CN104065547A - A method for selecting physical hosts inside a computing center - Google Patents

A method for selecting physical hosts inside a computing center Download PDF

Info

Publication number
CN104065547A
CN104065547A CN201410284277.1A CN201410284277A CN104065547A CN 104065547 A CN104065547 A CN 104065547A CN 201410284277 A CN201410284277 A CN 201410284277A CN 104065547 A CN104065547 A CN 104065547A
Authority
CN
China
Prior art keywords
height
subtree
path
tree
mintree
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
CN201410284277.1A
Other languages
Chinese (zh)
Other versions
CN104065547B (en
Inventor
沈玉龙
宗旋
张琪
姜晓鸿
裴庆祺
张华庆
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.)
Xidian University
Kunshan Innovation Institute of Xidian University
Original Assignee
Xidian University
Kunshan Innovation Institute of Xidian 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 Xidian University, Kunshan Innovation Institute of Xidian University filed Critical Xidian University
Priority to CN201410284277.1A priority Critical patent/CN104065547B/en
Publication of CN104065547A publication Critical patent/CN104065547A/en
Application granted granted Critical
Publication of CN104065547B publication Critical patent/CN104065547B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Computer And Data Communications (AREA)

Abstract

The invention discloses a method for selecting physical hosts inside a computing center, belongs to the technical field of data processing, and helps to solve the problem in reducing the communication cost between virtual machines inside the computing center to the largest degree. The realization step of the method comprises the following steps: inputting tree and correlation parameters; carrying out numerical value initialization, carrying out tree traversal and recording the minimum subtree. The method in the invention effectively helps to overcome the defects in the problem of reducing the communication cost between the virtual machines inside the computing center in the prior art; on the basis of determining a network architecture inside the computing center, appropriate areas and server sets are selected in the computing center so as to provide service for client requests and application programs; and thus, the communication cost between the virtual machines is reduced to the largest degree.

Description

The system of selection of a kind of computer center internal physical main frame
Technical field
The invention belongs to technical field of data processing, relate to the system of selection of a kind of computer center internal physical main frame, be specifically related to a kind of method that how to reduce the communication cost between computer center's internal virtual machine.
Background technology
Computer center need to provide for concrete application program or subtask the support of concrete virtual machine number.Distribute the virtual machine set of Yi Ge computer center to be placed on computer center inner different frame or server.Communication in frame between virtual machine can complete fast.The communication of convergence switch can occupy a part for the frame network bandwidth, and along with the increase of the distance between virtual machine, the available bandwidth between convergence switch reduces gradually.The available bandwidth of processing the virtual machine of application program or subtask depends on the position of the physical host that virtual machine is placed.The efficiency of data center of computer center also depends on the selection that virtual machine is placed.
Each application program or subtask are all added up at the virtual machine number of this computer center's demand, obtain all virtual machine numbers that certain computer center need to provide.Whole virtual machine numbers that Ruo Gai computer center need to provide are less than this computer center's capacity, just need to how to select the region in computer center, not only make this local meet virtual machine demand, and the communication cost between maximized minimizing virtual machine.
For the problem that how to reduce the communication cost problem between virtual machine, several representational strategies have been there are at present.The first is random selection, in computer center's internal random, selects a server to place virtual machine, if this server capacity is not enough, selects at random next server, until placed whole virtual machines.This Placement Strategy is simple, but does not consider to reduce the communication distance between virtual machine.The second is that greed is selected, and server is sorted from big to small according to capacity, always selects the server of heap(ed) capacity to place virtual machine.This Placement Strategy, does not consider from overall angle, frame capacious or server potential range very far away.The third is to select highly minimum subtree, allows the server in this subtree place all virtual machines.This Placement Strategy, in the height of subtree, has reduced the communication distance between virtual machine by server centered to a great extent, but occurs what be usually shown in by right and wrong in computer center for highly the same subtree, needs this strategy of better criteria optimization.
Above-mentioned the third strategy is based on the hierarchical tree type network system, to select the physical host of computer center inside.Hierarchical tree type system is the most typical architecture at present, is also that industrial quarters is generally applied the architecture of generally admitting with academia.Each frame of hierarchical tree type system comprises a plurality of servers, between server and server, by TOR (top-of-the-rack) switch, communicates.TOR switch is positioned at the top of frame.Communicating by letter between frame and frame undertaken by convergence switch.Therefore, two servers in adjacent frame need to communicate, just need to be through the such paths of source TOR switch, convergence switch and object TOR switch.If machine is positioned at farther place, just need to be by the convergence switch of multilayer.The communication delay that runs on the application program in virtual machine depends on the position of the physical host that virtual machine is placed.
Summary of the invention
For overcoming the deficiency that reduces the communication cost problem between virtual machine in prior art, the invention provides the system of selection of a kind of computer center internal physical main frame, on the basis of network architecture of determining computer center inside, select suitable region and server set wherein to provide service, the communication cost between maximized minimizing virtual machine for client requests and application program computer center is inner.
For achieving the above object, the technical scheme that the present invention takes is:
A system of selection for computer center's internal physical main frame, comprises the steps:
Step 1, input tree and relevant parameter
A given triplet sets G=(T, r, g), wherein, T represents tree, the root node of r representative tree, g represents the virtual machine number that computer center need to provide; In tree, there are three variable function: weight (), height () and path (), wherein, weight (r) represents to take the virtual machine number that tree that r is root node can provide, height (r) represents to take the height of the tree that r is root node, and path (r) represents to take the path farthest that tree that r is root node exists;
The root node of the subtree of tree is with r nrepresent, n is positive integer, and the span of n is [1, n], the variable in subtree: weight (r n), height (r n) and path (r n), wherein, weight (r n) represent with r nfor the virtual machine number that the subtree of root node can provide, height (r n) represent with r nfor the height of the subtree of root node, path (r n) represent with r nthe path farthest existing for the subtree of root node;
Given variable minHeight, minTree, minPath, longestPath, h1 and h2, wherein, minHeight represents the height of local optimum subtree in ergodic process, minTree represents the subtree of local optimum in ergodic process, minPath represents the path of local optimum subtree in ergodic process, longestPath represents the longest path length of local optimum subtree in ergodic process, and h1 represents the first high subtree height in ergodic process, and h2 represents the second high subtree height in ergodic process;
Step 2, numerical value initialization
Initializing variable, i.e. height (r)=0, path (r)=0, height (r n)=0, path (r n)=0, minHeight is initialized as infinity, and minTree is initialized as sky, longestPath=0, h1=0, h2=0;
Step 3, traversal of tree
Adopt the method for postorder traversal, from root node r, start all nodes of traverse tree;
Step 4, record minimum subtree
Judgement weight (r nwhether)>=g sets up, if weight is (r n)>=g is false, and n=n+1, goes to step 3) continue to travel through; If weight is (r n)>=g sets up, execution step 5);
If minHeight>=height is (r n), minHeight=height (r n); If minPath>=path is (r n), minPath=path (r n); And record minimum subtree minTree=r n;
Step 5, judge whether traversal of tree finishes
Judgement height (r n) whether >h1 set up, if height is (r n) >h1 sets up, h2=h1, h1=height (r n); If height is (r n) >h1 is false, then continuation judgement height (r n) whether >h2 set up, if height is (r n) >h2 establishment, h2=height (r n);
If longestPath>path is (r n), longestPath=path (r n), obtain the shortest longest path length;
If tree not traversal is finished, n=n+1, goes to step 3) continue to travel through;
Judge whether h2+h1>longestPath sets up, if h2+h1>longestPath establishment, path (r n)=h2+h1; If h2+h1>longestPath is false, path (r n)=longestPath;
Step 6, draw optimum subtree
Judge whether minTree exists, if minTree exists, weight (r)=weight (minTree), height (r n)=height (minTree); If minTree does not exist, whether virtual machine number that must decision tree T is greater than g, if so, minTree=r, weight (r)=weight (minTree), height (r)=height (minTree), otherwise there is no optimum subtree.
Beneficial effect of the present invention:
The present invention has proposed the improvement of different emphasis on the basis of the optimum subtree of the searching based on height, and considered fault-tolerance simultaneously, first describe tree network topological structure in detail, proposed on this basis to find improving one's methods of optimum subtree, use different standards to weigh the quality of subtree, and analyzed in theory the reasonability of improved algorithm.The present invention has overcome the defect that reduces the communication cost problem between virtual machine in prior art effectively, on the basis of network architecture of determining computer center inside, select suitable region and server set wherein to provide service, the maximized communication cost having reduced between virtual machine for client requests and application program computer center is inner.
Below with reference to embodiment, the present invention is elaborated further.
Accompanying drawing explanation
The system of selection flow chart of Tu1Shi computer center internal physical main frame.
Fig. 2 is that single computer center is used algorithms of different Processing tasks figure.
Fig. 3 is a plurality of co-treatment user of computer center request figure.
Fig. 4 is that single computer center processes different request figure.
The request figure of Tu5Shi computer center processing demands different virtual machine number.
Embodiment
Embodiment 1:
The system of selection of 1 pair of computer center's internal physical main frame describes by reference to the accompanying drawings, and this system of selection comprises the steps:
Step 1, input tree and relevant parameter
A given triplet sets G=(T, r, g), wherein, T represents tree, the root node of r representative tree, g represents the virtual machine number that computer center need to provide; In tree, there are three variable function: weight (), height () and path (), wherein, weight (r) represents to take the virtual machine number that tree that r is root node can provide, height (r) represents to take the height of the tree that r is root node, and path (r) represents to take the path farthest that tree that r is root node exists;
The root node of the subtree of tree is with r nrepresent, n is positive integer, and the span of n is [1, n], the variable in subtree: weight (r n), height (r n) and path (r n), wherein, weight (r n) represent with r nfor the virtual machine number that the subtree of root node can provide, height (r n) represent with r nfor the height of the subtree of root node, path (r n) represent with r nthe path farthest existing for the subtree of root node;
Given variable minHeight, minTree, minPath, longestPath, h1 and h2, wherein, minHeight represents the height of local optimum subtree in ergodic process, minTree represents the subtree of local optimum in ergodic process, minPath represents the path of local optimum subtree in ergodic process, longestPath represents the longest path length of local optimum subtree in ergodic process, and h1 represents the first high subtree height in ergodic process, and h2 represents the second high subtree height in ergodic process;
Step 2, numerical value initialization
Initializing variable, i.e. height (r)=0, path (r)=0, height (r n)=0, path (r n)=0, minHeight is initialized as infinity, and minTree is initialized as sky, longestPath=0, h1=0, h2=0;
Step 3, traversal of tree
Adopt the method for postorder traversal, from root node r, start all nodes of traverse tree;
Step 4, record minimum subtree
Judgement weight (r nwhether)>=g sets up, if weight is (r n)>=g is false, and n=n+1, goes to step 3) continue to travel through; If weight is (r n)>=g sets up, execution step 5);
If minHeight>=height is (r n), minHeight=height (r n); If minPath>=path is (r n), minPath=path (r n); And record minimum subtree minTree=r n;
Step 5, judge whether traversal of tree finishes
Judgement height (r n) whether >h1 set up, if height is (r n) >h1 sets up, h2=h1, h1=height (r n); If height is (r n) >h1 is false, then continuation judgement height (r n) whether >h2 set up, if height is (r n) >h2 establishment, h2=height (r n);
If longestPath>path is (r n), longestPath=path (r n), obtain the shortest longest path length;
If tree not traversal is finished, n=n+1, goes to step 3) continue to travel through;
Judge whether h2+h1>longestPath sets up, if h2+h1>longestPath establishment, path (r n)=h2+h1; If h2+h1>longestPath is false, path (r n)=longestPath;
Step 6, draw optimum subtree
Judge whether minTree exists, if minTree exists, weight (r)=weight (minTree), height (r n)=height (minTree); If minTree does not exist, whether virtual machine number that must decision tree T is greater than g, if so, minTree=r, weight (r)=weight (minTree), height (r)=height (minTree), otherwise there is no optimum subtree.
On this basis, also can do further optimization to the method for the present embodiment.
(1) take farthest path selects as the first criteria optimization subtree
In said method, the height of subtree is the standard of judgement subtree quality, and the longest path length existing in subtree of usining is selected in the subtree of equal height as the second standard simultaneously.This optimization can be reduced to the communication cost between the virtual machine of user application service to a certain extent, improves the performance of application program.But the longest path length existing in subtree of take is that the first standard is weighed the good and bad of subtree, can reduce more the communication cost between virtual machine, improves greatly the performance of application program.
Select take the longest path length that exists in subtree as the raising of the first standard application programs performance more helpful, use longest path as standard, not only to provide the upper limit of communication cost between virtual machine, also provided the upper limit of processing the communication cost total amount that all virtual machines of user's request occur in whole stalk tree, user always selects the best upper limit as the standard of weighing subtree quality, therefore determines that the longest path length existing in subtree is as the first standard of weighing subtree quality.The second standard is used the height of subtree, and the subtree of Select Subtree height minimum in the identical subtree group of longest path length is processed user's request with it.
(2) take root node selects to the maximum distance of subtree as the second criteria optimization subtree
By above-mentioned analysis, determined and usingd the length of the point-to-point transmission longest path that exists in subtree as the first standard of Select Subtree, and usingd the height of subtree as the second standard.Here, analyze and to using subtree height as the reasonability of the second standard, and improve, select more suitably the second standard.
Select suitable subtree as the local of placing virtual machine, process the task of user's submission or the subtask that cloud platform distributes.There will be the virtual machine of processing same subtask to be placed in the subtree of a plurality of computer centers internal condition algorithm selection, between these virtual machines, need to intercom mutually frequently.Unique gateway of tactful Shi Rang of the present invention computer center is that top-level router root node is short as much as possible to the distance of selected subtree, with the virtual machine that is conducive to be distributed between different subtrees, communicates by letter faster.Using the root node of whole tree to the length in the path farthest of leaf node in subtree the second standard as judgement subtree quality.This path farthest is from top to bottom the longest distance that the leaf node in selected subtree arrives whole root vertex, is the upper bound of leaf node and communication port communication distance.And the standard of this dimension of subtree height can not define any communication distance, provide optimization, the root node of therefore selecting whole tree to the longest path length of leaf node in selected subtree as the second standard.Arrive here, two criterions have just all been determined.The longest path length existing in subtree is considering of transverse dimensions, and whole root vertex is considering of longitudinal dimension to the longest path length of leaf node in selected subtree.
When the inner Select Subtree of computer center is placed the virtual machine of processing subtask, if need to the selected subtree co-treatment subtask of other computer center inside, the subtree of selecting is meeting under the prerequisite of distributing virtual machine number, allows as far as possible selected subtree from a little close to communication port; If the required virtual machine in subtask is all distributed in this computer center, the subtree of selecting is a little away from root node from communication port as far as possible, leaves the computational resource near root node for subtree that communication requirement is higher.
Computer center is abstracted into tree network topological structure, and communication port is that top-level router means as root node, with root, represents.Hop (x) represents that node x is to the distance of whole root vertex.Calculate each node and to the algorithm of the distance of root node, adopt the method for breadth First traversal, algorithm flow is as follows:
The 1st step: queue Queue is initialized as sky, hop (root) ← 0;
The 2nd step: root node root enters queue Queue;
The 3rd step: if queue Queue is empty, forward the 7th step to; If queue Queue is not empty, forward the 4th step to;
The 4th step: head of the queue element goes out team, is assigned to node variable r;
The 5th step: if node r is leaf node, forward the 3rd step to; Otherwise, forward the 6th step to;
The 6th step: all child nodes x of node r, x ∈ children (r), adds 1, hop (x) ← hop (x)+1 to the distance of root node, and all child nodes x enter queue Queue; Forward the 3rd step to;
The 7th step: obtain all nodes to the distance of whole root vertex, finish.
By the present invention, obtain the height height (x) of each node x in whole tree, the length path (x) of shortest path between the leaf node existing in the subtree that the x of take is root node, as the first standard.By calculating the algorithm of each node to the distance of root node, obtain communication port root node to the hop of path farthest (the x)+height (x) of all leaf nodes in subtree, as the second standard.
Selection is from the nearest subtree of computer center's access point node, be because the subtask that the subtree of selecting is processed by and uniquely by selected subtree, processed.The subtask that if subtree is processed specifically by and only by the computer center at subtree place, processed, in the subtree of selecting so, computer center's access point node is more far better; If the subtask that subtree is processed also needs the assist process of other computer centers, better close to more from computer center's access point node so.Task S is comprised of m sub-task, needs respectively g 1, g 2..., g mindividual virtual machine.M tlv triple set of service corresponding to subtask is respectively C 1, C 2..., C m, in distributed cloud, have n computer center to be expressed as set { v 1, v 2..., v n.The tlv triple set of service that is designated the subtask of i is:
C i＝{(v 1,g 1i,s i),(v 2,g 2i,s i),…,(v n,g ni,s i)} (1)
Formula (1) meets constraints:
wherein i ∈ 1,2 ..., m}, k ∈ 1,2 ..., n} (2)
Formula (2) represents the summation of the virtual machine number that certain computer center need to provide for all subtasks.If being designated the computer center of k meets: at g kiin the situation of > 0, g ki=g i, wherein i ∈ 1,2 ..., m}.Illustrate that the k of computer center processes subtask i, the required virtual machine in subtask is all provided by the k of computer center.Only have in this case, in the equal subtree of path (x), select the maximum subtree of hop (x)+height (x), otherwise, the minimum subtree of hop (x)+height (x) all selected.
Embodiment 2: experiment and analysis
(1) find the emulation experiment of optimum subtree
In the operation simulation nucleus module DataCenterBroker of CloudSim, realize the virtual machine Placement Strategy of verifying proposition at computer center's optimum subtree algorithm of inner searching.
In simulation process, need to carry out necessary parameter setting and global configuration, in CloudSim emulation platform, All hosts configuration is the bandwidth of the internal memory of 5G, the external memory of 2TB and 1Gbps, host-processor is single core processor, and processor speed is 1100,2100 or 3200MIPS.Cpu busy percentage is in 0 situation, and node consumes 162 watts/hour of electric energy, and during CPU full load, node consumes 266 watts/hour of electric energy.The internal memory 1.5G of virtual machine in CloudSim emulation platform, external memory 100GB, bandwidth is 150Mbps, the CPU processing speed that each virtual machine needs is 270,550,650 or 1050MIPS.
For the performance of assessment algorithm, the present invention will contrast at the algorithm of proposition in 2012 with Mansoor Alicherry and two scholars of T.V.Lakshman.Two scholars of Mansoor Alicherry and T.V.Lakshman select suitable subtree according to the height of subtree in computer center inside, are referred to as Height algorithm.The longest path length that exists in subtree using herein as the first standard, using computer center gateway to the length in the path farthest of the leaf node of subtree as the second standard, thereby in the suitable subtree of the inner selection of computer center, be referred to as Heuristic algorithm.This experiment to this in two method be analyzed.
In experiment, created 1000 * 1000 grid and random user's request, user asks the virtual machine of some to be used for running application, and enters the processing selecting different subtree of algorithms of different and bears user's request.Random being distributed on this 1000 * 1000 network of Servers-all of Yi Ge computer center inside, all servers are all leaf nodes, with black circle, represent.Server on the grid line of same layer forms the brotgher of node, and father node, on the last layer grid point of leftmost node, represents with soft dot.All child nodes are connected with solid line with father node, thus the tree network structure chart of the whole computer center of bottom-up formation.Design 10 distributed cloud scenes, comprised respectively 1,2,3,4,5,6,7,8,9,10Ge computer center number.The number that each distributed cloud has server is the same.Therefore, computer center has the number of server and number that cloud has computer center is inversely proportional to.The number that in the cloud of You100Ge computer center, each computer center has server is between 50 to 100.The number that has server for each computer center of cloud of You50Ge computer center is randomly dispersed between 100 to 200.In Fig. 2, all data are all after 100 operations, to obtain the mean value of experimental data, and each experiment changes random seed, allows computer center have different number destination servers, so repeats the mean value of variable of the same race in experiment with computing result 100 times.
In this first experiment, Zhi Rangyige computer center processes user's request, and the task that user submits to is all processed by this computer center, and test does not have in the situation of communicating by letter with other computer centers, the performance that algorithm Heuristic and algorithm Height show.From Fig. 2, can go out, when single computer center is used Different Strategies processing user to ask, the communication cost total amount between the server set that algorithm Heuristic selects is less than algorithm Height.The virtual machine number of user's request is respectively 10,20,30,40,50,60,70,80,90,100, along with increasing of the virtual machine number of asking, represent to need more virtual machine to process user's request, run user trustship is in the application program of cloud platform, and the communication total amount between the server set that use algorithms of different is selected is also on the increase.
Now the virtual machine number of request is fixed on to 100, the computer center of measuring respectively by different numbers processes the communication cost total amount that the same task occurs, and computer center's number is respectively 1,2, and 3,4,5,6,7,8,9,10.100 virtual machines of user's request are provided jointly by a plurality of computer centers, and each computer center provides the virtual machine of some.While using the processing user task of computer center of different numbers in experiment, each computer center needs to provide participation Processing tasks, Heuristic algorithm and Height algorithm are identical for the virtual machine number that provides with same computer center under a scene, guarantee that the factor changing is just used two different algorithms to select suitable subtree.As can be seen from Figure 3, along with participating in, the computer center of service compute is more, and the required communication cost total amount of Processing tasks is larger, because the virtual machine that participates in calculating is distributed in different computer centers, communication distance is increased.The communication cost total amount that the same task of Heuristic algorithm process occurs is smaller than Height algorithm, and along with participating in the increase of the computer center of processing, the gap of two algorithms is also increasing gradually.
(2) build true Open Source Platform and test
The situation at server placement virtual machine of considering has blindness on emulation platform, and for the effect of verification algorithm on true cloud platform, this experiment is tested different virtual machine Placement Strategies on the cloud platform Eucalyptus that increases income simultaneously.
Eucalyptus is that a kind of cloud computing infrastructure of increasing income provides software.User can select source code or bag to install, Eucalyptus can be good in most of Linux issue versions according to.The eucalyptus in early stage is not docked with the service of Amazon, and Eucalyptus provides the interface that uses EC2 and S3 now, and user can use third party to serve easily.These Interface design are good, all existing instruments use these interfaces can with the cloud platform cooperating operation based on Eucalyptus.Eucalyptus platform also can support to operate in the operation of the virtual machine on Xen hypervisor or KVM.Different server clusters has privately owned separately internal network address, and Eucalyptus can be deployed to them in same cloud.Eucalyptus comprises Cloud Controller (CLC), Cluster Controller (CC), NodeController (NC), Walrus and five primary clusterings of Storage Controller (SC), thereby between them, uses the mutual close cooperation of soap message communication with WS-Security.
Experiment porch comprises 11 physical nodes, and each Node deployment is 4 core processors (Intel XeonE56062.13GHz), 8GB internal memory, the two 7200 SATA disks that turn 1TB, two Intel PCI-Express.All nodes connect by a gigabit ethernet switch.One of them node is as controlling node, and other nodes are used for creating virtual machine as calculating node.The power supply that each node consumes when cpu busy percentage is 0% and 100% is respectively 88W and 120W.Test when initial, wherein in 8 physical nodes, moving respectively 2-3 virtual machine, and in all the other two physical nodes, do not moving virtual machine, all physical servers are installed ubuntu operating system and the Linux2.6.3164-bit kernel that supports Xen3.3.
Use the java of TPC-W to realize analog subscriber request, representative of consumer trustship cloud put down in application program.TPC-W is a multi-level ecommerce weblication, comprising Tomcat application layer and mysql database layer.TPC does not issue the task code of benchmark program, from the unified angle of using, provide the standard criterion of benchmark program, any tester, writes out test program ownly and tests in unified platform deploy in conjunction with restrictions such as own software and hardware conditions according to standard.Use version that Tomcat-5.5.27 revised as the container of servlet and mysql-5.0.45 as backup database, thereby the application copy of establishment TPC-W.Tomcat server record service time and the statistics of each request.The virtual machine creating in Eucalyptus platform, before operation TPC-W application program, need to configure accordingly according to the appointment standard of TPC-W.
In Eucalyptus platform, create required virtual machine.After system made reflection buffer memory, virtual machine instance example becomes operation running state from suspension status.
This virtual machine instance has been assigned to the IP of 192.168.0.12 from the IP address pool (192.168.0.2-192.168.0.50) of Front end, wherein i-3E740761 is example number, in system, be unique, 172.19.1.4 is the IP of Virtual machine.Use can see that to issue orders this IP has distributed to this example.In all IP, there are three to dispense, one of them is exactly 192.168.0.12, and corresponding use-case numbering is i-3E740761.
The ssh key that use creates, to signing in in virtual machine instance, obtains a complete whole PC.Use ssh-i/home/xdcloud1/my.private ubuntu@192.168.0.12 order to log in.Whether connect and can inquiry continue to connect, input yes, signs in in the virtual machine instance of user ubuntu by name, can demonstrate the current service condition of virtual machine after logging in into.Login this virtual machine and configure accordingly and move TPW-W application program.
In Eucalyptus platform, do not have patterned monitoring software, used opening API that Eucalyptus provides and Web Service art designs and be easy to the graphic monitoring equipment of image data.Keeper is by inputting user name and password login cloud platform.Cloud computing management platform shows the virtual machine moving, and clicks the concise and to the point description that certain virtual machine can obtain the resource service condition of this virtual machine.At system mid point, certain wants to obtain the virtual machine of its resource service condition, can check the program of the resource using status in virtual machine, current operation and the resource taking thereof.
In experiment, 1000 * 1000 grid and TPC-W user's request of appointment have been created, the virtual machine that user asks some for the Servers-all of the Yi Ge computer center inside that runs application random be distributed in this 1000 * 1000 network, all servers are all leaf nodes, with black circle, represent.Server on the grid line of same layer forms the brotgher of node, and father node, on the last layer grid point of leftmost node, represents with soft dot.All child nodes are connected with solid line with father node, thus the tree network structure chart of the whole computer center of bottom-up formation.Distributed cloud scene is different, but each distributed cloud can have the number of server, is the same.In this experiment, test performance condition and the communication cost that single computer center processes client requests, the number that single computer center comprises server, is a random value between 5000 to 1000.In Fig. 4, all data are all after 100 operations, to obtain the mean value of experimental data, and each experiment changes random seed, allows computer center have different number destination servers, so repeats 100 times, and in experiment with computing result, variable of the same race is got its mean value.
In this experiment, only use Yi Ge computer center to process user's request, the task that user submits to is all processed by this computer center, test is not combined in the situation that service is provided with other computer centers, the performance that algorithm Heuristic, algorithm Height and algorithm Random show separately, statistics is deployed in the path farthest between all leaf nodes of the selected subtree of application program of cloud platform for handling part.From Fig. 4, can go out, when single computer center is used Different Strategies Select Subtree, in the selected subtree of the selection strategy of algorithm Height, between all leaf nodes, path is less than Random strategy farthest, because Random algorithm is always selected a server at random, between the server of selecting, the needs of communication path entered various level intermediary switch, and communication cost is larger in general.Heuristic method in this paper will be better than Height method.In experiment, the virtual machine number of user's request is respectively 20,40,60,80,100, along with increasing of the virtual machine number of asking, need more server to place virtual machine, need to there is more jumbo subtree to process application program, therefore selected subtree comprises more leaf node and intermediary switch, and the path farthest between all leaf nodes increases thereupon.
The virtual machine number of request is increased to 9 selections, in experiment, the virtual machine number of user's request is respectively 20, 30, 40, 50, 60, 70, 80, 90, 100, selection for each number is tested, the server number random value of computer center, span is between 5000 to 10000, this computer center is in order to provide certain virtual machine number need to select suitable subtree, measure the communication cost total amount that in selected subtree, the virtual machine on all leafy nodes occurs, change the number of servers that computer center has next time, do again same measurement, so repeat 100 times, calculate the mean value of communication cost total amount.Although algorithm is different, scene identical calculations center is identical, and the virtual machine number that provides of computer center is identical, guarantees that the factor changing is just used three different algorithm Select Subtrees.As can be seen from Figure 5, along with increasing of request virtual machine number, no matter three kinds of methods are used that Select Subtree to place virtual machine, communication cost all can increase, this is because demand virtual machine number increases, leaf node and intermediary switch in the subtree of selecting also increase, and the path farthest existing in subtree also can increase, and overall communication cost has just increased.The fault-tolerance of the task that Heuristic algorithm process is same is more better than other two algorithms, Height algorithm is better than Random algorithm, this is because Random algorithm is chosen the subtree that leaf node forms arbitrarily at random, the length of communication path farthest at inner any two leaf nodes of this subtree is usually very large, has higher blindness and directly uses the performance of this computer center there is no significant difference.This experiment user request virtual machine number be set in decimal magnitude, along with user asks virtual machine number, increase considerably, the algorithm that this experiment proposes is having the trend of continuous expansion than other two algorithms aspect the communication cost of selected subtree inside and delay.
The present invention has been done to further narration above, but the present invention is not limited to above-mentioned execution mode in conjunction with the embodiments, in the ken that one skilled in the relevant art possesses, can also under the prerequisite that does not depart from aim of the present invention, makes a variety of changes.

Claims (1)

  1. The system of selection of 1.Yi Zhong computer center internal physical main frame, is characterized in that comprising the steps:
    1) input tree and relevant parameter
    A given triplet sets G=(T, r, g), wherein, T represents tree, the root node of r representative tree, g represents the virtual machine number that computer center need to provide; In tree, there are three variable function: weight (), height () and path (), wherein, weight (r) represents to take the virtual machine number that tree that r is root node can provide, height (r) represents to take the height of the tree that r is root node, and path (r) represents to take the path farthest that tree that r is root node exists;
    The root node of the subtree of tree is with r nrepresent, n is positive integer, and the span of n is [1, n], the variable in subtree: weight (r n), height (r n) and path (r n), wherein, weight (r n) represent with r nfor the virtual machine number that the subtree of root node can provide, height (r n) represent with r nfor the height of the subtree of root node, path (r n) represent with r nthe path farthest existing for the subtree of root node;
    Given variable minHeight, minTree, minPath, longestPath, h1 and h2, wherein, minHeight represents the height of local optimum subtree in ergodic process, minTree represents the subtree of local optimum in ergodic process, minPath represents the path of local optimum subtree in ergodic process, longestPath represents the longest path length of local optimum subtree in ergodic process, and h1 represents the first high subtree height in ergodic process, and h2 represents the second high subtree height in ergodic process;
    2) numerical value initialization
    Initializing variable, i.e. height (r)=0, path (r)=0, height (r n)=0, path (r n)=0, minHeight is initialized as infinity, and minTree is initialized as sky, longestPath=0, h1=0, h2=0;
    3) traversal of tree
    Adopt the method for postorder traversal, from root node r, start all nodes of traverse tree;
    4) record minimum subtree
    Judgement weight (r nwhether)>=g sets up, if weight is (r n)>=g is false, and n=n+1, goes to step 3) continue to travel through; If weight is (r n)>=g sets up, execution step 5);
    If minHeight>=height is (r n), minHeight=height (r n); If minPath>=path is (r n), minPath=path (r n); And record minimum subtree minTree=r n;
    5) judge whether traversal of tree finishes
    Judgement height (r n) whether >h1 set up, if height is (r n) >h1 sets up, h2=h1, h1=height (r n); If height is (r n) >h1 is false, then continuation judgement height (r n) whether >h2 set up, if height is (r n) >h2 establishment, h2=height (r n);
    If longestPath>path is (r n), longestPath=path (r n), obtain the shortest longest path length;
    If tree not traversal is finished, n=n+1, goes to step 3) continue to travel through;
    Judge whether h2+h1>longestPath sets up, if h2+h1>longestPath establishment, path (r n)=h2+h1; If h2+h1>longestPath is false, path (r n)=longestPath;
    6) draw optimum subtree
    Judge whether minTree exists, if minTree exists, weight (r)=weight (minTree), height (r n)=height (minTree); If minTree does not exist, whether virtual machine number that must decision tree T is greater than g, if so, minTree=r, weight (r)=weight (minTree), height (r)=height (minTree), otherwise there is no optimum subtree.
CN201410284277.1A 2014-06-23 2014-06-23 A kind of system of selection for calculating central interior physical host Expired - Fee Related CN104065547B (en)

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 true CN104065547A (en) 2014-09-24
CN104065547B 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)

Cited By (3)

* Cited by examiner, † Cited by third party
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
CN106933654A (en) * 2017-03-17 2017-07-07 中山大学 A kind of virtual machine based on caching starts method
CN113535320A (en) * 2020-04-14 2021-10-22 深信服科技股份有限公司 Data access method, device, equipment and storage medium

Citations (3)

* Cited by examiner, † Cited by third party
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
US20130263120A1 (en) * 2012-03-30 2013-10-03 International Business Machines Corporation Virtual machine placement framework
CN103414752A (en) * 2013-07-16 2013-11-27 上海交通大学 Network-awareness cloud data center virtual machine allocation method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130263120A1 (en) * 2012-03-30 2013-10-03 International Business Machines Corporation Virtual machine placement framework
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)

* Cited by examiner, † Cited by third party
Title
MANSOOR ALICHERRY等: "Network Aware Resource Allocation in Distributed Clouds", 《2012 PROCEEDINGS IEEE INFOCOM》 *

Cited By (5)

* Cited by examiner, † Cited by third party
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
CN106933654A (en) * 2017-03-17 2017-07-07 中山大学 A kind of virtual machine based on caching starts method
CN106933654B (en) * 2017-03-17 2020-08-28 中山大学 Virtual machine starting method based on cache
CN113535320A (en) * 2020-04-14 2021-10-22 深信服科技股份有限公司 Data access method, device, equipment and storage medium
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

Also Published As

Publication number Publication date
CN104065547B (en) 2017-06-20

Similar Documents

Publication Publication Date Title
US11093298B2 (en) System and method for topology-aware job scheduling and backfilling in an HPC environment
EP3606008B1 (en) Method and device for realizing resource scheduling
Hu et al. Flutter: Scheduling tasks closer to data across geo-distributed datacenters
CA2503775C (en) High performance computing system and method
Bilal et al. Trends and challenges in cloud datacenters
JP4833965B2 (en) System and method for cluster management based on HPC architecture
JP4986844B2 (en) System and method for detecting and managing HPC node failures
WO2016101638A1 (en) Operation management method for electric power system cloud simulation platform
US20050235055A1 (en) Graphical user interface for managing HPC clusters
CN104065547A (en) A method for selecting physical hosts inside a computing center
Xu et al. Flexcloud: A flexible and extendible simulator for performance evaluation of virtual machine allocation
CN113504968A (en) Method for building power plant data center
CN105471986B (en) A kind of Constructing data center Scale Revenue Ratio method and device
CN110958192B (en) Virtual data center resource allocation system and method based on virtual switch
Meng et al. Reliability simulation in cloud computing system
Jang et al. Enhancing Node Fault Tolerance through High-Availability Clusters in Kubernetes
Pärssinen Analysis and Forming of Energy Efficiency and GreenIT Metrics Framework for Sonera Helsinki Data Center HDC
WO2017191489A1 (en) Distributed network-aware service placement utilizing constrained kernighan-lin heuristic models
Wang et al. The Application Research of Private Cloud in the Data Centers Colleges of Universities
Zhang et al. The research based on private cloud computing information interaction model
Lu et al. Research for Spatial Interpolation Based on Cloud Computing
Arjona Aroca Structural issues and energy efficiency in data centers

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
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20170620

Termination date: 20180623