CN104065547B - A kind of system of selection for calculating central interior physical host - Google Patents
A kind of system of selection for calculating central interior physical host Download PDFInfo
- Publication number
- CN104065547B CN104065547B CN201410284277.1A CN201410284277A CN104065547B CN 104065547 B CN104065547 B CN 104065547B CN 201410284277 A CN201410284277 A CN 201410284277A CN 104065547 B CN104065547 B CN 104065547B
- Authority
- CN
- China
- Prior art keywords
- subtree
- height
- path
- virtual machine
- tree
- 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.)
- Expired - Fee Related
Links
Abstract
The invention discloses a kind of system of selection for calculating central interior physical host, belong to technical field of data processing, solve the problems, such as to maximize the communication cost for reducing and calculating between central interior virtual machine.The method realizes that step includes：Input tree and relevant parameter, numerical value initialization, traversal of tree, the minimum subtree of record and the minimum subtree of record.The present invention effectively overcomes the defect for reducing the communication cost problem between virtual machine in the prior art, it is determined that on the basis of the network architecture for calculating central interior, interior portion selects suitable region and server set therein to be that client's request and application program provide service, the maximized communication cost reduced between virtual machine in the calculation.
Description
Technical field
The invention belongs to technical field of data processing, it is related to a kind of system of selection for calculating central interior physical host, tool
Body is related to a kind of method for how reducing and calculating the communication cost between central interior virtual machine.
Background technology
Calculating center needs to provide the support of specific virtual machine number for specific application program or subtask.Distribution one
The virtual machine set at calculating center is placed on and calculates the different frame of central interior or server.It is logical between virtual machine in frame
Letter can be rapidly completed.The communication of convergence switch can occupy a part for the frame network bandwidth, between virtual machine away from
From increase, the available bandwidth between convergence switch gradually decreases.The virtual machine for the treatment of application program or subtask it is available
Bandwidth depends on the position of the physical host that virtual machine is placed.The efficiency of calculating central data center also relies on virtual machine and puts
The selection put.
Each application program or subtask are all added up in the virtual machine number of this calculating focal need, certain is obtained
Individual calculating center needs all virtual machine numbers for providing.Should if the calculating center needs the whole virtual machine numbers for providing to be less than
Cardiac volume in calculating, it is necessary to how to select the region in calculating center, not only makes the local meet virtual machine demand, and most
The communication cost reduced between virtual machine of bigization.
For the problem for how reducing the communication cost problem between virtual machine, several representational plans are occurred in that at present
Slightly.The first is random selection, and interior portion randomly chooses a server and places virtual machine in the calculation, if this server
Off-capacity, then randomly choose next server, until having placed whole virtual machines.This Placement Strategy is simple and easy to apply,
But do not account for reducing the communication distance between virtual machine.Second is greedy selection, and server is carried out from big according to capacity
To small sequence, the server of maximum capacity is always selected to place virtual machine.This Placement Strategy, does not consider from global angle,
The big frame of capacity or server potential range it is very remote.The third is the minimum subtree of selection height, is allowed in this subtree
Server place all virtual machines.This Placement Strategy, by server centered in the height of subtree, is largely reduced
Communication distance between virtual machine, but it is in the heart in the calculation very common, it is necessary to more preferable the same subtree of height occur
This strategy of criteria optimization.
Above-mentioned the third strategy is the physical host that central interior is calculated based on the selection of the hierarchical tree-type network system.Hierarchical tree
Type system is presently the most typical architecture, is also the commonly used architecture generally accepted with academia of industrial quarters.
Each frame of hierarchical tree-type system includes multiple servers, and TOR (top-of-the- are passed through between server and server
Rack) interchanger is communicated.TOR interchangers are located at the top of frame.Communication between frame and frame is exchanged by converging
Machine is carried out.Therefore, if two servers in adjacent rack need to be communicated, it is necessary to by source TOR interchanger, converge
Interchanger and the such paths of purpose TOR interchangers.If machine is located at farther place, it is necessary to by the convergence of multilayer
Interchanger.The communication delay for running on the application program in virtual machine depends on the position of the physical host that virtual machine is placed.
The content of the invention
To overcome the shortcomings of to reduce the communication cost problem between virtual machine in the prior art, the invention provides one kind meter
The system of selection of central interior physical host is calculated, it is determined that on the basis of the network architecture for calculating central interior, calculating
Central interior selects suitable region and server set therein to be that client's request and application program provide service, most
The communication cost reduced between virtual machine of bigization.
To achieve the above object, the present invention is adopted the technical scheme that：
A kind of system of selection for calculating central interior physical host, comprises the following steps：
Step one, input tree and relevant parameter
A triplet sets G=(T, r, g) is given, wherein, T represents tree, and r represents the root node of tree, and g is represented in calculating
The heart needs the virtual machine number for providing；There are three variable functions in tree：Weight (), height () and path (), wherein,
Weight (r) represents the virtual machine number that the tree with r as root node can provide, and height (r) represents the tree with r as root node
Height, path (r) represent with r as root node tree exist farthest path；
The root node of the subtree of tree is with rnRepresent, n is positive integer, and the span of n is [1, n], the variable in subtree：
weight(rn)、height(rn) and path (rn), wherein, weight (rn) represent with rnFor the subtree of root node can be provided
Virtual machine number, height (rn) represent with rnIt is the height of the subtree of root node, path (rn) represent with rnIt is the son of root node
Set the farthest path for existing；
Given variable minHeight, minTree, minPath, longestPath, h1 and h2, wherein, minHeight tables
Show the height of local optimum subtree in ergodic process, minTree represents the subtree of local optimum in ergodic process, minPath tables
Show the path length of local optimum subtree in ergodic process, longestPath represent ergodic process in local optimum subtree most
Path length long, h1 represent ergodic process in the first subtree high highly, h2 represent ergodic process in the second subtree high highly；
Step 2, numerical value initialization
Initializing variable, i.e. height (r)=0, path (r)=0, height (rn)=0, path (rn)=0,
MinHeight is initialized as infinity, and minTree is initialized as sky, longestPath=0, h1=0, h2=0；
Step 3, traversal of tree
Using the method for postorder traversal, all nodes of tree are begun stepping through from root node r；
The minimum subtree of step 4, record
If minHeight >=height (rn), then minHeight=height (rn)；If minPath >=path (rn), then
MinPath=path (rn)；And record minimum subtree minTree=rn,
Judge weight (rnWhether) >=g sets up, if weight (rn) >=g is invalid, then n=n+1, goes to step 3) after
Continuous traversal；If weight (rn) >=g sets up, and performs step 5)；
Step 5, judge whether traversal of tree terminates
Judge height (rn)>Whether h1 sets up, if height (rn)>H1 sets up, then h2=h1, h1=height (rn)；
If height (rn)>H1 is invalid, is further continued for judging height (rn)>Whether h2 sets up, if height (rn)>H2 sets up, then h2
=height (rn)；
If longestPath>path(rn), then longestPath=path (rn), that is, obtain most short longest path path length
Degree；
If setting no traversal to terminate, n=n+1 goes to step 3) continue to travel through；
Judge h2+h1>Whether longestPath sets up, if h2+h1>LongestPath sets up, then path (rn)=h2+
h1；If h2+h1>LongestPath is invalid, then path (rn)=longestPath；
Step 6, draw optimal subtree
Judge that minTree whether there is, if minTree is present, weight (r)=weight (minTree), height
(rn)=height (minTree)；If minTree does not exist, it is necessary to which whether the virtual machine number of decision tree T is more than g, if
It is, minTree=r, weight (r)=weight (minTree) that height (r)=height (minTree) does not have otherwise
Optimal subtree；
Step 7, the physical host placement virtual machine selected in optimal subtree
The server included in optimal subtree is selected, used as the physical host for placing virtual machine, what treatment user submitted to appoints
Business or the subtask of cloud platform distribution.
Beneficial effects of the present invention：
The present invention proposes the improvement of different emphasis on the basis of the optimal subtree of searching based on height, and examines simultaneously
Fault-tolerance is considered, has described tree-network topology structure in detail first, the improvement for finding optimal subtree has been proposed on this basis
Method, using the quality of different criterion subtrees, and analyzes the reasonability of modified hydrothermal process in theory.The present invention has
The defect for reducing the communication cost problem between virtual machine in the prior art is overcome to effect, it is determined that calculating the net of central interior
On the basis of network architecture, interior portion selects suitable region and server set therein to be client in the calculation
Request and application program provide service, the maximized communication cost reduced between virtual machine.
The present invention is elaborated further below with reference to embodiment.
Brief description of the drawings
Fig. 1 is the system of selection flow chart for calculating central interior physical host.
Fig. 2 is that single calculating center uses algorithms of different process task figure.
Fig. 3 is that multiple calculating centers are jointly processed by user's request figure.
Fig. 4 is the different request figure of single calculating center processing.
Fig. 5 is the request figure for calculating center processing demand different virtual machine number.
Specific embodiment
Embodiment 1：
Illustrated with reference to the system of selection of 1 pair of calculating central interior physical host of accompanying drawing, the system of selection includes as follows
Step：
Step one, input tree and relevant parameter
A triplet sets G=(T, r, g) is given, wherein, T represents tree, and r represents the root node of tree, and g is represented in calculating
The heart needs the virtual machine number for providing；There are three variable functions in tree：Weight (), height () and path (), wherein,
Weight (r) represents the virtual machine number that the tree with r as root node can provide, and height (r) represents the tree with r as root node
Height, path (r) represent with r as root node tree exist farthest path；
The root node of the subtree of tree is with rnRepresent, n is positive integer, and the span of n is [1, n], the variable in subtree：
weight(rn)、height(rn) and path (rn), wherein, weight (rn) represent with rnFor the subtree of root node can be provided
Virtual machine number, height (rn) represent with rnIt is the height of the subtree of root node, path (rn) represent with rnIt is the son of root node
Set the farthest path for existing；
Given variable minHeight, minTree, minPath, longestPath, h1 and h2, wherein, minHeight tables
Show the height of local optimum subtree in ergodic process, minTree represents the subtree of local optimum in ergodic process, minPath tables
Show the path length of local optimum subtree in ergodic process, longestPath represent ergodic process in local optimum subtree most
Path length long, h1 represent ergodic process in the first subtree high highly, h2 represent ergodic process in the second subtree high highly；
Step 2, numerical value initialization
Initializing variable, i.e. height (r)=0, path (r)=0, height (rn)=0, path (rn)=0,
MinHeight is initialized as infinity, and minTree is initialized as sky, longestPath=0, h1=0, h2=0；
Step 3, traversal of tree
Using the method for postorder traversal, all nodes of tree are begun stepping through from root node r；
The minimum subtree of step 4, record
If minHeight >=height (rn), then minHeight=height (rn)；If minPath >=path (rn), then
MinPath=path (rn)；And record minimum subtree minTree=rn,
Judge weight (rnWhether) >=g sets up, if weight (rn) >=g is invalid, then n=n+1, goes to step 3) after
Continuous traversal；If weight (rn) >=g sets up, and performs step 5)；
Step 5, judge whether traversal of tree terminates
Judge height (rn)>Whether h1 sets up, if height (rn)>H1 sets up, then h2=h1, h1=height (rn)；
If height (rn)>H1 is invalid, is further continued for judging height (rn)>Whether h2 sets up, if height (rn)>H2 sets up, then h2
=height (rn)；
If longestPath>path(rn), then longestPath=path (rn), that is, obtain most short longest path path length
Degree；
If setting no traversal to terminate, n=n+1 goes to step 3) continue to travel through；
Judge h2+h1>Whether longestPath sets up, if h2+h1>LongestPath sets up, then path (rn)=h2+
h1；If h2+h1>LongestPath is invalid, then path (rn)=longestPath；
Step 6, draw optimal subtree
Judge that minTree whether there is, if minTree is present, weight (r)=weight (minTree), height
(rn)=height (minTree)；If minTree does not exist, it is necessary to which whether the virtual machine number of decision tree T is more than g, if
It is, minTree=r, weight (r)=weight (minTree) that height (r)=height (minTree) does not have otherwise
Optimal subtree；
Step 7, the physical host placement virtual machine selected in optimal subtree
The server included in optimal subtree is selected, used as the physical host for placing virtual machine, what treatment user submitted to appoints
Business or the subtask of cloud platform distribution.
On this basis, it is also possible to which the method to the present embodiment is further optimized.
(1) selected by the first criteria optimization subtree of farthest path
The height of subtree is the standard for judging that subtree is good and bad in the above method, while with longest path path length present in subtree
Degree is selected as the second standard in mutually level subtree.This optimization can to a certain extent be reduced to user's application
Communication cost between the virtual machine of procedure service, improves the performance of application program.But with longest path path length present in subtree
Spend if the quality for the first criterion subtree, can more reduce the communication cost between virtual machine, more
Raising application program performance.
From being that raising of first standard to application program capacity is more helpful with longest path length present in subtree,
The upper limit of communication cost between virtual machine is not only given as standard using longest path, to be also given process in whole stalk tree and is used
The upper limit of the communication cost total amount that all virtual machines of family request occur, user always selects the best upper limit as measurement
The good and bad standard of subtree, it is thus determined that first standard good and bad as subtree is weighed of longest path length present in subtree.The
Two standards then use the height of subtree, and the minimum subtree of Select Subtree height, uses in longest path length identical subtree group
It processes user's request.
(2) selected as the second criteria optimization subtree with root node to the maximum distance of subtree
By above-mentioned analysis, it is determined that with the length of point-to-point transmission longest path present in subtree alternatively subtree
One standard, and using the height of subtree as the second standard.Herein, reasonability of the analysis using subtree height as the second standard,
And be improved, select more suitably the second standard.
Select the task that suitable subtree is submitted to as the local for placing virtual machine, treatment user or the son that cloud platform is distributed
Task.Occur that the virtual machine for processing same subtask is placed on multiple subtrees for calculating central interior according to algorithms selection
On, need frequently to be in communication with each other between these virtual machines.Strategy of the invention be allow calculating center unique gateway i.e.
The distance of top-level router root node to selected subtree is short as far as possible, to be conducive to the virtual machine being distributed between different subtrees more
Fast communication.Using the whole root node of tree to the length in the farthest path of leaf node in subtree as judging the of subtree quality
Two standards.This farthest path from top to bottom be leaf node in selected subtree to the whole longest distance of root vertex,
It is the upper bound of leaf node and communication port communication distance.And the standard of subtree highly this dimension can not define any leading to
Communication distance, provides optimization, therefore select the whole root node of tree to the longest path length conduct of leaf node in selected subtree
Second standard.Arrive here, two criterions are just all determined.Longest path length present in subtree is examining for transverse dimensions
Amount, the longest path length of leaf node is considering for longitudinal dimension in whole root vertex to selected subtree.
When interior portion Select Subtree places the virtual machine for the treatment of subtask in the calculation, if necessary to calculate center with other
Internal selected subtree is jointly processed by subtask, and the subtree of selection allows as far as possible on the premise of distribution virtual machine number is met
Selected subtree is a little close to communication port；If all distribution, at this calculating center, is selected the virtual machine needed for subtask
Subtree as far as possible from communication port be away from root node it is a little, will be close to the computing resource of root node, to leave communication requirement for higher
Subtree.
Calculating center is abstracted into tree-network topology structure, and communication port is that top-level router is expressed as root node,
Represented with root.Hop (x) represents node x to the whole distance of root vertex.Calculate each node to the calculation of the distance of root node
The method that method uses breadth first traversal, algorithm flow is as follows：
1st step：Queue Queue is initialized as sky, hop (root) ← 0；
2nd step：Root node root enters enqueue Queue；
3rd step：If queue Queue is sky, the 7th step is gone to；If queue Queue is not sky, the 4th step is gone to；
4th step：Head of the queue element goes out team, is assigned to node variable r；
5th step：If node r is leaf node, the 3rd step is gone to；Otherwise, the 6th step is gone to；
6th step：All child nodes x, x ∈ children (r) of node r, the distance to root node add 1, and hop (x) ←
Hop (x)+1, all of child nodes x enters enqueue Queue；Go to the 3rd step；
7th step：All nodes to the whole distance of root vertex are obtained, is terminated.
Height height (x) of each node x in whole tree is obtained by the present invention, is existed in the subtree with x as root node
Leaf node between shortest path length path (x), as the first standard.By calculating each node to the distance of root node
Algorithm, obtains communication port root node to farthest path length hop (the x)+height (x) of all leaf nodes in subtree,
As the second standard.
Select from calculating the nearest subtree of center access point node because the subtask of the subtree treatment of selection by and only
One is processed by selected subtree.If specifically the subtask of subtree treatment as and the only calculating center processing as where subtree,
Center access point node is calculated in the subtree for so selecting more remote better；If the subtask of subtree treatment also needs to other calculating
The assist process at center, then from calculate center access point node more close to it is better.Task S is made up of m sub- task, is respectively necessary for
g1,g2,…,gmIndividual virtual machine.The corresponding triple set of service in m subtask is respectively C1,C2,…,Cm, have in distributed cloud
N calculating center is expressed as gathering { v1,v2,…,vn}.The triple set of service for being designated the subtask of i is：
Ci={ (v1,g1i,si),(v2,g2i,si),…,(vn,gni,si)} (1)
Formula (1) meets constraints：
Formula (2) represents that certain calculating center needs the summation of virtual machine number provided for all subtasks.If mark
For the calculating center of k meets：In gkiIn the case of ＞ 0, gki=gi,Wherein i ∈ 1,2 ..., m }.Illustrate calculating center k
Treatment subtask i, the virtual machine needed for subtask is all provided by calculating center k.Only in this case, in path (x)
Selected in equal subtreeMaximum subtree, otherwise, all selectsMost
Small subtree.
Embodiment 2：Experiment and analysis
(1) emulation experiment of optimal subtree is found
Realize that interior portion is found in the calculation in the operation simulation nucleus module DataCenterBroker of CloudSim
Optimal subtree algorithm verifies the virtual machine Placement Strategy of proposition.
Need to carry out necessary parameter setting and global configuration, All hosts in CloudSim emulation platforms in simulation process
Configuration is the bandwidth of the internal memory of 5G, the external memory of 2TB and 1Gbps, and host-processor is single core processor, and processor speed is
1100,2100 or 3200MIPS.In the case that cpu busy percentage is 0, node consumption watt/hour of electric energy 162, during CPU full loads,
Node consumes watt/hour of electric energy 266.The internal memory 1.5G of virtual machine, external memory 100GB, with a width of in CloudSim emulation platforms
150Mbps, the CPU processing speeds that each virtual machine needs are 270,550,650 or 1050MIPS.
For the performance of assessment algorithm, the present invention will be with two scholars of Mansoor Alicherry and T.V.Lakshman
The algorithm proposed in 2012 is contrasted.Two scholars of Mansoor Alicherry and T.V.Lakshman are according to subtree
Interior portion selects suitable subtree, referred to as Height algorithms to height in the calculation.Herein then with longest path present in subtree
Electrical path length is marked to the length in the farthest path of the leaf node of subtree using calculating center gateway as the first standard as second
Standard, so that interior portion selects suitable subtree, referred to as Heuristic algorithms in the calculation.This experiment to this in two method enter
Row comparative analysis.
In experiment, create 1000 × 1000 grid and random user's request, user's request certain amount it is virtual
Machine is used to run application program, and the treatment for entering algorithms of different selects different subtrees to undertake user's request.In being calculated at one
The Servers-all of interior portion is randomly distributed on this 1000 × 1000 network, and all of server is all leaf node, is used
Black circle is represented.Server on the grid line of same layer forms the brotgher of node, father node in leftmost node upper one
On layer mesh point, represented with soft dot.All child nodes are connected with father node with solid line, thus bottom-up to form whole
The tree network structure chart at calculating center.10 distributed cloud scenes are devised, respectively comprising 1,2,3,4,5,6,7,8,9,10
Individual calculating Center Number.The number that each distributed cloud possesses server is the same.Therefore, calculating center possesses server
The number that number possesses calculating center with a cloud is inversely proportional.Each calculating center possesses clothes in having 100 clouds at calculating center
The number of business device is between 50 to 100.Each calculating center of cloud for there is 50 calculating centers possess the number of server with
Machine is distributed between 100 to 200.All data are all to obtain the average value of experimental data after 100 operations in Fig. 2, real every time
Change random seed is tested, allows calculating center to possess different number of server, so repeated 100 times, it is same in experiment with computing result
Plant the average value of variable.
In this first experiment, a calculating center processing user's request, the task that user submits to is only allowed to be counted by this
Calculation center is all processed, in the case that test does not calculate center to center communications with other, algorithm Heuristic and algorithm Height tables
Reveal the performance come.Can go out from Fig. 2, when single calculating center processes user's request using Different Strategies, algorithm
Communication cost total amount between the server set of Heuristic selections is less than algorithm Height.The virtual machine number of user's request
Mesh is respectively 10,20,30,40,50,60,70,80,90,100, and with increasing for the virtual machine number asked, representing needs more
Many virtual machine treatment user's requests, operation user's trustship uses the service of algorithms of different selection in the application program of cloud platform
Communication total amount between device set is also on the increase.
The virtual machine number of request is fixed on 100 now, is measured respectively by different number of calculating center processing one
The communication cost total amount that the task of sample occurs, calculates center number and is respectively 1,2,3,4,5,6,7,8,9,10.User's request
100 virtual machines are provided jointly by multiple calculating centers, and each calculating center provides the virtual machine of certain amount.Make in experiment
During with the treatment user task at different number of calculating center, each calculating center is required for providing and participates in process task,
Heuristic algorithms and Height algorithms are identical for the offer virtual machine number with same calculating center under a scene
, it is ensured that the factor of change is using two different suitable subtrees of algorithms selection.From figure 3, it can be seen that with ginseng
More with the calculating center that service is calculated, the communication cost total amount needed for process task is bigger, because participating in the virtual machine for calculating
Different calculating centers are distributed in so that communication distance increases.The communication generation that the same task of Heuristic algorithm process occurs
Valency total amount is smaller than Height algorithm, and with the increase at the calculating center for participating in treatment, the gap of two algorithms also exists
Gradually increase.
(2) build true Open Source Platform and tested
There is blindness on emulation platform in view of the situation that virtual machine is placed in server, while for verification algorithm
Effect in true cloud platform, this experiment different virtual machine Placement Strategies is carried out on cloud platform Eucalyptus being increased income
Experiment.
Eucalyptus is that a kind of cloud computing infrastructure increased income provides software.User can select source code or bag
Install, Eucalyptus can be good in most of Linux releases according to.The eucalyptus of early stage does not have and Amazon
Service docking, present Eucalyptus provided using the interface of EC2 and S3, and user easily can be taken using third party
Business.These Interface designs are good, and all existing instruments can be assisted using these interfaces with the cloud platform based on Eucalyptus
Run.Eucalyptus platform can also be supported to operate in the operation of the virtual machine on Xen hypervisor or KVM.Different clothes
Business device cluster possesses each privately owned internal network address, and can be deployed in same cloud for they by Eucalyptus.
Eucalyptus includes Cloud Controller (CLC), Cluster Controller (CC), Node Controller
(NC), five primary clusterings of Walrus and Storage Controller (SC), using with WS-Security between them
Soap message communication so as to mutual close cooperation.
Experiment porch includes 11 physical nodes, and each Node deployment is 4 core processors (Intel Xeon E5606
2.13GHz), 8GB internal memories, two pieces of SATA disks of 7200 turns of 1TB, two pieces of Intel PCI-Express.All nodes pass through one thousand
Mbit ethernet interchanger is attached.One of node is used as control node, and other nodes are used for creating as node is calculated
Build virtual machine.The power supply respectively 88W and 120W that each node is consumed when cpu busy percentage is by 0% and 100%.Experiment is just
During the beginning, wherein 2-3 virtual machine is separately operable in 8 physical nodes, and it is virtual without operation in other two physical nodes
Machine, all physical servers install the ubuntu operating systems and Linux 2.6.31 64-bit kernel for supporting Xen 3.3.
Realize that analog subscriber is asked using the java of TPC-W, represent user's trustship and equal the application program in cloud.TPC-
W is a multi-level ecommerce weblication, including Tomcat application layers and mysql database layers.TPC does not have
There is the task code of issue benchmark program, from the angle that unification is used, provide the standard criterion of benchmark program, any test
Person combines the limitation such as oneself software and hardware condition according to specification, and test program and the deployment survey on the unified platform are write out in oneself ground
Examination.The version changed using Tomcat-5.5.27 is as the container and mysql-5.0.45 of servlet as Backup Data
Storehouse, so as to create the application copy of TPC-W.The service time of each request of Tomcat server records and statistics.
The virtual machine created in Eucalyptus platforms is before TPC-W application programs are run, it is necessary to enter according to the specified specification of TPC-W
The corresponding configuration of row.
Required virtual machine is created in Eucalyptus platforms.After system sets up image caching, virtual machine instance example
Become to run running states from suspension status.
This virtual machine instance is assigned to from the IP address pond (192.168.0.2-192.168.0.50) of Front end
The IP of 192.168.0.12, wherein i-3E740761 is example number, is in systems unique, 172.19.1.4 is then face
To the IP of virtual machine.Can see this IP using issuing orders is strictly to be allocated to this example.In all of IP, there is three
Individual to have dispensed, one of them is exactly 192.168.0.12, and corresponding use example number is i-3E740761.
Using the ssh keys for creating to signing in virtual machine instance in, obtain a complete whole PC.Use
Ssh-i/home/xdcloud1/my.private ubuntu@192.168.0.12 orders are logged in.Setting up connection can ask whether
Continue to connect, be input into yes, sign in in the virtual machine instance of the entitled ubuntu of user, virtual machine can be shown after logging in into
Current service condition.This virtual machine is logined to be configured accordingly and run TPW-W application programs.
Do not have patterned monitoring software in Eucalyptus platforms, the opening API provided using Eucalyptus
It is easy to the graphic monitoring equipment of gathered data with Web Service Technology designs.Keeper is by inputing user name and password
Log in cloud platform.Cloud computing management platform shows the virtual machine being currently running, and clicking on certain virtual machine, can to obtain this virtual
The brief description of the resource service condition of machine.Certain desired virtual machine for obtaining its resource service condition is put in systems, can be with
The resource of the resource using status, the program of current operation and its occupancy checked in virtual machine.
In experiment, the grid that creates 1000 × 1000 and the TPC-W user's requests specified, user's request certain amount
Virtual machine be used for run application program one calculate central interior Servers-all be randomly distributed over this 1000 × 1000
On network, all of server is all leaf node, is represented with black circle.Server on the grid line of same layer forms brother
Younger brother's node, father node is represented on the last layer mesh point of leftmost node with soft dot.All child nodes are saved with father
Point is connected with solid line, thus bottom-up to form the whole tree network structure chart for calculating center.Distributed cloud scene is different, but
The number that each distributed cloud can possess server is the same.Single calculating center processing client is tested in this experiment
The performance condition and communication cost of request, number of the single calculating center comprising server is between 5000 to 1000
Random value.All data are all to obtain the average value of experimental data after 100 operations in Fig. 4, and experiment every time changes random seed,
Allow calculating center to possess different number of server, so repeat 100 times, to measure its average for change of the same race in experiment with computing result
Value.
In this experiment, only using a calculating center processing user's request, the task that user submits to is calculated by this
Center is all processed, in the case that test does not provide service with other calculating center complex, algorithm Heuristic, algorithm
The performance that Height and algorithm Random are each showed, counts son selected by the application program that cloud platform is deployed in process
Farthest path length between all leaf nodes of tree.Can go out from Fig. 4, single calculating center selects son using Different Strategies
During tree, farthest path length is less than Random plans between all leaf nodes in the selected subtree of the selection strategy of algorithm Height
Slightly, because Random algorithms always randomly choose a server, entered not the need for communication path between the server of selection
With the intermediary switch of level, communication cost is larger in general.Set forth herein Heuristic methods to be better than
Height methods.In experiment, the virtual machine number of user's request is respectively 20,40,60,80,100, with the virtual machine of request
Number increases, it is necessary to more server places virtual machine, it is desirable to have the subtree of more Large Copacity processes application program, therefore
The farthest path that selected subtree is included between more leaf nodes and intermediary switch, all leaf nodes increases therewith.
The virtual machine number of request is increased into 9 selections, in experiment the virtual machine number of user's request be respectively 20,
30th, 40,50,60,70,80,90,100, the selection for each number is tested, and the server number for calculating center is random
Value, between 5000 to 10000, this calculating center needs selection suitable to span in order to provide certain virtual machine number
Subtree, the communication cost total amount that the virtual machine in the selected subtree of measurement on all leafy nodes occurs changes and calculates next time
The number of servers that center possesses, then same measurement is done, so repeat 100 times, calculate the average value of communication cost total amount.
Although algorithm is different, scene identical calculations center is identical, and the offer virtual machine number for calculating center is identical, it is ensured that change
Factor be using three different algorithms selection subtrees.From figure 5 it can be seen that with the increasing of request virtual machine number
Many, no matter three kinds of methods place virtual machine using that Select Subtree, and communication cost can all increase, because demand virtual machine
Number increases, and the leaf node and intermediary switch in the subtree of selection also increase, farthest path length present in subtree
Can increase, overall communication cost is just increased.The fault-tolerance of the same task of Heuristic algorithm process is than other two
Algorithm is better, and Height algorithms are better than Random algorithm, because Random algorithms are randomly selected arbitrarily
The subtree of leaf node composition, the farthest communication path length of any two leaf node is usually very big inside this subtree,
There is no significant difference with blindness higher and directly using this performance for calculating center.This experiment user asks virtual machine
Number is set in decimal magnitude, and as user's request virtual machine number is increased considerably, the algorithm that this experiment is proposed is selected
Communication cost and delay aspect inside subtree has the trend of continuous expansion compared to other two algorithms.
Further narration has been done to the present invention above in conjunction with embodiment, but the present invention is not limited to above-mentioned implementation method,
In the ken that one skilled in the relevant art possesses, can also be made on the premise of present inventive concept is not departed from
Various change.
Claims (1)
1. a kind of system of selection for calculating central interior physical host, it is characterised in that comprise the following steps：
1) input tree and relevant parameter
A triplet sets G=(T, r, g) is given, wherein, T represents tree, and r represents the root node of tree, and g represents calculating center and needs
The virtual machine number to be provided；There are three variable functions in tree：Weight (), height () and path (), wherein,
Weight (r) represents the virtual machine number that the tree with r as root node can provide, and height (r) represents the tree with r as root node
Height, path (r) represent with r as root node tree exist farthest path；
The root node of the subtree of tree is with rnRepresent, n is positive integer, and the span of n is [1, n], the variable in subtree：weight
(rn)、height(rn) and path (rn), wherein, weight (rn) represent with rnFor the virtual machine that the subtree of root node can be provided
Number, height (rn) represent with rnIt is the height of the subtree of root node, path (rn) represent with rnFor the subtree of root node is present
Farthest path；
Given variable minHeight, minTree, minPath, longestPath, h1 and h2, wherein, minHeight is represented time
The height of local optimum subtree during going through, minTree represents the subtree of local optimum in ergodic process, and minPath is represented time
The path length of local optimum subtree during going through, longestPath represents the longest path of local optimum subtree in ergodic process
Electrical path length, h1 represent ergodic process in the first subtree high highly, h2 represent ergodic process in the second subtree high highly；
2) numerical value initialization
Initializing variable, i.e. height (r)=0, path (r)=0, height (rn)=0, path (rnAt the beginning of)=0, minHeight
Beginning turns to infinity, and minTree is initialized as sky, longestPath=0, h1=0, h2=0；
3) traversal of tree
Using the method for postorder traversal, all nodes of tree are begun stepping through from root node r；
4) minimum subtree is recorded
If minHeight >=height (rn), then minHeight=height (rn)；If minPath >=path (rn), then
MinPath=path (rn)；And record minimum subtree minTree=rn,
Judge weight (rnWhether) >=g sets up, if weight (rn) >=g is invalid, then n=n+1, goes to step 3) continuation time
Go through；If weight (rn) >=g sets up, and performs step 5)；
5) judge whether traversal of tree terminates
Judge height (rn)>Whether h1 sets up, if height (rn)>H1 sets up, then h2=h1, h1=height (rn)；If
height(rn)>H1 is invalid, is further continued for judging height (rn)>Whether h2 sets up, if height (rn)>H2 sets up, then h2=
height(rn)；
If longestPath>path(rn), then longestPath=path (rn), that is, obtain most short longest path length；
If setting no traversal to terminate, n=n+1 goes to step 3) continue to travel through；
Judge h2+h1>Whether longestPath sets up, if h2+h1>LongestPath sets up, then path (rn)=h2+h1；If
h2+h1>LongestPath is invalid, then path (rn)=longestPath；
6) optimal subtree is drawn
Judge that minTree whether there is, if minTree is present, weight (r)=weight (minTree), height (rn)=
height(minTree)；If minTree does not exist, it is necessary to which whether the virtual machine number of decision tree T is more than g, if it is,
MinTree=r, weight (r)=weight (minTree), height (r)=height (minTree), it is otherwise not optimal
Subtree；
7) physical host in optimal subtree is selected to place virtual machine
Select the server included in optimal subtree, as place virtual machine physical host, treatment user submit to task or
The subtask of cloud platform distribution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410284277.1A CN104065547B (en) | 2014-06-23 | 2014-06-23 | A kind of system of selection for calculating central interior physical host |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410284277.1A CN104065547B (en) | 2014-06-23 | 2014-06-23 | A kind of system of selection for calculating central interior physical host |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104065547A CN104065547A (en) | 2014-09-24 |
| CN104065547B true CN104065547B (en) | 2017-06-20 |
Family
ID=51553088
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410284277.1A Expired - Fee Related CN104065547B (en) | 2014-06-23 | 2014-06-23 | A kind of system of selection for calculating central interior physical host |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104065547B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529241A (en) * | 2016-11-18 | 2017-03-22 | 郑州云海信息技术有限公司 | Virtual machine user account and password resetting method and system |
| CN106933654B (en) * | 2017-03-17 | 2020-08-28 | 中山大学 | Virtual machine starting method based on cache |
| CN113535320B (en) * | 2020-04-14 | 2024-02-23 | 深信服科技股份有限公司 | Data access method, device, equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102929687A (en) * | 2012-10-12 | 2013-02-13 | 山东省计算中心 | Energy-saving virtual machine placement method for cloud computing data center |
| CN103414752A (en) * | 2013-07-16 | 2013-11-27 | 上海交通大学 | Network-awareness cloud data center virtual machine allocation method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8959523B2 (en) * | 2012-03-30 | 2015-02-17 | International Business Machines Corporation | Automated virtual machine placement planning using different placement solutions at different hierarchical tree levels |
-
2014
- 2014-06-23 CN CN201410284277.1A patent/CN104065547B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102929687A (en) * | 2012-10-12 | 2013-02-13 | 山东省计算中心 | Energy-saving virtual machine placement method for cloud computing data center |
| CN103414752A (en) * | 2013-07-16 | 2013-11-27 | 上海交通大学 | Network-awareness cloud data center virtual machine allocation method |
Non-Patent Citations (1)
| Title |
|---|
| Network Aware Resource Allocation in Distributed Clouds;Mansoor Alicherry等;《2012 Proceedings IEEE INFOCOM》;20120330;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104065547A (en) | 2014-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10621009B2 (en) | System and method for topology-aware job scheduling and backfilling in an HPC environment | |
| TWI287195B (en) | Method, system and computer-readable medium storing thereon logic instructions for scheduling in a high-performance computing (HPC) system | |
| CA2503775C (en) | High performance computing system and method | |
| CN104335535B (en) | Use the method, apparatus and system of spanning tree and network switch element resource routing iinformation stream in a network | |
| JP4451806B2 (en) | On-demand instantiation in high performance computing (HPC) systems | |
| JP4986844B2 (en) | System and method for detecting and managing HPC node failures | |
| CN103827825B (en) | Virtual resource object component | |
| CN106385329B (en) | Processing method, device and the equipment of resource pool | |
| CN107330056A (en) | Wind power plant SCADA system and its operation method based on big data cloud computing platform | |
| JP2007533034A (en) | Graphical user interface for managing HPC clusters | |
| WO2005106695A2 (en) | System and method for cluster management for parallel task allocation in a multiprocessor architecture | |
| CN104270421A (en) | Multi-user cloud platform task scheduling method supporting bandwidth guarantee | |
| CN104065547B (en) | A kind of system of selection for calculating central interior physical host | |
| CN107133083A (en) | A kind of virtual data center resource provider method based on virtualization technology | |
| CN208013975U (en) | The hardware device of on-line intelligence ability platform | |
| Li et al. | Cost-minimizing bandwidth guarantee for inter-datacenter traffic | |
| CN106775987A (en) | A kind of dispatching method of virtual machine for improving resource efficiency safely in IaaS cloud | |
| Kaplan et al. | Unveiling the interplay between global link arrangements and network management algorithms on dragonfly networks | |
| Yang et al. | Collective energy-efficiency approach to data center networks planning | |
| CN107360031A (en) | It is a kind of based on optimization overhead gains than mapping method of virtual network | |
| CN106645868A (en) | Short-circuit current scanning method and device | |
| Xu et al. | Flexcloud: A flexible and extendible simulator for performance evaluation of virtual machine allocation | |
| CN103051474A (en) | Method of large scale cluster network cabling | |
| Luo et al. | Traffic-aware VDC embedding in data center: A case study of fattree | |
| CN105471986B (en) | A kind of Constructing data center Scale Revenue Ratio method and device |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170620 Termination date: 20180623 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |