CN103176849B - A kind of dispositions method of the cluster virtual machine based on resource classification - Google Patents
A kind of dispositions method of the cluster virtual machine based on resource classification Download PDFInfo
- Publication number
- CN103176849B CN103176849B CN201310078450.8A CN201310078450A CN103176849B CN 103176849 B CN103176849 B CN 103176849B CN 201310078450 A CN201310078450 A CN 201310078450A CN 103176849 B CN103176849 B CN 103176849B
- Authority
- CN
- China
- Prior art keywords
- physical machine
- node
- virtual machine
- machine
- physical
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a kind of dispositions method of the cluster virtual machine based on resource classification, comprise copy mechanism and mirror image multidiameter delay pass through mechanism two parts; Its mechanism of copy by mirror image, effectively avoids the interference that mirror file system fault is transmitted mirror image management and mirror image; Simultaneously by mirror image multidiameter delay pass through mechanism, the speed that mirror image transmits can be accelerated greatly, shorten the response time of clustered deploy(ment); The present invention effectively can utilize the physical resource of whole physical machine system by clustered deploy(ment) node selection algorithm in addition, and avoid the utilization of physical machine part resource very high, another part is then quite idle, can also realize the load balancing of whole system.
Description
Technical field
The invention belongs to Computer Service technical field, be specifically related to a kind of dispositions method of the cluster virtual machine based on resource classification.
Background technology
The flow process that cluster virtual machine is disposed generally is divided into: cluster virtual machine resource characteristic is analyzed and physical machine resource load is analyzed; The analysis of cluster virtual machine resource characteristic based on the appointment of user, also can make analysis according to system according to monitor data in the past, to determine its resource type; The analysis of physical machine resource load is to determine which physical machine is applicable to disposing the virtual machine of these clusters; Virtual machine image prepares, and this comprises the preparation of mirror configuration file, and the transmission of mirror image; Finally complete the startup of virtual machine application cluster.
Clustered deploy(ment) strategy relates to a series of process such as establishment of mirror image management, mirror image transmission, clustered deploy(ment) node selection, virtual machine.Wherein:
Mirror image management is the prerequisite that cluster virtual machine is disposed, want the efficiency that raising system disposes cluster virtual machine, many improvements can be done above mirror image, customized, mirror image as mirror image are according to resource types classified, the replication policy of mirror image, be all improve the key mechanism of system effectiveness.All mirror images are all stored in mirror site, in order to realize the safe and reliable of mirror image, are also generally adopt mirror image copies strategy.Current many products that can be used for doing mirror site, as NFS(network file system(NFS)), financial telecommunications association of SWIFT(global cooperative bank) system, can not only conserve space, and improve efficiency.
Cluster virtual machine is disposed node selection and is referred to from physical machine system, to select some physical machine nodes to be used for disposing virtual machine.Current had multiple deploying virtual machine selection strategy, the strategy of general employing comprises the greedy selection strategy of disposing of order and disposes selection strategy with balanced, no matter be adopt which kind of strategy, all need the relevant information obtaining candidate physical machine from information or Performance Center, comprise CPU, internal memory, the network bandwidth, I/O service condition.
The establishment of cluster virtual machine comprises the startup of virtual machine configuration generation and virtual machine, required some configuration parameters used when virtual machine configuration refers to virtual machine activation, comprise memory requirements, virtual machine UUID(global unique identification symbol), CPU core number, image file deposit position, network configuration information etc.The startup of virtual machine comprises establishment configuration file, copies image file, calls virtual machine platform interface.
A good clustered deploy(ment) strategy needs to meet user's request with the shortest time, even if user's request is assigned in corresponding physical machine within the minimized time; Maximum system throughput, makes the resource utilization of system reach maximum; Be with good expansibility; Minimize clustered deploy(ment) and operate the overhead brought to system.Faced by existing dispositions method still there are some defects in new technological challenge, number of patent application be 201110401608.1 Chinese patent application disclose a kind of dispositions method and device of virtual machine, method comprises: receive the request disposing virtual machine, carries the virtual machine image file mark of disposing virtual machine and using in described deployment request; Obtain the storage information of corresponding virtual machine image file in distributed file system according to described virtual machine image file mark, described distributed file system is stored by this locality of multiple computing node and forms; According to the load information of described storage information and described multiple computing node, select the computing node disposing described virtual machine, and the computing node deploy virtual machine selected.The process employs distributed file system, then these file system are when the huge capacity demand of numerous cluster mirror image, be difficult to meet the demands, and when disposing cluster, traditional distributed file system, when setting up multiple transmission to a mirror image and connecting, is easy to become bottleneck; Meanwhile, distributed file system, once run into fault, very likely causes the loss of image file; In addition, only investigate the utilization factor of physical nodes to a certain resource during traditional clustered deploy(ment), fully do not analyze the resource characteristic of cluster, therefore can not make full use of the various resource of computing machine.
Summary of the invention
For the above-mentioned technical matters existing for prior art, the invention provides a kind of dispositions method of the cluster virtual machine based on resource classification, adopt the strategy of resource classification, cluster virtual machine is divided into different resource type, the load balancing of whole physical machine system can be realized.
Based on a dispositions method for the cluster virtual machine of resource classification, comprise the steps:
(1) embody rule corresponding to cluster virtual machine, determines the resource type of cluster virtual machine;
(2) according to the resource type of cluster virtual machine, choose k physical machine node successively from physical machine system, and virtual machine each in cluster virtual machine is distributed to this k physical machine node respectively, k is the number of virtual machine in cluster virtual machine;
(3) configuration file corresponding for cluster virtual machine and image file are passed to each physical machine node selected from template base.
Described resource type has three classes, is respectively computation-intensive, stores intensive and traffic-intensive type.
In described step (2), from physical machine system, choose physical machine node and virtual machine to be distributed to the method for physical machine node as follows:
A. according to the resource type of cluster virtual machine, the load information value F of every platform physical machine in physical machine system is calculated;
B. for arbitrary physical machine in physical machine system, calculate the load information value L of this physical machine, judge whether its load information value L is greater than given overloading threshold, if so, then eliminate this physical machine, if not, then retain this physical machine; Travel through every platform physical machine according to this;
C. from all physical machine retained, choose the maximum physical machine of load information value F as a physical machine node, and appoint from cluster virtual machine and get a virtual machine and distribute to this physical machine node;
D. return and perform step a, cycling is until be assigned virtual machine each in cluster virtual machine.
If the resource type of described cluster virtual machine is computation-intensive, then load information value F tries to achieve according to following formula:
F=α(1-c)+β(m+n)+node*γ
If the resource type of described cluster virtual machine is intensive for storing, then load information value F tries to achieve according to following formula:
F=α(1-m)+β(c+n)+node*γ
If the resource type of described cluster virtual machine is traffic-intensive type, then load information value F tries to achieve according to following formula:
F=α(1-n)+β(c+m)+node*γ
Wherein: c, m and n are respectively the current cpu busy percentage of physical machine, memory usage and network bandwidth utilization factor, the virtual machine number of node for present physical machine is loaded with, α, β and γ are given weight coefficient and are practical experience value.
Described load information value L tries to achieve according to following formula:
L=a
1c+a
2m+a
3n+node*γ
Wherein: c, m and n are respectively the current cpu busy percentage of physical machine, memory usage and network bandwidth utilization factor, the virtual machine number of node for present physical machine is loaded with, a
1, a
2, a
3given weight coefficient is and for practical experience value with γ.
Preferably, in described step (3), method image file being passed to each physical machine node selected from template base is as follows:
A. the traffic load value T of each physical machine node selected is calculated;
B. for arbitrary physical machine node, judge whether its traffic load value T is greater than given load threshold, if so, then retain this physical machine node, if not, then eliminate this physical machine node; Travel through each physical machine node selected according to this;
C. build a transmit queue, the physical machine node retained is deposited in described transmit queue by traffic load value T putting in order from small to large;
If D. image file has i copy in template base, then from transmit queue, extract front i+1 the physical machine node of arrangement, and the image file in template base and i copy thereof are passed to this i+1 physical machine node respectively; After end of transmission, i+1 the physical machine node then making the image file in template base and i copy thereof and obtain image file is all as transmission sources, from transmit queue, extract front 2i+2 the physical machine node of arrangement again, make 2i+2 transmission sources transmit image file respectively to this 2i+2 physical machine node; According to this propagate until in transmit queue each physical machine node all obtain image file, i be greater than 0 natural number;
E. for the physical machine node eliminated in step B, then these physical machine nodes are made to obtain image file or its copy by transmitting from template base.
Described traffic load value T tries to achieve according to following formula:
T=a
4C+a
5N
Wherein: C and N is respectively the cpu busy percentage and network bandwidth utilization factor that corresponding to physical machine node, physical machine is current, a
4and a
5be given weight coefficient and for practical experience value.
Advantageous Effects of the present invention is as follows:
(1) the present invention is by the copy mechanism of mirror image, effectively avoids the interference that mirror file system fault is transmitted mirror image management and mirror image;
(2) the present invention is by mirror image multidiameter delay pass through mechanism, can accelerate the speed that mirror image transmits greatly, shortens the response time of clustered deploy(ment);
(3) the present invention effectively can utilize the physical resource of whole physical machine system by clustered deploy(ment) node selection algorithm, and avoid the utilization of physical machine part resource very high, another part is then quite idle, can also realize the load balancing of whole system.
Accompanying drawing explanation
Fig. 1 is the steps flow chart schematic diagram of dispositions method of the present invention.
Fig. 2 is the schematic diagram that mirror image of the present invention transmits.
Fig. 3 is that the experimental result of mirror image transmission method of the present invention and conventional image transmission method contrasts schematic diagram.
Fig. 4 (a) is for cluster distribution method of the present invention and conventional greedy distribution method are about the contrast schematic diagram of cpu data.
Fig. 4 (b) is for cluster distribution method of the present invention and conventional greedy distribution method are about the contrast schematic diagram of internal storage data.
Fig. 4 (c) is for cluster distribution method of the present invention and conventional greedy distribution method are about the contrast schematic diagram of network bandwidth data.
Fig. 4 (d) is for cluster distribution method of the present invention and conventional greedy distribution method are about the contrast schematic diagram of I/O data.
Embodiment
In order to more specifically describe the present invention, below in conjunction with the drawings and the specific embodiments, the inventive method is described in detail.
As shown in Figure 1, a kind of dispositions method of the cluster virtual machine based on resource classification, comprises the steps:
(1) embody rule corresponding to cluster virtual machine, determines the resource type of cluster virtual machine;
Resource type has three classes, is respectively computation-intensive, stores intensive and traffic-intensive type;
Computation-intensive concentrates on Distributed Calculation, parallel computation, real-time calculating, typical application comprises: the Distributed Computing Platform of MapRedcue(Google), BOINC(Berkeley open network computing platform), CORBA(Common Object Request Broker Architecture), the distributed paralleling calculation platform of Dryad(Microsoft).
Data-intensive application mainly concentrates on mass file field of storage and cache field, and typical data-intensive applications comprises: the distribution file storage system of GFS(Google), the mass data distributed file system of CEPH(Linux), the distributed file system of HDFS(Hadoop), Memcached(distributed memory target cache system), the distributed cache system of Membase(NoSQL family).
Traffic-intensive type calculates and mainly concentrates on the calculating of magnanimity streaming and large-scale complex event handling, the StreamInsight that typical application is released before being Microsoft.
These application class above-mentioned are all the resource types judging application in experience, when an application is new opplication, just need to judge according to operational effect, rule of thumb show, the average utilization of the CPU of various application, internal memory, the network bandwidth, I/O can be used; After the utilization factor testing out the CPU of the application that will dispose, internal memory, the network bandwidth, I/O, deduct their average utilization respectively, there is the type of mxm., just can be defined as the resource type applied.
(2) according to the resource type of cluster virtual machine, choose k physical machine node successively from physical machine system, and virtual machine each in cluster virtual machine is distributed to this k physical machine node respectively, k is the number of virtual machine in cluster virtual machine; In present embodiment, in cluster virtual machine to be disposed, the number k of virtual machine is 24.
Generally before deployment, first filter clustered deploy(ment) request, if calculate physical machine system because the restriction of the aspects such as CPU, internal memory, file size, the network bandwidth by algorithm, cause when can not complete this clustered deploy(ment), should filter and this time dispose request, and to user feedback.Concrete deployment implementation is as follows:
A. according to the resource type of cluster virtual machine, the load information value F of every platform physical machine in physical machine system is calculated;
If the resource type of cluster virtual machine is computation-intensive, then load information value F tries to achieve according to following formula:
F=α(1-c)+β(m+n)+node*γ
If the resource type of cluster virtual machine is intensive for storing, then load information value F tries to achieve according to following formula:
F=α(1-m)+β(c+n)+node*γ
If the resource type of cluster virtual machine is traffic-intensive type, then load information value F tries to achieve according to following formula:
F=α(1-n)+β(c+m)+node*γ
Wherein: c, m and n are respectively the current cpu busy percentage of physical machine, memory usage and network bandwidth utilization factor, the virtual machine number of node for present physical machine is loaded with, α, β and γ are given weight coefficient; α=β=0.5 in present embodiment, γ=0.2.
B. for arbitrary physical machine in physical machine system, calculate the load information value L of this physical machine, judge whether its load information value L is greater than given overloading threshold, if so, then eliminate this physical machine, if not, then retain this physical machine; Travel through every platform physical machine according to this, in present embodiment, overloading threshold is set as 83;
Load information value L tries to achieve according to following formula:
L=a
1c+a
2m+a
3n+node*γ
Wherein: a
1, a
2and a
3be given weight coefficient, a in present embodiment
1=0.4, a
2=a
3=0.3.
C. from all physical machine retained, choose the maximum physical machine of load information value F as a physical machine node, and appoint from cluster virtual machine and get a virtual machine and distribute to this physical machine node, simultaneously by this physical machine node city one priority query;
D. return and perform step a, cycling is until be assigned virtual machine each in cluster virtual machine.
(3) configuration file corresponding for cluster virtual machine and image file are passed to each physical machine node selected from template base;
Wherein, because configuration file is generally smaller, so physical machine node can directly be acquired by transmission from template base; And image file is general comparatively large, therefore present embodiment adopts following transfer mode from template base, image file will be passed to each physical machine node:
The traffic load value T of each physical machine node A. selected according to following formulae discovery;
T=a
4C+a
5N
Wherein: C and N is respectively the cpu busy percentage and network bandwidth utilization factor that corresponding to physical machine node, physical machine is current, a
4and a
5be given weight coefficient, a in present embodiment
4=a
5=0.5.
B. for arbitrary physical machine node, judge whether its traffic load value T is greater than given load threshold, if so, then retain this physical machine node, if not, then eliminate this physical machine node; Travel through each physical machine node selected according to this; In present embodiment, load threshold is set as 0.7, remains 22, eliminated 2 in 24 physical machine nodes.
C. build a transmit queue, the physical machine node retained is deposited in transmit queue by traffic load value T putting in order from small to large;
D. as shown in Figure 2, if image file has i copy (in present embodiment i=2) in template base, from transmit queue, then extract front i+1 the physical machine node of arrangement, and the image file in template base and i copy thereof are passed to this i+1 physical machine node respectively; After end of transmission, i+1 the physical machine node then making the image file in template base and i copy thereof and obtain image file is all as transmission sources, from transmit queue, extract front 2i+2 the physical machine node of arrangement again, make 2i+2 transmission sources transmit image file respectively to this 2i+2 physical machine node; Propagate until each physical machine node all obtains image file in transmit queue according to this;
E. for 2 the physical machine nodes eliminated in step B, then these 2 physical machine nodes are made to obtain image file and copy 1 thereof respectively by transmitting from template base.
Finally start all virtual machines in cluster virtual machine, complete clustered deploy(ment).
We make present embodiment and traditional NFS single-point mirror image transmission and Swift mirror image dispose transmission to compare by experiment below, and these two kinds of control methodss are all the modes that cloud platform generally uses.When after the test result obtaining present embodiment mirror image parallel duplex transmission policy and traditional NFS mirror image transmission policy and Swift mirror site, the time of contrast three.In checking, the size of image file is 3G, mirror image quantity is respectively 1,2,4,5,6,7,8,16,32,64, when the copy of present embodiment and conventional image transmission method is all set to 2, passing time experimental result as shown in Figure 3, horizontal ordinate is mirror image quantity, and ordinate is passing time.
When transmit quantity be all 1 time, the mirror image passing time of present embodiment and the mirror image passing time of NFS system all similar; When mirror image transmission quantity is less, the mirror image way to manage performance of Swift mirror image way to manage and present embodiment is all similar; But when the number ratio transmitted is larger, the mirror image passing time of NFS system and the linear growth of quantity, the passing time of Swift mirror site also increases comparatively large, and the mirror image multidiameter delay passing time of present embodiment is then that logarithmic relationship increases.Therefore, experimental result shows, when mirror image transmit quantity larger time, the mirror image transmission policy of present embodiment has more excellent transmission performance for the deployment of large-scale virtual machine cluster.
We compare with regard to CPU, internal memory, the network bandwidth, I/O tetra-kinds of data the allocation algorithm of present embodiment and conventional greedy allocation algorithm below, as shown in Figure 4.According to these four groups of correlation datas, can obtain the CPU of two kinds of allocation algorithms, internal memory, the network bandwidth, the data of I/O and the error mean values of mean value, every numerical value is as shown in table 1:
Table 1
| Greedy allocation algorithm | Present embodiment | |
| CPU | 13.61375 | 5.6125 |
| Internal memory | 21.6258 | 1.7115 |
| The network bandwidth | 7.6295 | 0.24575 |
| I/O | 2.275 | 0.40345 |
As can be seen from the table, the cluster allocation algorithm based on resource classification that present embodiment proposes, compared with the greedy allocation algorithm of order, the load performance of whole physical machine system can be made to be in optimum condition, and every resource makes sufficient utilization.
Claims (1)
1., based on a dispositions method for the cluster virtual machine of resource classification, comprise the steps:
(1) embody rule corresponding to cluster virtual machine, determines the resource type of cluster virtual machine;
(2) according to the resource type of cluster virtual machine, choose k physical machine node successively from physical machine system, and virtual machine each in cluster virtual machine is distributed to this k physical machine node respectively, k is the number of virtual machine in cluster virtual machine; Specific implementation process is as follows:
2.1, according to the resource type of cluster virtual machine, calculate the load information value F of every platform physical machine in physical machine system;
If the resource type of described cluster virtual machine is computation-intensive, then load information value F tries to achieve according to following formula:
F＝α(1-c)+β(m+n)+node*γ
If the resource type of described cluster virtual machine is intensive for storing, then load information value F tries to achieve according to following formula:
F＝α(1-m)+β(c+n)+node*γ
If the resource type of described cluster virtual machine is traffic-intensive type, then load information value F tries to achieve according to following formula:
F＝α(1-n)+β(c+m)+node*γ
Wherein: c, m and n are respectively the current cpu busy percentage of physical machine, memory usage and network bandwidth utilization factor, the virtual machine number of node for present physical machine is loaded with, α, β and γ are given weight coefficient;
2.2 for arbitrary physical machine in physical machine system, goes out the load information value L of this physical machine according to following formulae discovery, judges whether its load information value L is greater than given overloading threshold, if so, then eliminates this physical machine, if not, then retain this physical machine; Travel through every platform physical machine according to this;
L＝a
1c+a
2m+a
3n+node*γ
Wherein: a
1, a
2and a
3be given weight coefficient, c, m and n are respectively the current cpu busy percentage of physical machine, memory usage and network bandwidth utilization factor, the virtual machine number of node for present physical machine is loaded with, and γ is given weight coefficient;
2.3 choose the maximum physical machine of load information value F as a physical machine node from all physical machine retained, and appoint from cluster virtual machine and get a virtual machine and distribute to this physical machine node;
2.4 return execution step 2.1, and cycling is until be assigned virtual machine each in cluster virtual machine;
(3) configuration file corresponding for cluster virtual machine and image file are passed to each physical machine node selected from template base; The method wherein image file being passed to each physical machine node selected is as follows:
The traffic load value T of the 3.1 each physical machine nodes selected according to following formulae discovery;
T＝a
4C+a
5N
Wherein: C and N is respectively the cpu busy percentage and network bandwidth utilization factor that corresponding to physical machine node, physical machine is current, a
4and a
5be given weight coefficient;
3.2 for arbitrary physical machine node, judges whether its traffic load value T is greater than given load threshold, if so, then retains this physical machine node, if not, then eliminates this physical machine node; Travel through each physical machine node selected according to this;
3.3 build transmit queues, deposit in described transmit queue by the physical machine node retained by traffic load value T putting in order from small to large;
If image file has i copy in 3.4 template base, then from transmit queue, extract front i+1 the physical machine node of arrangement, and the image file in template base and i copy thereof are passed to this i+1 physical machine node respectively; After end of transmission, i+1 the physical machine node then making the image file in template base and i copy thereof and obtain image file is all as transmission sources, from transmit queue, extract front 2i+2 the physical machine node of arrangement again, make 2i+2 transmission sources transmit image file respectively to this 2i+2 physical machine node; According to this propagate until in transmit queue each physical machine node all obtain image file, i be greater than 0 natural number;
3.5 for the physical machine node eliminated in step 3.2, then make these physical machine nodes obtain image file or its copy by transmitting from template base.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310078450.8A CN103176849B (en) | 2013-03-12 | 2013-03-12 | A kind of dispositions method of the cluster virtual machine based on resource classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310078450.8A CN103176849B (en) | 2013-03-12 | 2013-03-12 | A kind of dispositions method of the cluster virtual machine based on resource classification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103176849A CN103176849A (en) | 2013-06-26 |
| CN103176849B true CN103176849B (en) | 2016-04-27 |
Family
ID=48636749
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310078450.8A Active CN103176849B (en) | 2013-03-12 | 2013-03-12 | A kind of dispositions method of the cluster virtual machine based on resource classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103176849B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103580923B (en) * | 2013-11-08 | 2017-02-15 | 华为技术有限公司 | Software deploying method and device |
| CN103902353A (en) * | 2014-03-31 | 2014-07-02 | 华为技术有限公司 | Virtual machine deployment method and device |
| CN105519075A (en) * | 2014-06-05 | 2016-04-20 | 华为技术有限公司 | Resource scheduling method and apparatus |
| CN105867998A (en) * | 2016-03-24 | 2016-08-17 | 国云科技股份有限公司 | Virtual machine cluster deployment algorithm |
| CN105912382A (en) * | 2016-04-07 | 2016-08-31 | 浪潮电子信息产业股份有限公司 | Mirror image management device, system and method |
| CN106951559B (en) * | 2017-03-31 | 2020-08-25 | 联想(北京)有限公司 | Data recovery method in distributed file system and electronic equipment |
| CN107203394B (en) * | 2017-04-18 | 2021-01-01 | 中国科学院计算技术研究所 | High-safety docker container batch deployment method and device |
| CN107276835A (en) * | 2017-08-11 | 2017-10-20 | 郑州云海信息技术有限公司 | A kind of clustered deploy(ment) collocation method and system |
| CN108536729B (en) * | 2018-02-24 | 2022-06-14 | 国家计算机网络与信息安全管理中心 | Cross-partition mirror image file synchronization method and device |
| CN109558214B (en) * | 2018-12-05 | 2021-06-15 | 腾讯科技（深圳）有限公司 | Host machine resource management method and device in heterogeneous environment and storage medium |
| CN111158909B (en) * | 2019-12-27 | 2023-07-25 | 中国联合网络通信集团有限公司 | Cluster resource allocation processing method, device, equipment and storage medium |
| CN116582546A (en) * | 2023-07-12 | 2023-08-11 | 深圳市智博通电子有限公司 | Network intercommunication method of virtual machine cluster based on super fusion node |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101311917A (en) * | 2007-05-24 | 2008-11-26 | 中国科学院过程工程研究所 | Particle model faced multi-tier direct-connection cluster paralleling computing system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8621477B2 (en) * | 2010-10-29 | 2013-12-31 | International Business Machines Corporation | Real-time monitoring of job resource consumption and prediction of resource deficiency based on future availability |
-
2013
- 2013-03-12 CN CN201310078450.8A patent/CN103176849B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101311917A (en) * | 2007-05-24 | 2008-11-26 | 中国科学院过程工程研究所 | Particle model faced multi-tier direct-connection cluster paralleling computing system |
Non-Patent Citations (1)
| Title |
|---|
| 庄威等.云环境下基于多属性层次分析的虚拟机部署与调度策略.《西安交通大学学报》.2013, * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103176849A (en) | 2013-06-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103176849B (en) | A kind of dispositions method of the cluster virtual machine based on resource classification | |
| Liu et al. | Adaptive asynchronous federated learning in resource-constrained edge computing | |
| CN105049268B (en) | Distributed computing resource distribution system and task processing method | |
| CN110134495B (en) | Container cross-host online migration method, storage medium and terminal equipment | |
| CN102724277B (en) | The method of live migration of virtual machine and deployment, server and group system | |
| CN101957863B (en) | Data parallel processing method, device and system | |
| CN101296114B (en) | Parallel pattern matching method and system based on stream | |
| Abad et al. | Package-aware scheduling of faas functions | |
| CN110308984B (en) | Cross-cluster computing system for processing geographically distributed data | |
| CN103595780B (en) | Cloud computing resource scheduling method based on the weight that disappears | |
| CN104881322B (en) | A kind of cluster resource dispatching method and device based on vanning model | |
| CN105138679B (en) | A kind of data processing system and processing method based on distributed caching | |
| CN108900626B (en) | Data storage method, device and system in cloud environment | |
| Ubarhande et al. | Novel data-distribution technique for Hadoop in heterogeneous cloud environments | |
| Wang et al. | Hybrid pulling/pushing for i/o-efficient distributed and iterative graph computing | |
| Rupprecht et al. | SquirrelJoin: Network-aware distributed join processing with lazy partitioning | |
| WO2020134364A1 (en) | Virtual machine migration method, cloud computing management platform, and storage medium | |
| CN109614227A (en) | Task resource concocting method, device, electronic equipment and computer-readable medium | |
| CN103761146A (en) | Method for dynamically setting quantities of slots for MapReduce | |
| US20220300323A1 (en) | Job Scheduling Method and Job Scheduling Apparatus | |
| Chen et al. | Latency minimization for mobile edge computing networks | |
| Schneider et al. | Dynamic load balancing for ordered data-parallel regions in distributed streaming systems | |
| Li et al. | Replica-aware task scheduling and load balanced cache placement for delay reduction in multi-cloud environment | |
| Li et al. | Neighborhood search-based job scheduling for IoT big data real-time processing in distributed edge-cloud computing environment | |
| US20210390405A1 (en) | Microservice-based training systems in heterogeneous graphic processor unit (gpu) cluster and operating method thereof |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |