , When i=2 &Sigma; q = 1 n x 2 q = 1 , When i=3 &Sigma; q = 1 n x 3 q = 1 ; Objective function min (
Figure BDA000032763986000713
) be illustrated in to ask in all virtual machine placement sequences and use the minimum virtual machine of the network equipment to put in order.
103: calculate virtual machine by cost matrix D and business demand matrix F and tentatively place sequence the dispatching center;
In order to reduce network equipment use amount, the present invention adopts greedy random adaptation searching method (GRASP) in the virtual machine distribution method of a virtual machine collection of a physical machine carrying [4]And simulated annealing (SA) [5]Algorithm comes the solving virtual machine tentatively to place sequence.The concrete operations of this step, for conventionally known to one of skill in the art, the embodiment of the invention is not done this and is given unnecessary details.
104: according to the constraint condition of bandwidth, the service communication data configuration in corresponding switch and link, is obtained the initial service route information table;
When virtual machine in physical machine preliminary place finish after, initial routing iinformation professional in each virtual machine is: from source virtual machine (business data flow send end) via the link of the leftmost side of data center network and switch to the purpose virtual machine, the information that has professional corresponding link and switch is stored to the initial service route information table.Having new service communication data to add fashionablely is placed into it in link of having enabled and switch, when enabled link bandwidth reaches the defined bandwidth of RSVP, new service communication data are selected the link and the switch that are not used at random, and the new employed link of service communication data and switch is added in the initial service route information table.
The present invention preferentially is placed into the service communication data in the link of having enabled and switch, closes not to be used link, switch and switch ports themselves.Must satisfy the defined bandwidth of RSVP when Route Selection, namely formula 3, thereby guaranteed the service quality of the unobstructed and network of link, and virtual machine configuration and the method for routing having avoided comprising congestion link are introduced data center.
&Sigma; 1 &le; i , j &le; n R ij l &le; &alpha; * C l - - - ( 3 )
Wherein
Figure BDA00003276398600082
Be Business Stream from virtual machine i to the traffic rate of virtual machine j on link l, α is the constant that arranges according to RSVP, C lIt is the bandwidth of link l.Wherein, RSVP is conventionally known to one of skill in the art, and the embodiment of the invention is not done this and given unnecessary details.
105: the dispatching center adds up by the network operation historical data to the data central record, determines call duration time threshold value (TH);
Business data flow call duration time threshold T H is core parameter of the present invention, the dispatching center has ubiquity and stable business data flow rule working time by the network operation historical data of data central record being added up draw, and is formula 4 to the data statistic analysis formula.
Flow v≤Flow v+1 v∈(1,N)
&Sigma; u = 1 M Flow u &le; &Sigma; v = 1 N Flow v 2 M≤N (4)
max{T(Flow u)} u∈(1,M)
In the network operation historical data of data center's record, business data flow working time and Business Stream have corresponding relation.Flow in the formula 4 vThe amount of communication data of representing the v business data flow has N bar data stream altogether.Flow v≤ Flow V+1V ∈ (1, N) sort from small to large to the traffic of all data stream by expression.
Figure BDA00003276398600084
M≤N represents to be less than or equal to from the traffic summation of No. 1 to M number business data flow half of whole business data flow traffics.The M value has different values for different statisticss, can find the solution from formula 4 to obtain.Publicity max{T (Flow u) (1, M) maximal value of Business Stream working time of satisfying formula 4 preceding two constraint conditions is asked in expression to u ∈, and this value is exactly call duration time threshold T H, T (Flow in the formula 4 u) working time of expression u business data flow.
106: communicate between current virtual machine and the purpose virtual machine, judge that current virtual machine carries out the call duration time of institute's bearer service and whether surpass the call duration time threshold value, if, with current virtual machine (vm) migration to purpose virtual machine in logic in the nearest target physical machine, execution in step 107 then; If not, execution in step 107;
Wherein, logic refers to recently between host's physical machine of target physical machine and purpose virtual machine and can be communicated with by less switch, and can keep smooth and easy communication under the link congestion condition satisfying after institute's loaded service migration.Determine the call duration time threshold value and will carry the method that the virtual machine that surpasses call duration time threshold value business moves to be called the JAVPRS algorithm.The GRASP-JAVPRS algorithm is: at first utilizes GRASP algorithm solving virtual machine tentatively to place sequence, and then according to call duration time threshold value and link bandwidth constraint condition virtual machine moved, and the professional routing iinformation behind the calculating virtual machine (vm) migration.The GRASP algorithm does not communicate the time gate limit value and determines and virtual machine (vm) migration that other steps are identical with the GRASP-JAVPRS algorithm.SA algorithm and SA-JAVPRS algorithm repeat no more.
For example: referring to Fig. 3, S1, S2, S3 etc. represent the different switch of sequence number (Switch) respectively, L11, L12, L22 etc. represent the different link of sequence number (Link) respectively, virtual machine A is configured in the physical machine 1, virtual machine B is configured in the physical machine 14, virtual machine C is configured in the physical machine 2, and virtual machine D is configured in the physical machine 4.Virtual machine A communicates by letter with virtual machine B, and call duration time is 30 seconds; Virtual machine C communicates by letter with virtual machine D, and call duration time is 10 seconds; The call duration time threshold value is 25 seconds.According to this method, the call duration time between virtual machine C and the virtual machine D does not surpass the call duration time threshold value, thus do not need to move, wherein, routing iinformation between virtual machine C and the virtual machine D is: { physical machine 2, L12, S1, L13, S3, L23, S2, physical machine 4).Call duration time surpasses the call duration time threshold value between virtual machine A and the virtual machine B, virtual machine A need be moved.If virtual machine A is migrated to physical machine 13, judge that can physical machine 13 and service communication bandwidth between the physical machine 14 satisfy formula 3, as if satisfying formula 3, then physical machine 13 is purpose virtual machine B nearest physical machine in logic; If do not satisfy formula 3, then if virtual machine A is migrated to physical machine 15, judge that can physical machine 15 and service communication bandwidth between the physical machine 14 satisfy formula 3, as if satisfying formula 3, then physical machine 15 is purpose virtual machine B nearest physical machine in logic; If do not satisfy formula 3, then judge that with identical method can virtual machine A is migrated to physical machine 16 satisfy formula 3; If physical machine 16 can not satisfy condition, then virtual machine is not moved.
Primary routing information is between virtual machine A and the virtual machine B:
{ physical machine 1, L11, S1, L13, S3, L35, S5, L59, S9, L91, S11, L41, physical machine 14}; Suppose that by top judgement physical machine 13 is purpose virtual machine B nearest physical machine in logic, then the routing iinformation between virtual machine A and the virtual machine B is: { physical machine 13, L31, S11, L41, physical machine 14}.
107: the dispatching center judges whether virtual machine that all carryings surpass call duration time threshold value business moves and finishes, if, execution in step 108; If not, re-execute step 106;
In the practical application, finish each bearer service corresponding deadline all, the dispatching center compared deadline and call duration time threshold value, when existing the deadline to surpass the call duration time threshold value, to corresponding virtual machine execution in step 106 of deadline, virtual machine is moved.Realize the monitoring of dispatching center by this step, reduced the usage quantity of switch.
108: the dispatching center upgrades virtual machine and tentatively places sequence and initial service route information table, closes the switch that is not used in the data center.
Embodiment 2
201: initialization data center, dispatching center, obtain the computational fields internal information;
Wherein, various information in the computational fields generally includes: data center's topological structure, physical machine number, switch number and each link bandwidth, physical machine is used for carrying the virtual machine (the corresponding virtual machine collection of each physical machine, virtual machine concentrate the number of virtual machine to be determined by the load-bearing capacity of physical machine) of calculating and communication task; It is unobstructed that switch and link bandwidth information are used for guaranteeing communicate by letter, and determine path for the planning of route, and the occurrence of those information is definite by the dispatching center.
Dispatching center described in the embodiment of the invention is for finishing the large server etc. of scheduler task, and the model embodiment of the invention of above-mentioned physical machine and switch does not limit this, all can as long as can finish the device of above-mentioned functions.
202: the dispatching center determines cost matrix D between each physical machine to generate the business demand matrix F automatically according to user's request by data center's topological structure;
Wherein, the detail operations of this step sees embodiment 1 for details, and present embodiment is not done this and given unnecessary details.
203: calculate virtual machine by cost matrix D and business demand matrix F and tentatively place sequence the dispatching center;
In order to reduce network equipment use amount, the present invention adopts greedy random adaptation searching method (GRASP) in the virtual machine distribution method of a virtual machine collection of a physical machine carrying [4]And simulated annealing (SA) [5]Algorithm comes the solving virtual machine tentatively to place sequence.The concrete operations of this step, for conventionally known to one of skill in the art, the embodiment of the invention is not done this and is given unnecessary details.
204: according to the constraint condition of bandwidth, the service communication data configuration in corresponding switch and link, is obtained the initial service route information table;
The present invention preferentially is placed into the service communication data in the link of having enabled and switch, closes not to be used link, switch and switch ports themselves.When Route Selection, must satisfy the defined bandwidth of RSVP, i.e. formula 3 among the embodiment 1, thus having guaranteed the service quality of the unobstructed and network of link, virtual machine configuration and the method for routing having avoided comprising congestion link are introduced data center.The detail operations of this step is referring to embodiment 1, and present embodiment is not done this and given unnecessary details.
205: switch detects each professional traffic rate and transfers to the dispatching center, and the dispatching center determines the call duration time T between each business Er, and obtain the amount of communication data Flow of business data flow by the business demand matrix F v, the dispatching center determines call duration time threshold value (TH) according to the amount of communication data of the call duration time between each business and business data flow;
By each the element R among detected each the traffic communications rate matrix R(matrix R of switch ErRepresent the traffic rate that each is professional), again in conjunction with the business demand matrix F, calculate the call duration time of each Business Stream under non-congestion situation, see computing formula (5).
T er=F er/R er (5)
T in the formula 5 ErRepresent the call duration time between professional e and the professional r, F ErElement for the business demand matrix F.With the call duration time T between each business ErRunning time T (Flow as business data flow v), can obtain business data flow call duration time threshold T H according to formula (4) again.
206: communicate between current virtual machine and the purpose virtual machine, judge that current virtual machine carries out the call duration time of institute's bearer service and whether surpass the call duration time threshold value, if, with current virtual machine (vm) migration to purpose virtual machine in logic in the nearest target physical machine, execution in step 207 then; If not, execution in step 207;
Wherein, logic refers to recently between host's physical machine of target physical machine and purpose virtual machine and can be communicated with by less switch, and can keep smooth and easy communication after institute's loaded service migration under the link congestion condition.
207: the dispatching center judges whether virtual machine that all carryings surpass call duration time threshold value business moves and finishes, if, execution in step 208; If not, re-execute step 206;
In the practical application, finish each bearer service corresponding deadline all, the dispatching center compared deadline and call duration time threshold value, when existing the deadline to surpass the call duration time threshold value, to corresponding virtual machine execution in step 206 of deadline, virtual machine is moved.Realize the monitoring of dispatching center by this step, reduced the usage quantity of switch.
208: the dispatching center upgrades virtual machine and tentatively places sequence and initial service route information table, closes the switch that is not used in the data center.
Wherein, the execution in step among the embodiment 2 is except specified otherwise, and detailed operation steps is all identical with embodiment 1, please refer to the operation steps of embodiment 1, and present embodiment is not done this and given unnecessary details.
Embodiment 3
The embodiment of the invention also comprises to be improved switch, referring to Fig. 4, switch generally includes: optical module or netting twine interface (adopting the RJ45 socket usually), power module, processor, exchange chip, control panel, the readable flash memory of writing, promoter region read-only memory and dynamic RAM, network information aware switch in the present embodiment also comprises: Information Statistics, analysis module, this module is for detection of the working time of each professional traffic rate and definite each business data flow, processor will be transferred to working time of each business data flow, processor transfers to the dispatching center by exchange chip and optical module or netting twine interface with working time of each business data flow, between each module in the above-mentioned switch by being electrically connected.
During specific implementation, Information Statistics, analysis module can be FPGA, and the embodiment of the invention is that example describes with the EP1C12F256C6N model FPGA that adopts altera corp to produce, and can also be the FPGA of other models, and the embodiment of the invention does not limit this.
301: initialization data center, dispatching center, obtain the computational fields internal information;
Wherein, various information in the computational fields generally includes: data center's topological structure, physical machine number, network information aware switch number and each link bandwidth, physical machine is used for carrying the virtual machine (the corresponding virtual machine collection of each physical machine, virtual machine concentrate the number of virtual machine to be determined by the load-bearing capacity of physical machine) of calculating and communication task; It is unobstructed that network information aware switch and link bandwidth information are used for guaranteeing communicate by letter, and determine path for the planning of route, and the occurrence of those information is definite by the dispatching center.
302: the dispatching center determines cost matrix D between each physical machine to generate the business demand matrix F automatically according to user's request by data center's topological structure;
303: calculate virtual machine by cost matrix D and business demand matrix F and tentatively place sequence the dispatching center;
304: according to the constraint condition of bandwidth, the service communication data configuration in corresponding network information aware switch and link, is obtained the initial service route information table;
305: network information aware switch detects the call duration time T between each professional traffic rate and definite each business Er, with the call duration time T between each business ErTransfer to the dispatching center, the dispatching center obtains the amount of communication data Flow of business data flow by the business demand matrix F v, the dispatching center is according to the call duration time T between each business ErDetermine call duration time threshold value (TH) with the amount of communication data of business data flow;
The ethernet frame structure is seen Fig. 6,
PRE: guide's byte, 7 10101010
SFD: start of frame delimiter, 10101011
DA: target MAC (Media Access Control) address SS: source MAC
L/T: frame length (value＜=1500)/type (value〉1500)
DATA: data field
PAD: fill field
CRC: check field
" Information Statistics, analysis module " analyzes the packet of collecting in the unit interval from SS to DA, just the data flow communication speed from SS to DA is deposited to a two-dimensional array
Message={SS 1,DA 1,R er(1);SS 2,DA 2,R er(2);......;SS w,DA w,R er(w)}
SS wherein 1The source MAC of expression article one business data flow, its physical machine is numbered traffic rate R just ErIn e; DA 1The target MAC (Media Access Control) address of expression article one business data flow, its physical machine is numbered traffic rate R just ErIn r, R Er(1) expression is from SS 1To DA 1Traffic rate (implication of other characters is similar, and present embodiment is not done this and given unnecessary details), w represents that statistical formula is formula 6 through this switch w bar business data flow altogether.
R er ( w ) = &Sigma; t = 0 TD [ Bytee ( PR E t ( w ) ) + Bytee ( SF D t ( w ) ) + Bytee ( D A t ( w ) ) ] TD +
&Sigma; t = 0 TD [ Bytee ( SS t ( w ) ) + Bytee ( L T t ( w ) ) + Bytee ( DAT A t ( w ) ) + Bytee ( CR C t ( w ) ) ] TD
(6)
TD represents that " Information Statistics, analysis module " the past, once the statistics end was to the time period between this statistics end in the formula 6, and Bytee () is the function of each several part byte number in the calculating ethernet frame, for example Bytee (PRE t(w)) being illustrated in t guide's byte number in the ethernet frame of professional w constantly, is 7 bytes, is a fixed value.The molecule sum of formula 6 is represented the total amount of byte that the statistics switch receives/sends in the TD time period.The detail operations of this step is referring to embodiment 2, and present embodiment is not done this and given unnecessary details.
306: communicate between current virtual machine and the purpose virtual machine, judge that current virtual machine carries out the call duration time of institute's bearer service and whether surpass the call duration time threshold value, if, with current virtual machine (vm) migration to purpose virtual machine in logic in the nearest target physical machine, execution in step 307 then; If not, execution in step 307;
307: the dispatching center judges whether virtual machine that all carryings surpass call duration time threshold value business moves and finishes, if, execution in step 308; If not, re-execute step 306;
308: the dispatching center upgrades virtual machine and tentatively places sequence and initial service route information table, closes the network information aware switch that is not used in the data center.
Wherein, the execution in step among the embodiment 3 is except specified otherwise, and detailed operation steps is all identical with embodiment 1, please refer to the operation steps of embodiment 1, and present embodiment is not done this and given unnecessary details.
Verify a kind of feasibility that reduces the method for data center's traffic load and energy consumption provided by the invention with concrete experiment below, see for details hereinafter and describe:
Adopt 6 port Fat-Tree topologys as the network structure of data center in emulation experiment, the physical machine number is the function that each switch ports themselves is counted the k value, i.e. k 3/ 4, thus the physical machine number is 54; Adopt Montage science calculation workflow to come each service communication amount of initialization, Montage[3] be to be created by the instrument of increasing income of NASA/IPAC, can be produced the mosaic picture of starry sky by the picture of input FITS form, a Montage example sees Table 1; The switch number of use from the source to the purpose defines cost matrix D (formula (1)); Use four kinds of algorithm: GRASP, GRASP-JAVPRS, SA and SA-JAVPRS, four kinds of algorithms have identical routing policy, and wherein GRASP and SA only find the solution the QAP problem, no longer readjust the virtual machine assigned sequence; And GRASP-JAVPRS and SA-JAVPRS find the solution the QAP problem earlier, according to this method the virtual machine assigned sequence are adjusted then; The Business Stream generator adopts CBR, and CBR speed is used three kinds of truncation normal distribution data (5,01,0,8), (5,10,0,8), (5,70,0,8) simulate the data stream of three kinds of different qualities, wherein (5,01,0,8) expression CBR speed obedience average is 5, and variance is 1, minimum value is greater than 0, and maximal value is no more than 8 normal distribution (implication of character by that analogy in other two normal distribution data), and each layer switch energy consumption sees Table 2.
Table 1
Figure BDA00003276398600141
Each layer switch each several part energy consumption of table 2
The switch level Frame (watt) Ply-yarn drill (watt) Each port (watt)
Access Layer 146 0 0.42
Convergence-level 1558 1212 27
Core layer 1558 1212 27
This method is from energy consumption, time-delay and 3 aspects of remaining bandwidth are estimated, wherein, Fig. 7 (a) is the mean value of data center's energy consumption test findings under three groups of same communication rate distribution, as can be seen from the figure under identical QAP algorithm situation, network equipment energy consumption obviously reduces after using the JAVPRS algorithm, and the increase along with data flow communication speed variance, getting more and more of energy saving, for example: obey (5,10,0 at the CBR traffic rate, 8) during the truncation normal distribution, when GRASP algorithm energy consumption is 140.39 watts of *, when GRASP-JAVPRS algorithm energy consumption is 130.13 watts of *, when the GRASP-JAVPRS algorithm is saved 140.39-130.13=10.26 watt of * than GRASP algorithm; Obey (5,70,0 at the CBR traffic rate, 8) during the truncation normal distribution, when GRASP algorithm energy consumption is 213.20 watts of *, when GRASP-JAVPRS algorithm energy consumption is 124.73 watts of *, when the GRASP-JAVPRS algorithm is saved 213.20-124.73=88.47 watt of * than GRASP algorithm; Be when being the 10 o'clock about 88.47-10.26=78.21 of more piece watt of * than CBR traffic rate variance at 70 o'clock in CBR traffic rate variance.
Fig. 7 (b) uses the JAVPRS algorithm than the chart of percentage comparison of primal algorithm (GRASP or SA) the about electric energy of more piece under each traffic rate of CBR.From Fig. 7 (b) as can be seen, save (140.39-130.13)/140.39*100%=7.3% than GRASP algorithm when GRASP-JAVPRS algorithm when CBR traffic rate variance is 10; From Fig. 7 (a) as can be seen, be that 10 o'clock SA algorithm energy consumptions are in CBR traffic rate variance: during 190.74 watts of *, when SA-JAVPRS algorithm energy consumption was 184.88 watts of *, the SA-JAVPRS algorithm was saved (190.74-184.88)/190.74*100%=3.1% than SA algorithm; From Fig. 7 (a) as can be seen, it is 70 o'clock in CBR traffic rate variance, when GRASP algorithm energy consumption is 213.20 watts of *, when GRASP-JAVPRS algorithm energy consumption is 124.73 watts of *, GRASP-JAVPRS saves (213.20-124.73)/213.20*100%=41.5% than GRASP algorithm, when SA algorithm energy consumption was 196.00 watts of * simultaneously, when SA-JAVPRS algorithm energy consumption was 164.53 watts of *, the SA-JAVPRS algorithm was saved (196.00-164.53)/196.00*100%=16.1% than SA algorithm.This just illustrates that service communication speed variance is more big, and just certain service communication time is departed from average when big, and JAVPRS algorithm energy-saving effect is more obvious.This be because the JAVPRS algorithm with the business reorganization of call duration time overtime threshold value in suitable virtual machine, most of target virtual machine and purpose virtual machine are connected on the same access switch, thereby make this part business no longer take too much switch, and access switch is longer service time, thereby this part business only increases a spot of energy consumption.
Time-delay is defined as from the generation of packet and arrives the time interval of destination to it.It not only comprises the transmission time, is also included within the formation time-delay in the switch buffer memory.Buffer memory time-delay is subjected to mainly that packet is congested to be influenced.As can be seen from Figure 8, increase along with the traffic rate variance, use more that the packet time-delay reduces behind this algorithm, for example: be 10 o'clock in CBR traffic rate variance, GRASP algorithm data packet delay is 0.059 second, GRASP-JAVPRS algorithm data packet delay is 0.056 second, and the packet time-delay reduces (0.059-0.056)/0.059*100%=5.1% after application this method; It is 70 o'clock in CBR traffic rate variance, GRASP algorithm data packet delay is 0.059 second, GRASP-JAVPRS algorithm data packet delay is 0.055 second, the packet time-delay reduces (0.059-0.055)/0.059*100%=6.8% after using this method, the GRASP-JAVPRS algorithm CBR traffic rate variance be 70 o'clock be the packet time-delay reduction of Duoing in 10 o'clock than variance:
[(0.059-0.055)-(0.059-0.056)]/(0.059-0.056)*100%=33.3%。Illustrate that the JAVPRS algorithm effectively reduces network congestion.
Remaining bandwidth refers to available bandwidth in the occupied communication link.Definition bottom link is the communication link between server and the access switch, and intermediate links is the communication link between access switch and the convergence switch, and the top layer link is the communication link between convergence switch and the core.Each layer link remaining bandwidth that emulation obtained when Fig. 9, Figure 10 were respectively (5,10,0,8) and (5,70,0,8) the truncation normal distribution of data flow communication speed obedience.Transverse axis is the simulation run time, is unit with the second.The longitudinal axis is remaining bandwidth number percent.
Data among Fig. 9 and Figure 10 are each layer link remaining bandwidth number percent when the time being 3 seconds, data are distinguished the bottom link behind the corresponding JAVPRS of the application algorithm from top to bottom, the bottom link of primal algorithm (GRASP or SA), intermediate links behind the application JAVPRS algorithm, the intermediate links of primal algorithm, top layer link behind the application JAVPRS algorithm, the top layer link of primal algorithm.From Fig. 9 (a) as can be seen, GRASP-JAVPRS algorithm top layer link remaining bandwidth has more the remaining bandwidth of 50.81%-43.20%=7.61% than GRASP algorithm top layer link remaining bandwidth; GRASP-JAVPRS algorithm intermediate links remaining bandwidth has more the remaining bandwidth of 66.86%-63.68%=3.18% than GRASP algorithm intermediate links remaining bandwidth; GRASP-JAVPRS algorithm bottom link remaining bandwidth has more the remaining bandwidth of 70.65%-70.48%=0.17% than GRASP algorithm bottom link remaining bandwidth; From Figure 10 (b) as can be seen, when the time was 3 seconds, SA-JAVPRS algorithm top layer link remaining bandwidth had more the remaining bandwidth of 50.86%-39.12%=11.74% than SA algorithm top layer link remaining bandwidth; SA-JAVPRS algorithm intermediate links remaining bandwidth has more the remaining bandwidth of 65.41%-62.36%=3.05% than SA algorithm intermediate links remaining bandwidth; SA-JAVPRS algorithm bottom link remaining bandwidth has more the remaining bandwidth of 70.57%-70.13%=0.44% than SA algorithm bottom link remaining bandwidth.Identical rule can draw from Fig. 9 (b) and Figure 10 (a).
No matter be GRASP-JAVPRS algorithm or SA-JAVPRS algorithm as can be seen from Fig. 9, Figure 10, no matter be bottom or middle layer or top layer, use the JAVPRS algorithms in most times and all obtain more remaining bandwidth when not using the JAVPRS algorithm, this is because the JAVPRS algorithm is inserted in the long still little business of traffic rate of call duration time in the link of operation, improve link bandwidth utilization rate on the one hand, saved out available link for the big business of traffic rate on the other hand.From middle layer and top layer as can be seen the JAVPRS algorithm have clear superiority aspect the link bandwidth utilization rate improving.
List of references
[1]Heller,B.,Seetharaman,S.,Mahadevan,P.,Yiakoumis,Y.,Sharma,P.,Banerjee,S.,McKeown,N.:ElasticTree:Saving Energy inDataCenter Networks.In USENIX NSDI,April2010
[2]X.Meng,V.Pappas,and L.Zhang.:Improving the scalability of data center networks with traffic-aware virtual machine placement.In Proceedings of the29th IEEE Conference on Information Communications(INFOCOM’10),pp.1154-1162,2010
[3]Shirayanagi,H.,Yamada,H.,Kono,K.:Honeyguide:A VM migration-aware network topology for saving energy consumption in data center networks.In IEEE Symposium on Computers and Communications,pp.460-467,2012
[4]Oliveira,C.,Pardalos,P.,Resende,M.:GRASP with path-relinking for the quadratic assignment problem.In3rd International Workshop on Experimental and Efficient Algorithms,pp.356-368,2004
[5]Misevicius,A.:A modified simulated annealing algorithm for the quadratic assignment problem.In INFORMATICA,vol.14,pp497-514,2003
It will be appreciated by those skilled in the art that accompanying drawing is the synoptic diagram of a preferred embodiment, the invention described above embodiment sequence number does not represent the quality of embodiment just to description.
The above only is preferred embodiment of the present invention, and is in order to limit the present invention, within the spirit and principles in the present invention not all, any modification of doing, is equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (8)

1. a method that reduces data center's traffic load and energy consumption is characterized in that, described method comprises:
(1) sequence is tentatively placed by cost matrix and business demand matrix computations virtual machine by the dispatching center;
(2) according to the constraint condition of bandwidth, the service communication data configuration in corresponding switch and link, is obtained the initial service route information table;
(3) judge that current virtual machine carries out the call duration time of institute's bearer service and whether surpass the call duration time threshold value, if, with current virtual machine (vm) migration to purpose virtual machine in logic in the nearest target physical machine, execution in step (4) then; If not, execution in step (4);
(4) judge whether virtual machine that all carryings surpass call duration time threshold value business moves and finish the dispatching center, if, execution in step (5); If not, re-execute step (4);
(5) dispatching center upgrades described virtual machine and tentatively places sequence and described initial service route information table, closes the switch that is not used in the data center.
2. a kind of method that reduces data center's traffic load and energy consumption according to claim 1 is characterized in that, tentatively place before the step of sequence by cost matrix and business demand matrix computations virtual machine described dispatching center, and described method also is included as:
Initialization data center, dispatching center obtains the computational fields internal information; The dispatching center determines described cost matrix between each physical machine by data center's topological structure, generates described business demand matrix automatically according to user's request.
3. a kind of method that reduces data center's traffic load and energy consumption according to claim 1 is characterized in that the constraint condition of described bandwidth is specially:
&Sigma; 1 &le; i , j &le; n R ij l &le; &alpha; * C l
Wherein
Figure FDA00003276398500012
Be Business Stream from virtual machine i to the traffic rate of virtual machine j on link l, α is the constant that arranges according to RSVP, C lIt is the bandwidth of link l.
4. a kind of method that reduces data center's traffic load and energy consumption according to claim 1, it is characterized in that, describedly judge that current virtual machine carries out the call duration time of institute's bearer service and whether surpass before the step of call duration time threshold value, described method also comprises: the dispatching center determines described call duration time threshold value.
5. a kind of method that reduces data center's traffic load and energy consumption according to claim 4 is characterized in that, described dispatching center determines that the process of described call duration time threshold value is specially:
The dispatching center adds up by the network operation historical data to the data central record, determines described call duration time threshold value.
6. a kind of method that reduces data center's traffic load and energy consumption according to claim 4 is characterized in that, described dispatching center determines that the process of described call duration time threshold value is specially:
Switch detects each professional traffic rate and transfers to the dispatching center, the dispatching center determines the call duration time between each business, and obtaining the amount of communication data of business data flow by described business demand matrix, the dispatching center determines described call duration time threshold value according to the amount of communication data of the call duration time between described each business and described business data flow.
7. a kind of method that reduces data center's traffic load and energy consumption according to claim 4 is characterized in that, described dispatching center determines that the process of described call duration time threshold value is specially:
Network information aware switch detects the call duration time between each professional traffic rate and definite each business, call duration time between described each business is transferred to the dispatching center, the dispatching center obtains the amount of communication data of business data flow by described business demand matrix, and the dispatching center determines described call duration time threshold value according to the amount of communication data of the call duration time between described each business and described business data flow.
8. a kind of method that reduces data center's traffic load and energy consumption according to claim 7, described network information aware switch comprises: optical module or netting twine interface, power module, processor, exchange chip, control panel, readable flash memory, promoter region read-only memory and the dynamic RAM write, it is characterized in that, also comprise: Information Statistics, analysis module
Described Information Statistics, analysis module are also determined the working time of each business data flow for detection of each professional traffic rate, to transfer to working time of each business data flow processor, processor will transfer to working time of each business data flow the dispatching center by exchange chip and optical module or netting twine interface.
CN201310211652.5A 2013-05-30 2013-05-30 A kind of method reducing data center's traffic load and energy consumption Active CN103294521B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201310211652.5A CN103294521B (en) 2013-05-30 2013-05-30 A kind of method reducing data center's traffic load and energy consumption

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201310211652.5A CN103294521B (en) 2013-05-30 2013-05-30 A kind of method reducing data center's traffic load and energy consumption

Publications (2)

Publication Number Publication Date
CN103294521A true CN103294521A (en) 2013-09-11
CN103294521B CN103294521B (en) 2016-08-10

Family

ID=49095456

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201310211652.5A Active CN103294521B (en) 2013-05-30 2013-05-30 A kind of method reducing data center's traffic load and energy consumption

Country Status (1)

Country Link
CN (1) CN103294521B (en)

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103559084A (en) * 2013-10-17 2014-02-05 电子科技大学 Virtual machine migration method of energy-saving data center
CN103581048A (en) * 2013-11-08 2014-02-12 国家电网公司 Method for having control over overload of distributed resources of electric power communication network
CN103905337A (en) * 2014-03-31 2014-07-02 华为技术有限公司 Network resource processing device, method and system
CN104035868A (en) * 2014-06-12 2014-09-10 天津大学 Data center calculation method for block bordered diagonal model decomposition coordination algorithm
CN104050038A (en) * 2014-06-27 2014-09-17 国家计算机网络与信息安全管理中心 Virtual machine migration method based on policy perception
CN104298539A (en) * 2014-10-13 2015-01-21 南京大学 Network awareness based virtual machine dispatching and re-dispatching method
WO2015172661A1 (en) * 2014-05-13 2015-11-19 天津大学 Compression sensing reconstruction method for monitoring microgrid harmonic wave
CN105426243A (en) * 2015-11-19 2016-03-23 国云科技股份有限公司 Openstack based physical machine remote maintenance method
CN105812170A (en) * 2014-12-31 2016-07-27 华为技术有限公司 Data center-based fault analysis method and device
CN106304308A (en) * 2016-09-19 2017-01-04 南京邮电大学 A kind of multi-service deposit system medium cloud business energy optimization dispatching method
CN106447127A (en) * 2016-10-21 2017-02-22 黄东 Energy consumption optimization planning method for smart grid
CN107077387A (en) * 2014-09-09 2017-08-18 微软技术许可有限责任公司 Resources control for virtual data center
CN108108224A (en) * 2017-12-05 2018-06-01 西南交通大学 Virtual machine placement method in cloud data center based on ant colony optimization algorithm
CN109314630A (en) * 2018-09-06 2019-02-05 深圳市汇顶科技股份有限公司 Chain processing method, equipment and storage medium
CN109325344A (en) * 2018-09-13 2019-02-12 郑州云海信息技术有限公司 The virtual machine migration method and system of side-channel attack are defendd in a kind of cloud environment
CN109428841A (en) * 2017-08-30 2019-03-05 英特尔公司 For the technology of automated network congestion management
CN110673640A (en) * 2019-10-21 2020-01-10 深圳市道通智能航空技术有限公司 Unmanned aerial vehicle control method, device, equipment and storage medium
CN112214276A (en) * 2019-07-11 2021-01-12 富士通株式会社 Information processing apparatus, system, and method, and non-transitory computer-readable recording medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101719081A (en) * 2009-12-01 2010-06-02 北京大学 Method for scheduling virtual machines
CN102096461A (en) * 2011-01-13 2011-06-15 浙江大学 Energy-saving method of cloud data center based on virtual machine migration and load perception integration

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101719081A (en) * 2009-12-01 2010-06-02 北京大学 Method for scheduling virtual machines
CN102096461A (en) * 2011-01-13 2011-06-15 浙江大学 Energy-saving method of cloud data center based on virtual machine migration and load perception integration

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
孙鑫: "面向云环境数据中心的高效资源调度机制研究", 《中国博士学位论文全文数据库》 *

Cited By (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103559084A (en) * 2013-10-17 2014-02-05 电子科技大学 Virtual machine migration method of energy-saving data center
CN103581048A (en) * 2013-11-08 2014-02-12 国家电网公司 Method for having control over overload of distributed resources of electric power communication network
CN103581048B (en) * 2013-11-08 2016-04-13 国家电网公司 A kind of distributed resource overload controlling method of powerline network
CN103905337A (en) * 2014-03-31 2014-07-02 华为技术有限公司 Network resource processing device, method and system
US10200287B2 (en) 2014-03-31 2019-02-05 Huawei Technologies Co., Ltd. Network resource processing apparatus, method, and system
CN103905337B (en) * 2014-03-31 2018-01-23 华为技术有限公司 A kind of processing unit of Internet resources, method and system
WO2015149491A1 (en) * 2014-03-31 2015-10-08 华为技术有限公司 Network resource processing device, method and system
WO2015172661A1 (en) * 2014-05-13 2015-11-19 天津大学 Compression sensing reconstruction method for monitoring microgrid harmonic wave
CN104035868B (en) * 2014-06-12 2016-08-24 天津大学 Diagonal angle edged model decomposition coordinates the data center's method for solving calculated
US10235341B2 (en) 2014-06-12 2019-03-19 Tianjin University Method for solving the decomposition-coordination calculation based on block bordered diagonal form (BBDF) model using data center
CN104035868A (en) * 2014-06-12 2014-09-10 天津大学 Data center calculation method for block bordered diagonal model decomposition coordination algorithm
CN104050038B (en) * 2014-06-27 2018-04-10 国家计算机网络与信息安全管理中心 A kind of virtual machine migration method based on policy-aware
CN104050038A (en) * 2014-06-27 2014-09-17 国家计算机网络与信息安全管理中心 Virtual machine migration method based on policy perception
CN107077387A (en) * 2014-09-09 2017-08-18 微软技术许可有限责任公司 Resources control for virtual data center
CN104298539B (en) * 2014-10-13 2017-09-22 南京大学 Scheduling virtual machine and dispatching method again based on network aware
CN104298539A (en) * 2014-10-13 2015-01-21 南京大学 Network awareness based virtual machine dispatching and re-dispatching method
US10831630B2 (en) 2014-12-31 2020-11-10 Huawei Technologies Co., Ltd. Fault analysis method and apparatus based on data center
CN105812170B (en) * 2014-12-31 2019-01-18 华为技术有限公司 Failure analysis methods and device based on data center
CN105812170A (en) * 2014-12-31 2016-07-27 华为技术有限公司 Data center-based fault analysis method and device
CN105426243A (en) * 2015-11-19 2016-03-23 国云科技股份有限公司 Openstack based physical machine remote maintenance method
CN106304308A (en) * 2016-09-19 2017-01-04 南京邮电大学 A kind of multi-service deposit system medium cloud business energy optimization dispatching method
CN106304308B (en) * 2016-09-19 2019-06-28 南京邮电大学 A kind of multi-service and deposit system medium cloud business energy optimization dispatching method
CN106447127A (en) * 2016-10-21 2017-02-22 黄东 Energy consumption optimization planning method for smart grid
CN109428841A (en) * 2017-08-30 2019-03-05 英特尔公司 For the technology of automated network congestion management
CN108108224B (en) * 2017-12-05 2019-10-01 西南交通大学 Virtual machine placement method in cloud data center based on ant colony optimization algorithm
CN108108224A (en) * 2017-12-05 2018-06-01 西南交通大学 Virtual machine placement method in cloud data center based on ant colony optimization algorithm
CN109314630A (en) * 2018-09-06 2019-02-05 深圳市汇顶科技股份有限公司 Chain processing method, equipment and storage medium
CN109314630B (en) * 2018-09-06 2022-01-25 深圳市汇顶科技股份有限公司 Link processing method, device and storage medium
CN109325344A (en) * 2018-09-13 2019-02-12 郑州云海信息技术有限公司 The virtual machine migration method and system of side-channel attack are defendd in a kind of cloud environment
CN112214276A (en) * 2019-07-11 2021-01-12 富士通株式会社 Information processing apparatus, system, and method, and non-transitory computer-readable recording medium
CN110673640A (en) * 2019-10-21 2020-01-10 深圳市道通智能航空技术有限公司 Unmanned aerial vehicle control method, device, equipment and storage medium

Also Published As

Publication number Publication date
CN103294521B (en) 2016-08-10

Similar Documents

Publication Publication Date Title
CN103294521A (en) Method for reducing communication loads and energy consumption of data center
CN106411770B (en) A kind of data center network energy-saving routing algorithm based on SDN framework
Li et al. Software defined green data center network with exclusive routing
Kliazovich et al. GreenCloud: a packet-level simulator of energy-aware cloud computing data centers
Wang et al. Dynamic SDN controller assignment in data center networks: Stable matching with transfers
CN103559084B (en) A kind of virtual machine migration method at Energy-saving Data center
CN104572307B (en) The method that a kind of pair of virtual resource carries out flexible scheduling
CN108566659A (en) A kind of online mapping method of 5G networks slice based on reliability
Yang et al. Energy-efficient data center networks planning with virtual machine placement and traffic configuration
CN103077082A (en) Method and system for distributing data center load and saving energy during virtual machine migration
CN103457852A (en) Invulnerability mapping method of multicast virtual network
Dalvandi et al. Time-aware vmflow placement, routing, and migration for power efficiency in data centers
Wang et al. Achieving energy efficiency in data centers using an artificial intelligence abstraction model
CN108664116A (en) Adaptive electricity saving method, device and the cpu controller of network function virtualization
Kliazovich et al. Simulation and performance analysis of data intensive and workload intensive cloud computing data centers
Chen et al. Deep learning-based traffic prediction for energy efficiency optimization in software-defined networking
CN107835130A (en) A kind of flow allocation method and device
CN103595610B (en) A kind of the anti-of non-deterministic source demand multicast virtual network ruins mapping method
Cheung et al. DFFR: A flow-based approach for distributed load balancing in Data Center Networks
Yang et al. Collective energy-efficiency approach to data center networks planning
CN106209390A (en) Network device management and power-economizing method and resource management apparatus, route energy saver
CN106209683B (en) Data transmission method and system based on data center&#39;s wide area network
Zhuang et al. Optimization of mobile MEC offloading with energy harvesting and dynamic voltage scaling
Luo et al. Traffic-aware VDC embedding in data center: A case study of fattree
CN105577834B (en) Two layers of bandwidth allocation methods of cloud data center with Predicable performance and system

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant