CN106550036B - One kind is towards energy-efficient heuristic cloud computing resources distribution and dispatching method - Google Patents
One kind is towards energy-efficient heuristic cloud computing resources distribution and dispatching method Download PDFInfo
- Publication number
- CN106550036B CN106550036B CN201610966411.5A CN201610966411A CN106550036B CN 106550036 B CN106550036 B CN 106550036B CN 201610966411 A CN201610966411 A CN 201610966411A CN 106550036 B CN106550036 B CN 106550036B
- Authority
- CN
- China
- Prior art keywords
- scheduling
- server
- schedule
- task
- initial
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/4557—Distribution of virtual machine instances; Migration and load balancing
Abstract
The invention discloses one kind towards energy-efficient heuristic cloud computing resources distribution and dispatching method, this method comprises: being that the task requests that all users submit generate a task initial schedule first with first-fit algorithm FF or improved optimal adaptation algorithm MBFD;Secondly, proposing a kind of heuristic mutation operations optimization method, iterative optimization initial schedule is final to obtain the scheduling that significantly reduced consumption of data center, improve user's request receptance.The task requests that the present invention can be issued effectively for cloud computing user search out the approximate optimal solution of resource allocation scheduling, effectively reduce consumption of data center.
Description
Technical field
The invention belongs to computer field, it is absorbed in cloud computing resources distribution and Mission Scheduling, more particularly to a kind of
Towards energy-efficient heuristic cloud computing resources distribution and dispatching method.
Background technique
There are considerable cloud computing resources distribution and dispatching method at present, specific as follows:
Using dynamic voltage frequency regulation technology (Dynamic Voltage and Frequency Scaling, abbreviation
) and dynamic power management technology (Dynamic Power Management, abridge DPM) DVFS.They are all a kind of hardware skills
Art.The processor of part server supports DVFS technology at present, it refers to when processor utilization is lower, reduces processor
Voltage or frequency, reduce the performance of processor, keep it in low power consumpting state;And when the load of processor is higher, it mentions
The voltage or frequency of high disposal device, improve the performance of processor, and at this point, the power consumption of processor is also relatively high.By this
Load for processor carries out the dynamic mode for adjusting processor voltage or frequency, realizes the reduction of energy consumption.DPM technology is
Refer to when server component does not load, it is switched to dormant state or closed state, to realize energy conservation.But at present very
The multi-vendor stabilization to guarantee server performance, is not equipped with these technologies, therefore, in the enterprise for being assembled with common server
Or in the private clound inside group, these technologies can not be applied.
Energy consumption is reduced by server Dynamic Integration.When lower due to server resource utilization rate, can still consume compared with
High energy consumption, therefore server resource utilization rate can be improved, energy resource efficiency is improved, energy consumption is reduced.Virtualization technology
Appearance allows virtual machine to carry out dynamic migration, and the integration of server may be implemented using this technology, low-load is serviced
On virtual machine (vm) migration to other low-load servers on device, idling can be carried to server closing, and achievement unit sub-service in this way
Device resource utilization of being engaged in improves.Although server resource utilization rate can be improved in this server Dynamic Integration technology, empty
Quasi- machine migration necessarily will affect user experience, and such as the extension of task execution time, therefore this method is not appropriate for being applied to one
The higher scene of a little time-constrains.
Summary of the invention
The object of the present invention is to provide it is a kind of it is didactic towards energy-efficient cloud resource for giving dispatching method, this method energy
Enough effective task requests issued for cloud computing user search out the approximate optimal solution of resource allocation scheduling, effectively reduce number
According to power consumption.
Realizing the specific technical solution of the object of the invention is:
One kind is that this method includes in detail below towards energy-efficient heuristic cloud computing resources distribution and dispatching method, feature
Step:
Step 1: initial phase
It the use of improved optimal adaptation method MBFD or for the first time adaptive method FF is cloud to accelerate the generation of task schedule
It calculates all user tasks that data center receives and generates initial schedule.
Step 2: the optimizing phase
Several mutually independent schedule optimizers are instantiated, each schedule optimizer is initial with what is generated in step 1 respectively
It is scheduling to input and independently carries out local optimum, all independent schedule optimizer cooperations generate the approximation in global search space
Optimal scheduling;Wherein, it specifically includes:
(1) single schedule optimizer optimization: single schedule optimizer is based on the initial schedule inputted, iterative benefit
Local optimum is carried out to it with task choosing, server grouping and scheduling three operations of evolution;
(2) more schedule optimizer cooperative optimizations: after each schedule optimizer generates the scheduling of its local optimum, all
It relatively and selects energy consumption minimum and the highest scheduling of task receptance in the scheduling that schedule optimizer generates, is adjusted as initial
Degree carries out the operation of step (1);After each schedule optimizer generates the scheduling of its local optimum, produced in all schedule optimizers
It relatively and selects energy consumption minimum and the highest scheduling of task receptance in raw scheduling, as initial schedule, then is walked
Suddenly the operation of (1);In cycles, when operating procedure (1) number reaches 10-15 times, the approximation in global search space is obtained
Optimal scheduling.
The generation initial schedule, that is, the original allocation for generating a physical server to user task map.
The list schedule optimizer, according to initial schedule, determines that data center is each use after receiving initial schedule
The server resource of family task distribution, meanwhile, each server and the distributing to it of the task constitute the part on the server
Scheduling.
The task choosing operation is that each server selects a goal task using initial schedule: in each clothes
Be engaged on device, be Overlapped Execution time on each task computation it and the server between all other task and, there is minimum
The task of Overlapped Execution time sum is selected as the goal task of the server.
Data center server is carried out random grouping two-by-two, obtains two different services by the server division operation
Device group;Using the initial schedule of input step 1, the local scheduling information in every server group on two servers constitutes the service
The local scheduling of device group.
The scheduling, which is developed, to be operated, and in every server group, execution evolution operation generates 4 different filial generations and dispatches;Often
A filial generation scheduling generates as follows:
(1) it is dispatched the initial schedule of step 1 as first filial generation;
(2) goal task of first server is moved on second server, and modified corresponding in initial schedule
Second filial generation scheduling is obtained in the local scheduling information of the operation;
(3) goal task of second server is moved on first server, and modified corresponding in initial schedule
Third filial generation scheduling is obtained in the local scheduling information of the operation;
Goal task is exchanged between (4) two servers, and modifies the local scheduling for corresponding to the operation in initial schedule
Information obtains the 4th filial generation scheduling;
After generating 4 filial generation scheduling, single schedule optimizer emulates each filial generation scheduling, calculates consumption information and appoints
Business receptance, and select energy consumption minimum and the original initial scheduling of the highest filial generation scheduling substitution of task receptance.
The iterative utilization task choosing, server grouping and scheduling three number of operations of evolution are 30-50 times.
List schedule optimizer of the present invention is an iterative operation, it selects energy consumption minimum, task connects after an iteration
By the highest filial generation scheduling of rate as new initial schedule, operation that iterative task selection, server are grouped and scheduling is developed, until
Certain the number of iterations terminates.
The more schedule optimizer cooperative optimizations of the present invention need each independent progress randomized grouping behaviour of schedule optimizer
Make, just guarantees to search for approximate optimal schedule in global space in this way.
The beneficial effects of the present invention are: the present invention can be in the pact of cloud computation data center fixed physical server resource
Approximate optimal schedule is searched under the conditions of beam, is greatly reduced energy consumption, is improved user's request receptance.
Detailed description of the invention
Fig. 1 is flow chart of the present invention;
Fig. 2 is that server of the present invention is grouped schematic diagram;
Fig. 3 is present invention scheduling evolution operational flowchart;
Fig. 4 is that the single schedule optimizer of the present invention optimizes implementation procedure figure.
Specific embodiment
In conjunction with following specific embodiments and attached drawing, the present invention is described in further detail.
The present invention is a kind of towards energy-efficient heuristic cloud resource distribution and dispatching method, the task requests submitted for user
Generate energy-efficient scheduling.Initial schedule is generated by initial method first, then by evolutionary optimization process to initial schedule
Optimize: for server selection target task, server being grouped at random, and to each grouping generate local scheduling after
In generation, replaces the scheduling in the original grouping by emulation selection optimal partial scheduling.Multiple optimization actuators work independently, and find
Global optimum's scheduling.
The present invention includes the following steps that detailed process is as shown in Figure 1:
Step 1: initial phase
In order to accelerate the generation of task schedule, fit using for example improved optimal adaptation method (MBFD) of existing method or for the first time
Induction method (FF) is that all user tasks that cloud computation data center receives generate initial schedule.
Step 2: the optimizing phase
The multiple mutually independent schedule optimizers of the phase instanceization, each schedule optimizer in step 1 respectively to generate
Initial schedule be that input independently carries out local optimum, all schedule optimizers cooperations generate the approximation in global search spaces
Optimal scheduling;It specifically includes:
(1) single schedule optimizer optimization: single schedule optimizer is based on the initial schedule that the stage inputs, iteration
Three operations of evolution that are grouped and dispatch using task choosing, server of formula carry out local optimum to it.
(2) more schedule optimizer cooperative optimizations: after each schedule optimizer generates the scheduling of its local optimum, all
In the scheduling that schedule optimizer generates relatively and the scheduling that selects energy consumption minimum, it is used as to initial schedule again, carries out step
(1) operation;After each schedule optimizer generates the scheduling of its local optimum, compare in the scheduling that all schedule optimizers generate
Compared with and select energy consumption minimum and the highest scheduling of task receptance, as initial schedule, then carry out the operation of step (1);
In cycles, when operating procedure (1) number reaches 10-15 times, the approximate optimal schedule in global search space is obtained.
Wherein, initial phase in step 1: data center's scheduler program needs to request to produce for received all users
A raw scheduling, that is, generate the mapping pair of user's request to server, each user request is mapped to a service
On device.But if enumerating all combinations of user's request to server mappings, this is a kind of nonsensical way, therefore is needed
Want some way to for one energy-saving distribution of generation as fast as possible.Using existing certain methods, such as optimal adaptation method
MBFD or for the first time adaptive method FF, for all users request generate an initial schedule, however usually both initial methods
The scheduling of generation still remains very big optimization space, if energy consumption is still very high, cannot all receive user's request, therefore can be with
It takes certain methods to advanced optimize initial schedule, achieves the optimal of these targets.
Wherein, the task choosing in (1) of step 2: server S 1 and S2 there are two in some current server group, two
All several tasks of original allocation on server.A goal task is selected for each server with the following method: every
On a server, be each task computation it the sum of with the parallel execution time of other all tasks, for example, some task 1
Execute time interval be [], the execution time interval of task 2 be [], if meeting relationship between them,And, then the two tasks there are one section it is parallel execute time interval [], they and
Row execute the time be.There is also other parallel execution relationships between two tasks, no longer enumerate here.When being each
Task computation goes out after it executes the sum of time parallel with other all tasks, and parallel the sum of time the smallest task that executes is selected to make
For goal task.This is based on such idea, if there are the parallel of a task and other tasks on some server
It is shorter to execute the time, then this server is larger a possibility that resource utilization is lower during this task execution,
So this task immigration is gone out, the server may be made to reach dormant state in this period.Meanwhile with other tasks
Parallel when executing on time shortest task immigration to other servers, a possibility that making other server overloads, is relatively low.It is logical
A goal task can be selected for each server by crossing above procedure.As shown in figure 3, server S 1 has selected task k, service
Device S2 selects task m.
Wherein, in (1) of step 2 server grouping: due to user request to server mappings it is all combine into
It is a very time-consuming task that row, which is enumerated, and is difficult to realize that energy consumption and user request the optimal of receptance simultaneously.Therefore it takes
Didactic method, the local scheduling being absorbed in initial schedule, by optimizing realization to overall scheduling to local scheduling
Optimization.Specific practice is as follows: carrying out random grouping two-by-two to the server in initial schedule, includes two clothes in each group
It is engaged in device, there are several to distribute its task on each server.As shown in Fig. 2, server 1 and n-1 constitute a server
Group, server k and m constitute a server group.Two-server and set of tasks therein constitute in each server group
One local scheduling.
Wherein, scheduling in (1) of step 2 is developed: when Servers-all has all been randomly formed two server groups, next
Local optimum can be carried out to the local scheduling in this server group using the thought to develop in evolution algorithmic.Specific practice
It is as shown in Figure 3: just carry out evolution operation after server group selected target task, in server group two servers by exchange or
The mode of migration task generates 4 filial generation scheduling, in order to guarantee that local scheduling will not degenerate after developing, using primitive scheduling as one
A filial generation saves.Then, one of server moves to its goal task on another server, generates second filial generation
Scheduling.Similar, goal task is moved to and generates third filial generation scheduling on first server by second server.Most
Afterwards, two servers are exchanged with each other goal task, generate the 4th filial generation scheduling.Even if as shown in figure 3, first filial generation scheduling
The copy of primitive scheduling, server S 1, which moves to task k, generates second filial generation scheduling, server S 2 in server S 2
Task m is moved to and generates another filial generation on S1, finally, S1 and S2 switching task k and m generate the 4th scheduling.Work as service
After device group generates 4 filial generation scheduling, each filial generation local scheduling is emulated, calculates energy consumption and task receptance, selection
Energy consumption is minimum, the highest filial generation scheduling replacement primitive scheduling of task receptance.
Wherein, single schedule optimizer optimization in (1) of step 2: since single schedule optimizer is in initial schedule
Servers-all is all grouped, then it can be by develop on every a pair of of the server of operation generation to each grouping
Local scheduling, all local scheduling merging just constitute the new scheduling that data center requests all users.In order to this
Scheduling advanced optimizes, and single schedule optimizer is allowed to be scheduling to initial schedule, repeated packets and the behaviour that develops again with newly generated
Make fixed number of times, it is final to generate the scheduling with high energy efficiency.Algorithm shown in Fig. 4 describes the mistake of single schedule optimizer in detail
Journey, wherein servers is data center's Servers-all, and tasks is all users request that data center receives,
Mapping is the initial schedule that initialization algorithm generates, i.e. to the mapping pair of server, niter is that dull degree is excellent for user's request
Change the number that device is iterated optimization.
Wherein, more schedule optimizer cooperative optimizations in (2) of step 2: since single schedule optimizer is directed to oneself
Scheduling is iterated optimization, it is easy to fall into local optimum, in order to avoid this kind of situation, while start multiple execution
Device, allow they it is independent initial schedule is optimized, due to being grouped the presence of randomness, multiple schedule optimizers cooperate
Local optimum can be avoided as far as possible.
The present invention can search optimal scheduling, pole under the constraint condition of cloud computation data center fixed server resource
The earth reduces energy consumption, improves user and requests receptance.
Claims (3)
1. a kind of towards energy-efficient heuristic cloud computing resources distribution and dispatching method, which is characterized in that this method includes following
Specific steps:
Step 1: initial phase
It the use of improved optimal adaptation method MBFD or for the first time adaptive method FF is all users that cloud computation data center receives
Task generates initial schedule;
Step 2: the optimizing phase
Instantiate several mutually independent schedule optimizers, each schedule optimizer initial schedule to generate in step 1 respectively
Local optimum is independently carried out for input, all independent schedule optimizer cooperations generate the near-optimization in global search space
Scheduling;It specifically includes:
(1) single schedule optimizer optimization: single schedule optimizer based on the initial schedule inputted, appoint by iterative utilization
Business selection, server grouping and scheduling three operations of evolution carry out local optimum to it;
(2) more schedule optimizer cooperative optimizations: after each schedule optimizer generates the scheduling of its local optimum, in all scheduling
Relatively and select energy consumption minimum and the highest scheduling of task receptance in the scheduling that optimizer generates, as initial schedule,
Carry out the operation of step (1);After each schedule optimizer generates the scheduling of its local optimum, generated in all schedule optimizers
It relatively and selects energy consumption minimum and the highest scheduling of task receptance in scheduling, as initial schedule, then carries out step (1)
Operation;In cycles, when operating procedure (1) number reaches 10-15 times, the near-optimization tune in global search space is obtained
Degree;
Wherein:
The task choosing operation is that each server selects a goal task using initial schedule: in each server
On, be Overlapped Execution time on each task computation it and the server between all other task and, there is minimum overlay
The executing time sum of the task is selected as the goal task of the server;
Data center server is carried out random grouping two-by-two, obtains two different server groups by the server division operation;
Using the initial schedule of input step 1, the local scheduling information in every server group on two servers constitutes the server group
Local scheduling;
The scheduling, which is developed, to be operated, and in every server group, execution evolution operation generates 4 different filial generations and dispatches;Every height
Generation scheduling generates as follows:
(1) it is dispatched the initial schedule of step 1 as first filial generation;
(2) goal task of first server is moved on second server, and modifies to correspond in initial schedule and is somebody's turn to do
The local scheduling information of operation obtains second filial generation scheduling;
(3) goal task of second server is moved on first server, and modifies to correspond in initial schedule and is somebody's turn to do
The local scheduling information of operation obtains third filial generation scheduling;
Goal task is exchanged between (4) two servers, and modifies the local scheduling information for corresponding to the operation in initial schedule
Obtain the 4th filial generation scheduling;
After generating 4 filial generation scheduling, single schedule optimizer emulates each filial generation scheduling, calculates consumption information and task connects
By rate, and select energy consumption minimum and the original initial scheduling of the highest filial generation scheduling substitution of task receptance.
2. the method as described in claim 1, which is characterized in that the generation initial schedule generates a physical server
Original allocation to user task maps.
3. the method as described in claim 1, which is characterized in that the list schedule optimizer, after receiving initial schedule, root
According to initial schedule, determine that data center is the server resource of each user task distribution, meanwhile, each server with distribute to
Its task constitutes the local scheduling on the server.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610966411.5A CN106550036B (en) | 2016-10-28 | 2016-10-28 | One kind is towards energy-efficient heuristic cloud computing resources distribution and dispatching method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610966411.5A CN106550036B (en) | 2016-10-28 | 2016-10-28 | One kind is towards energy-efficient heuristic cloud computing resources distribution and dispatching method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106550036A CN106550036A (en) | 2017-03-29 |
CN106550036B true CN106550036B (en) | 2019-05-17 |
Family
ID=58394511
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610966411.5A Active CN106550036B (en) | 2016-10-28 | 2016-10-28 | One kind is towards energy-efficient heuristic cloud computing resources distribution and dispatching method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106550036B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107436811A (en) * | 2017-07-07 | 2017-12-05 | 广东工业大学 | It is related to the task immigration method of task scheduling in mobile cloud problem |
CN110008015B (en) * | 2019-04-09 | 2022-09-30 | 中国科学技术大学 | Online task dispatching and scheduling method with bandwidth limitation in edge computing system |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103677957A (en) * | 2013-12-13 | 2014-03-26 | 重庆邮电大学 | Cloud-data-center high-energy-efficiency virtual machine placement method based on multiple resources |
CN103957261A (en) * | 2014-05-06 | 2014-07-30 | 湖南体运通信息技术有限公司 | Cloud computing resource distributing method based on energy consumption optimization |
WO2015099701A1 (en) * | 2013-12-24 | 2015-07-02 | Intel Corporation | Cloud compute scheduling using a heuristic contention model |
CN105159762A (en) * | 2015-08-03 | 2015-12-16 | 冷明 | Greedy strategy based heuristic cloud computing task scheduling method |
-
2016
- 2016-10-28 CN CN201610966411.5A patent/CN106550036B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103677957A (en) * | 2013-12-13 | 2014-03-26 | 重庆邮电大学 | Cloud-data-center high-energy-efficiency virtual machine placement method based on multiple resources |
WO2015099701A1 (en) * | 2013-12-24 | 2015-07-02 | Intel Corporation | Cloud compute scheduling using a heuristic contention model |
CN103957261A (en) * | 2014-05-06 | 2014-07-30 | 湖南体运通信息技术有限公司 | Cloud computing resource distributing method based on energy consumption optimization |
CN105159762A (en) * | 2015-08-03 | 2015-12-16 | 冷明 | Greedy strategy based heuristic cloud computing task scheduling method |
Non-Patent Citations (2)
Title |
---|
云数据中心高能效的虚拟机放置算法;豆育升,等;《小型微型计算机系统》;20141130;第35卷(第11期);第81~84页 |
基于启发式算法的数据迁移策略;孙阳,等;《吉林建筑大学学报》;20160630;第33卷(第3期);第2543~2547页 |
Also Published As
Publication number | Publication date |
---|---|
CN106550036A (en) | 2017-03-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Hussain et al. | Energy and performance-efficient task scheduling in heterogeneous virtualized cloud computing | |
Meshkati et al. | Energy-aware resource utilization based on particle swarm optimization and artificial bee colony algorithms in cloud computing | |
Luo et al. | Hybrid shuffled frog leaping algorithm for energy-efficient dynamic consolidation of virtual machines in cloud data centers | |
Guan et al. | Energy efficient virtual network embedding for green data centers using data center topology and future migration | |
Ghetas | A multi-objective Monarch Butterfly Algorithm for virtual machine placement in cloud computing | |
Gao et al. | An energy-aware ant colony algorithm for network-aware virtual machine placement in cloud computing | |
Xu et al. | Migration cost and energy-aware virtual machine consolidation under cloud environments considering remaining runtime | |
Chen et al. | EONS: minimizing energy consumption for executing real-time workflows in virtualized cloud data centers | |
CN104679594A (en) | Middleware distributed calculating method | |
Hanani et al. | A multi-parameter scheduling method of dynamic workloads for big data calculation in cloud computing | |
Rasouli et al. | EPBLA: energy-efficient consolidation of virtual machines using learning automata in cloud data centers | |
Peng et al. | A reinforcement learning-based mixed job scheduler scheme for cloud computing under SLA constraint | |
CN106550036B (en) | One kind is towards energy-efficient heuristic cloud computing resources distribution and dispatching method | |
Lin et al. | A heuristic task scheduling algorithm for heterogeneous virtual clusters | |
CN107908464A (en) | A kind of cloud computing workflow energy-saving scheduling method for considering reliability | |
Lv et al. | TBTOA: A dag-based task offloading scheme for mobile edge computing | |
Bui et al. | Energy efficiency in cloud computing based on mixture power spectral density prediction | |
Guo et al. | Hydrozoa: Dynamic hybrid-parallel dnn training on serverless containers | |
Hu et al. | Research of scheduling strategy on OpenStack | |
Jiang et al. | Network-aware virtual machine migration based on gene aggregation genetic algorithm | |
Rasouli et al. | Virtual machine placement in cloud systems using learning automata | |
Badr et al. | Task consolidation based power consumption minimization in cloud computing environment | |
Liang et al. | A high-applicability heterogeneous cloud data centers resource management algorithm based on trusted virtual machine migration | |
CN104683480A (en) | Distribution type calculation method based on applications | |
Su et al. | A power-aware virtual machine mapper using firefly optimization |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |