CN105302630A - Dynamic adjustment method and system for virtual machine - Google Patents
Dynamic adjustment method and system for virtual machine Download PDFInfo
- Publication number
- CN105302630A CN105302630A CN201510703546.8A CN201510703546A CN105302630A CN 105302630 A CN105302630 A CN 105302630A CN 201510703546 A CN201510703546 A CN 201510703546A CN 105302630 A CN105302630 A CN 105302630A
- Authority
- CN
- China
- Prior art keywords
- virtual machine
- server host
- server
- main frame
- constituent element
- 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
Links
Landscapes
- Debugging And Monitoring (AREA)
Abstract
The invention provides a dynamic adjustment method and system for a virtual machine. The method comprises: obtaining a server host in first and second running states, wherein in the first running state, an overload server host cannot meet an SLA default rate, and in the second running state, the load utilization rate of a no-load server host is lower than a preset threshold; and solving a global optimal solution of the virtual machine required to be migrated by using an extremal optimization based particle swarm algorithm. The virtual machine is adjusted according to an optimal solution result; and the steps are repeatedly performed every a predetermined cycle in a predetermined time window so as to finish dynamic adjustment of the virtual machine. The VMDA problem is efficiently solved by using the improved algorithm fusing extremal optimization with particle swarm.
Description
Technical field
The present invention relates to field of cloud computer technology, particularly relate to a kind of dynamic adjusting method and system thereof of virtual machine.
Background technology
Cloud data center is made up of the hardware facility of numerous high configuration usually.The computing power of data center has become the leading indicator that cloud service provider is concerned about, and along with the appearance of more and more large-scale data center, the energy resource consumption of data center is increasing.The high-performance hardware facility of One's name is legion consumes a large amount of energy, and the CO2 emission of generation causes greenhouse effect, produces great impact to climatic environment, and largely have impact on the operation benefits of cloud service provider.The low-yield utilization factor of data center mainly due to server poor efficiency or idle time, but still consume huge energy.Such as, have scholars to add up energy that the idle server of discovery one (do not have close or switching state) consumes often to account for when full load is run catabiotic about 70%.
In reality, the hardware facility in cloud data center not keeps static constant for a long time, and on the contrary, most of hardware state may often will be changed.First, new server may join in system, and then may needing of before existing carries out reconfiguring, repair or replacing; Secondly, resource pool often may change running status to adapt to the change in resources requirement of elastic cloud environment intermittent; 3rd, real-time migration (LiveMigration, LM) technology makes virtual machine (VirtualMachine, VM) can realize reconfiguring fast and integrating in different physical nodes, realizes the targets such as load balance with this; 4th, some servers to need virtual machine (vm) migration to other suitable server host, and the maintenance that needs to shut down after completing dynamic migration such as to restart at the operation, makes this part server unavailable within a certain period of time.Similar, certain server needs interim unlatching to happen suddenly to process some uncertain access peak.The access being more than server exists uncertain, in addition, also there is the possibility of various change at each server internal, such as change the processing unit number (ProcessingElements, PEs), memory size, hard-disc storage, bandwidth etc. of server.In addition, current server generally supports dynamic voltage frequency zoom technology (DynamicVoltageFrequencyScalingTechnique, DVFS), server dynamically can change voltage thus adjust operation frequency according to current load, reaches energy-conservation object.The dynamic of resource and uncertainty are necessary to adopt dynamic regulation mechanism, the hardware resource of dynamic monitoring system, detect and find that the state of resource changes (Energy-aware), to occurring that the VM of new application rationally places and to service level agreement (ServiceLevelAgreements, SLA) VM in the extremely low server of the VM broken a contract or cpu busy percentage is optimized configuration, makes the configuration of whole VM reach optimum as far as possible.Dynamic regulation mechanism must realize the efficient management of automatic dynamic as far as possible in without the interference situation of managerial personnel.
In cloud environment, for meeting uncertain resource bid peak, there is supply (over-provisioned) state in data center, a large amount of energy dissipations produces usually thus.
Current Intel Virtualization Technology in the data the heart start to be applied, system is supported in the virtual machine above it of real-time migration between physical node, thus realizes performance boost or the object such as energy-conservation.When the resource of the actual use of virtual machine is less than the resource distributing to it, virtual machine is by adjusting and merging, reconfigure to other server node, the server node that free time gets off is switched to energy saver mode by ACPI (AdvancedConfigurationandPowerInterface) interface specification, realizes the object of saving energy consumption.Current cloud data center resource scheduling strategy mainly concentrates on elevator system performance, safeguards SLA, and seldom from saving energy consumption angle.
DVFS realizes one of energy-conservation Main Means of hardware facility at present.DVFS be the application program run according to chip to the different needs of computing power, the running frequency of dynamic adjustments chip and voltage (for same chip, frequency is higher, and required voltage is also higher), thus reach energy-conservation object.Reduce frequency and can reduce power, but merely reducing frequency can not save energy.Because for a given task, only low-frequencyly reduce voltage falling simultaneously, low-energy consumption could be fallen veritably.Can DVFS implement to depend on that next success prediction processor needs number and the time of Processing tasks.And usually in real-time system, clock frequency and voltage are not linear relationships, there is very large uncertainty between task execution time, energy ezpenditure, processor voltage, inappropriate frequent Voltage Cortrol can make processor performance decline on the contrary.In most of cloud environment, task quantitative forecast is also difficult to determine.
DVFS needs to carry out power management by BIOS usually, and the design circuit of different manufacturers exists very big-difference.In order to there be a common power-management interface between operating system and hardware facility, that brings with the disunity interface that different vendor formulated on power management before improving is difficult to compatibling problem, and the companies such as Intel, Microsoft, Toshiba have formulated ACPI (AdvancedConfigurationandPowerInterface) specification jointly.ACPI improves original pattern (APM) of being carried out power management by BIOS, provides the interface specification of a more outstanding powder source management mode and configuration management.ACPI defines maximum six kinds of power supply statuss, the energy consumption power of processor, internal memory and hard disk that different states is corresponding different and running status.At present, most processor all supports several states such as operation, free time, dormancy, cut out.
Existing scholar is as follows about some progress of the problems referred to above:
The people such as Rusu propose the managing power consumption strategy ensured based on QoS for server cluster system.System is divided into management drawn game portion, rear end to manage two modules.Local management supports DVFS, and when back end manager detects that system needs close or open certain server, local management device controls power supply by DVFS module, and Server switching to corresponding states.This system does not comprise real-time migration of virtual machine technology, and whether server closing opens the calculated off-line depending on rear end, and saving energy consumption is limited.
Also have scholar in distributed cloud computing system, propose the strategy of a kind of available energy dissipation management, author is according to the relation of task processing time and energy ezpenditure, the objective function optimized is defined as relative superiority (RelativeSuperiority, RS) expression formula, for will distributing of task, first calculate this task matching RS value on each server, and this task is finally distributed to the maximum server of RS value.
But this algorithm acquiescence Servers-all is all in and activates and good running status, does not consider the isomerism of system, meanwhile, only considered the assignment problem newly increasing virtual machine in literary composition, and real process also needs to consider the problems such as SLA promise breaking virtual machine adjustment.
The scholars such as Kusic are sorting consistence problem managing power consumption problem definition under virtual isomerous environment and adopt limited controls in advance filtering (LimitedLookaheadControl, LLC) to process.Processing target is the profit realizing maximum resource ISP, meets energy ezpenditure and SLA simultaneously and to break a contract minimized requirement.System adopts Kalman filter to estimate the quantity of future customers request and the state of prognoses system, realizes necessary resource consolidation accordingly.But this system is difficult to realize in IaaS cloud environment, and model too complex, the system for 15 nodes regulates the needs time of 30 minutes at every turn, is difficult to be applied to extensive real-time cloud data center.
The people such as Verma carry out modeling to the virtual machine Energy-aware Dynamic Arrangement problem under virtual isomerous environment, it is become continuous optimization problems: at each time frame, virtual machine Placement Problems can be considered the minimum and optimization problem of maximizing performance of energy ezpenditure, author adopts heuritic approach to solve this bin packing, and redistributes with the VM that real-time migration strategy completes each time frame.In their follow-up work, they also adopt static policies (moon, year adjustment), semi-static strategy (day, week adjustment) and dynamic strategy (point, hour adjust) to carry out periodic adjustment.But these algorithms do not consider SLA promise breaking problem: system performance can decline along with the change of load, and SLA can not be guaranteed.
The people such as Berral have studied VM Dynamic Integration problem, are meeting under SLA prerequisite, and they adopt machine learning techniques to process energy ezpenditure control problem.This processing mode only considered the application of some specific occasions, the application scenario of the such as limited constraint of the band such as high-performance calculation (HighPerformanceComputing, HPC), inapplicable for common mixed load application scenario.
The people such as Beloglazov adopt optimal adaptation degree sort descending algorithm (MBFD) to solve the dynamic assignment problem of VM, the VM Dynamic Integration that maximization is energy-conservation but this algorithm can not be realized ideal.
From the above, mainly there are the following problems for the resource management techniques scheme of saving based on energy consumption at present: 1. energy consumption technology of saving only is considered to save energy mostly, do not consider the promise breaking problem of SLA very well, or both can not be taken into account very well; 2. mainly concentrate on the assignment problem that newly increases virtual machine, and consider seldom for the operation conditions of the virtual machine distributed and the server load condition of correspondence; 3. as the VMDA problem of resource management core support technology, existing scheme adopts traditional heuritic approach to solve mostly, and solution efficiency is low.When server host scale thousands of in the face of data center, operation efficiency is not enough, is difficult to the scheduling of resource target reached rapidly and efficiently.
Existing desirable scheduling (adjustment) target is: under the prerequisite safeguarding SLA, guarantee system performance, realize energy-conservation object.
The system of the scheduling that can realize ideal generally can face following Railway Project: the energy consumption adjustment that 1. each server is too much can reduce the stability that server runs; 2. in dynamic environment, closing server faces the risk that QoS can not be guaranteed.Merge owing to there is virtual machine and integrate, the service that some virtual machine runs can not obtain enough resources in access peak period, thus can not meet the requirement of QoS; 3. in the virtualized environment expanded in server host quantity, common dispatching algorithm is difficult to guarantee that resource management is fast and effectively dispatched, and guarantees that the application performance management design of system SLA exists very large challenge.Realize some requirement above, core needs a set of efficient Resource dynamic allocation algorithm, i.e. virtual machine dynamic assignment (VirtualMachineDynamicAllocation, VMDA) algorithm fast.
This algorithm also requires that external service performance is excellent, can ensure the QoS that user specifies while requiring that system meets energy-saving and emission-reduction, the green calculating of realization.
Easy proof, VMDA problem belongs to NP (NP-complete) problem completely, and the polynomial time algorithm that not yet existence one is feasible at present solves np complete problem.All there is the phase transition phenomena (phasetransition) of impact decision problem computational complexity in np complete problem: all np complete problems all at least exist a control variable, there is a transformation temperature (or claiming critical point) in this variable, this point is two regions optimizing spatial division, at area limit place annex, the probability separating existence can be met and there occurs unexpected transformation in certain critical point of control variable.This phenomenon is called the phase transition phenomena of np complete problem, maximum at the problem computation complexity at transformation temperature place.There is multiple such transformation temperature in VMDA, and optimum solution is often located near transformation temperature, and therefore, the optimizing algorithm solving this kind of problem must have the good ability solving Phase-change Problems.
Summary of the invention
In view of above-mentioned the deficiencies in the prior art part, the object of the present invention is to provide a kind of dynamic adjusting method and system thereof of virtual machine, to be intended to solve in prior art solution efficiency in virtual machine adjustment searching process low, the problem of virtual machine dynamic assignment rapidly and efficiently cannot be realized on a large amount of server host.
In order to achieve the above object, this invention takes following technical scheme:
A dynamic adjusting method for virtual machine, wherein, described method comprises:
Obtain the server host being in first and second ruuning situation;
Described first ruuning situation is: the Overloaded Servers main frame that can not meet SLA rate of violation;
Described second ruuning situation is: load utilization is lower than the unloaded server host of predetermined threshold value;
Move some virtual machines on described Overloaded Servers main frame in other server hosts;
By all virtual machine (vm) migrations on described unloaded server host in other servers, and described unloaded server host is switched to energy saver mode;
The globally optimal solution of the virtual machine using the PSO Algorithm based on extremal optimization to move; And
According to described optimum solution result, adjustment virtual machine; Further, in predetermined time window, repeat above-mentioned steps every the predetermined cycle thus complete the dynamic conditioning of virtual machine.
The dynamic adjusting method of described virtual machine, wherein, the described particle cluster algorithm based on extremal optimization is specially: in described particle cluster algorithm iteration renewal process, incorporate extremal optimization local search algorithm with probabilistic manner.
The dynamic adjusting method of described virtual machine, wherein, in described extremal optimization algorithm, each constituent element is corresponding with the virtual machine that need move; For each constituent element gives corresponding fitness, and the constituent element selecting fitness minimum and neighbours thereof make a variation.
The dynamic adjusting method of described virtual machine, wherein, the fitness of described constituent element calculates especially by following formula:
The dynamic adjusting method of described virtual machine, wherein, described method also comprises: traversal Servers-all main frame, is assigned to the virtual machine newly increased and can meets SLA rate of violation and save in the server host of energy consumption most.
A dynamic debugging system for virtual machine, wherein, described system comprises:
Server host ruuning situation acquisition module, for obtaining the server host being in first and second ruuning situation; Described first ruuning situation is: the Overloaded Servers main frame that can not meet SLA rate of violation; Described second ruuning situation is: load utilization is lower than the unloaded server host of predetermined threshold value;
Virtual machine (vm) migration module, for moving some virtual machines on described Overloaded Servers main frame in other servers to other server hosts and by all virtual machine (vm) migrations on described unloaded server host, and described unloaded server host is switched to energy saver mode;
Migration computing module, the globally optimal solution of the virtual machine that the PSO Algorithm for using based on extremal optimization need move; And
Virtual machine (vm) migration module also for, according to described optimum solution result, adjustment virtual machine;
Adjustment cycle module, in predetermined time window, repeats above-mentioned steps every the predetermined cycle thus completes the dynamic conditioning of virtual machine.
The dynamic debugging system of described virtual machine, wherein, described migration computing module specifically for: in described particle cluster algorithm iteration renewal process, incorporate extremal optimization local search algorithm with probabilistic manner.
The dynamic debugging system of described virtual machine, wherein, described migration computing module specifically for: in described extremal optimization algorithm, each constituent element is corresponding with the virtual machine that need move; For each constituent element gives corresponding fitness, and the constituent element selecting fitness minimum and neighbours thereof make a variation.
The dynamic debugging system of described virtual machine, wherein, described migration computing module is specifically for the fitness being calculated described constituent element by following formula:
λ
ifor the destination server main frame that the fitness of i constituent element, h (i) distribute for virtual machine i; The processor utilization that Utlz (h (i)) is destination server main frame h (i); Utlz (i) is the processor utilization of virtual machine i; α and β is weight parameter; C1 and c2 is weight primary constants.
The dynamic debugging system of described virtual machine, wherein, described system also comprises: assignment module, for traveling through Servers-all main frame, user newly being applied for the virtual machine increased is assigned to and can meet SLA rate of violation and save in the server host of energy consumption most.
Beneficial effect: the dynamic adjusting method of a kind of virtual machine provided by the invention and system thereof, employ a kind of innovatory algorithm having merged extremal optimization and population and solve VMDA problem efficiently, thus achieve the dynamic conditioning of virtual machine in a large amount of server host, maximise power-saving object can be realized while the customer service quality level of effective guarantee data center.
Accompanying drawing explanation
Fig. 1 is the method flow diagram of the dynamic adjusting method of the virtual machine of the specific embodiment of the invention.
Fig. 2 is the Scheduling Framework schematic diagram of the dynamic conditioning of the virtual machine of the specific embodiment of the invention.
Fig. 3 is the structured flowchart of the dynamic debugging system of the virtual machine of the specific embodiment of the invention.
Fig. 4 is that the dynamic adjusting method of the described virtual machine of application of the specific embodiment of the invention and the energy consumption consumption of other algorithms contrast schematic diagram.
Embodiment
The invention provides a kind of dynamic adjusting method and system thereof of virtual machine.For making object of the present invention, technical scheme and effect clearly, clearly, developing simultaneously referring to accompanying drawing, the present invention is described in more detail for embodiment.Should be appreciated that specific embodiment described herein only in order to explain the present invention, be not intended to limit the present invention.
As shown in Figure 1, be the dynamic adjusting method of the virtual machine of the specific embodiment of the invention.Described method comprises:
S1, obtain and be in the server host of first and second ruuning situation.
Wherein, described first ruuning situation is: the Overloaded Servers main frame that can not meet SLA rate of violation; Described second ruuning situation is: load utilization is lower than the unloaded server host of predetermined threshold value.For Overloaded Servers main frame, need to move some virtual machines on described Overloaded Servers main frame in other server hosts.
And for unloaded server host, then need all virtual machine (vm) migrations on described unloaded server host in other servers, and described unloaded server host is switched to energy saver mode (specifically as stated in the Background Art, such as reducing server host operation frequency etc.).
In the application of the cloud data center of reality, for Servers-all main frame, the server host that adjustable strategies only selects SLA (not meeting rate of violation to require) server host not up to standard and load too low performs VM integration, to reduce the quantity of VM real-time migration.
The server host of SLA promise breaking is selected in strategy, first matches to make the SLA of server service level and customization by formula (1) dynamic conditioning processor utilization upper limit threshold.
Wherein, upper_th is the processor utilization upper threshold of current server when meeting SLA; R
sLAfor server host target SLAs; r
sLAfor the SLAs of the current reality of server host, τ are an increment constant.
In the present embodiment example, estimate that new processor threshold value is similar to direct ratio with present threshold value and the product between target SLA and actual SLA ratio, and new threshold value is substituted old threshold value.It should be noted that upper_th value infinitely can not reduce or increase (being set in the present embodiment: upper_th ∈ [0.5,0.95]).
Then assess the SLA rate of violation of current server host and judge whether to meet appointment SLA requirement, if do not met, selecting part VM by formula (2) and add migration list and prepare them to move out of main frame.
Wherein, hvs is that the virtual machine that current server main frame is born is all; Utlz (h) is current hosts processor overall utilization level, and Utlz (v) takies the average utilization of current processor for virtual machine v within the scope of time window.
The present embodiment adopts the strategy minimizing virtual machine (vm) migration quantity to select the virtual machine needing in Overloaded Servers to move away, and the virtual machine quantity that needs are moved is minimum, the problem bringing system performance to decline when can reduce virtual machine (vm) migration thus.
After virtual machine (vm) migration is gone out, front server host carry residue virtual machine processor utilization sum be less than the threshold value of regulation, thus reach the requirement meeting regulation SLA.
For in the server host of first and second ruuning situation, need the concrete migration position carrying out the virtual machine moved, can be solved by step S2 computing and draw.
In actual application, meet the important requirement that QoS is cloud computing system.QoS is presented by SLA usually.SLA typically specify the indexs such as system computing capacity, reaction time, the maximum visit capacity of system, and these indexs corresponding to different application services have very large difference.
Concrete, formula (4) can be passed through and define certain server host SLA rate of violation (SLAViolation, SLAV) in certain hour window:
The SLA violations of last item caused by server host processor overload operation in formula (4), wherein, T
vfor in time window, current server main frame experiences the duration of 100% utilization factor, T
afor time window size, when main frame arrives 100% utilization factor interval scale now main frame overload operation, α is normalized parameter.Due to usually when server host reaches oepration at full load, be difficult to the access application that response is new in time, cause SLA to break a contract and occur.
The SLA promise breaking that in formula (4), a rear hydraulic performance decline caused for real-time migration causes, because in real-time migration process, needs the external access performance of the service of the upper operation of the VM of migration to decline to some extent, wherein C
vfor moving spent processor resource (MIPS), usual C
vvalue account for about 10%, C of real-time migration process controller total resources
afor total processor resource (MIPS) of virtual machine application.
That is as shown in Figure 2, the Scheduling Framework applying the dynamic adjusting method of virtual machine of the present invention is specific as follows:
User to cloud system application VM resource, and specifies SLA.Scheduling Framework, according to application requirement and server host operation conditions, is arranged to the VM of application on suitable server host.(it should be noted that the instrumentation of this VM completes within the very fast time in order to ensure response speed)
Scheduling Framework real-time monitoring hardware facility (a large amount of server host of such as data center) operation conditions, carries out dynamic VM integration according to the ruuning situation of main frame in certain hour window.Described time window specifically can determine according to actual service condition.
Server for following two kinds of situations carries out Dynamic Integration: (1) is for the server host that can not meet SLA, reduce server host processor utilization upper threshold level, some (individual) VM selected on this server host moves to other server, reduce its load, guarantee that SLA is ensured; (2) for the server host of utilization factor load lower than certain lower limit, the whole VM on this server are moved to other server, and this Server switching to energy-conservation (dormancy) state, realize energy-conservation object.
S2, the globally optimal solution of virtual machine using the PSO Algorithm based on extremal optimization to move.
In specific embodiment of the invention, the described particle cluster algorithm based on extremal optimization is specially: in described particle cluster algorithm iteration renewal process, incorporate extremal optimization local search algorithm with probabilistic manner.
Existing basic particle group algorithm ability of searching optimum still can, but its algorithm convergence efficiency is not high.For improving the efficiency of evolution (being convenient to apply on a large amount of server host) of particle cluster algorithm, keeping sample diversity, better local search technique can be incorporated to improve algorithm evolution efficiency under the multifarious prerequisite of the certain sample space of guarantee.
More specifically, in described extremal optimization algorithm, each constituent element is corresponding with the virtual machine that need move; For each constituent element gives corresponding fitness, and the constituent element selecting fitness minimum and neighbours thereof make a variation.
Particle cluster algorithm (PSO) is a kind of based on the group hunting evolution algorithm of mould because evolving.When it is according to flock of birds airflight, every bird constantly can reach the principle of optimum position and a kind of bionic Algorithm of putting forward to self-position adjustment according to the position of around bird.Described particle cluster algorithm comprises the following key character:
(1), in PSO, each particle (bird) will carry out mould because evolving when they fly.Each particle carrys out its state of flight of dynamic conditioning according to the flight path of self flight path and companion.
(2) PSO constantly upgrades for random several particles produced, in iteration renewal process, always by optimum solution p that particle itself finds
idwith the optimum solution p that whole population is found by the end of current iteration
gdthese two " extreme values " upgrade each particle, and in some occasion, also can be used as whole population with part neighbours, the extreme value in neighbours region is local extremum.
(3), in current kth time iteration, a certain particle i, once find this two optimal values, upgrades the speed of oneself and new position according to following formula:
In formula (5) and (6),
speed during iteration secondary to particle i kth; W is inertia weight;
it is the position of current particle; p
idfor the optimum solution found by particle itself, p
gdfor the optimum solution found by whole population, r
1and r
2it is the random number between 0 and 1; c
1and c
2it is Studying factors.
Wherein, w is great for the convergence effect of particle cluster algorithm.W value is larger, then global optimizing ability is stronger, and local optimal searching ability is more weak.Otherwise then local optimal searching ability strengthens, and global optimizing ability weakens.
Can by speed before adjusting the size of w and controlling on the impact of present speed, become of taking into account global search and Local Search and compromise.Therefore, in practical application, algorithm starting stage w is comparatively large, along with the w that carries out of iteration diminishes gradually.Because w is large, then speed v is just large, is conducive to the space that particle search is larger, may finds new solution territory; And w is less, then speed w is just little, is conducive to excavating in current solution space better separating
Shown in the false code of the above-mentioned particle cluster algorithm based on extremal optimization is specific as follows:
Described extremal optimization algorithm only has an individuality be made up of some constituent elements, each component of each constituent element homographic solution vector when evolving.This algorithm thinks that the contribution of each inner constituent element to the good and bad degree of individuality is different, and comes for each constituent element gives fitness according to the contribution of each constituent element to individual goal functional value, and the minimum constituent element of fitness is the poorest constituent element.The each iteration of extremal optimization always selects the poorest constituent element and neighbours thereof to make a variation.
The basic framework of described extremal optimization algorithm is as described below:
First, random generation one individual X=(x
1, x
2..., x
t), if optimum solution is X
best=X, target function value is denoted as C (X);
Then, to individual X, perform following operation:
A. constituent element x is evaluated
ifitness and be denoted as λ
i, i ∈ 1,2 ..., t};
B. according to fitness size, t constituent element is sorted, find out the constituent element x that fitness is the poorest
j, i.e. λ
j≤ λ
i, i=1,2 ..., t, then x
jbe the poorest constituent element;
C. in the neighborhood of X, select a neighbours X', make the poorest constituent element x by variation
jchange;
D. individual X=X' is upgraded;
If e. C (X) < C (X
best), then X
best=X;
Then. repeat the step finding out the poorest constituent element of fitness, until end condition meets;
Finally. return gained optimum solution X
best.
In the specific embodiment of the dynamic adjusting method of virtual machine of the present invention, the fitness of described constituent element calculates especially by following formula:
λ
ifor the destination server main frame that the fitness of i constituent element, h (i) distribute for virtual machine i; The processor utilization that Utlz (h (i)) is destination server main frame h (i); Utlz (i) is the processor utilization of virtual machine i; α and β is weight parameter; C1 and c2 is weight primary constants.
It is preferred that for maintenance algorithm global search, avoid being absorbed in local optimum, select need variation constituent element time, select according to the probability distribution of fitness order by power-law.If total t needs the VM that moves, each VM (constituent element) to choose probability to obey distribution as shown in formula (8) from big to small by constituent element fitness value:
p(r)∝r
-τ1≤r≤t,τ≥0(8)
S3, according to described optimum solution result, adjustment virtual machine.
S4, in predetermined time window, repeat above-mentioned steps every the predetermined cycle thus complete the dynamic conditioning of virtual machine.
The described predetermined cycle can be determined according to actual conditions.In specific implementation process, each server host software and hardware due to cloud data center is in the environment of constantly change, the service that the VM of each server carrying runs also is in the state of constantly change, Scheduling Framework real-time perception main frame operation conditions, and carry out dynamic VM integration with the certain hour window cycle according to the operation conditions of main frame.
Concrete, as shown in Figure 1, user is newly applied for the virtual machine increased, described method also comprises:
S5, traversal Servers-all main frame, be assigned to the virtual machine newly increased and can meet SLA rate of violation and save in the server host of energy consumption most.
Corresponding, in above-mentioned Scheduling Framework, whenever data center all allows user to propose new service request (VM), and after reaching SLA, cloud data center responds fast it and VM is assigned to suitable server host (Host).The scheduling strategy of the virtual machine newly applied for for user is: meet SLA by newly applying for that the VM increased is assigned to and can save on the server host of energy consumption.It need travel through a server host list and can obtain a result, and has the advantage of response fast.
In the concrete calculating of energy consumption, the server host energy ezpenditure of existing cloud data center formed primarily of the consumption of the hardware modules such as CPU, internal memory, disk and refrigeration system.Along with multi-core CPU is more and more universal, the energy ezpenditure of CPU account for major part, existing research display: server energy ezpenditure can approximately linear direct ratio with the energy ezpenditure of CPU, and the energy of an idle server consumption to account for when full load is run catabiotic 70%.
In addition, we also need energy ezpenditure when considering virtual machine (vm) migration, and real time virtual machine migrating technology allows to reset virtual machine between server host fast and flexible, and do not need service hang-up to complete.In real-time migration technology, the image file of virtual machine and data real time backup, at network attached storage (NetworkAttachedStorage, NAS), therefore, do not need during migration to copy virtual machine itself, only need to copy virutal machine memory.The virtual machine (vm) migration time is that memory size is divided by the network bandwidth.But, real time virtual machine transition process still can bring adverse influence to the service that virtual machine runs, existing research has been discussed in detail real-time migration technology to the impact of system performance and has carried out mathematical modeling, the quantity that when pointing out that performance impairment depends primarily on time and the migration of migration, virtual in-fight service memory pages upgrades.For common application service, such as web page server, real-time migration process approximately expends 10%CPU utilization factor.
Thus, define unit interval self-energy consumed power to be specifically defined as by following formula:
In formula (3), E
maxh () is energy ezpenditure power when server host h full load is run, Utlz (h) is unit time server host-processor average utilization, v is the virtual machine set of migration in unit time window, and T (i) is the transit time of virtual machine i.
As shown in Figure 3, be the dynamic debugging system of a kind of virtual machine of the specific embodiment of the invention, wherein, described system comprises:
Server host ruuning situation acquisition module 100, for obtaining the server host being in first and second ruuning situation; Described first ruuning situation is: the Overloaded Servers main frame that can not meet SLA rate of violation; Described second ruuning situation is: load utilization is lower than the unloaded server host of predetermined threshold value.
Virtual machine (vm) migration module 200, for moving some virtual machines on described Overloaded Servers main frame in other servers to other server hosts and by all virtual machine (vm) migrations on described unloaded server host, and described unloaded server host is switched to energy saver mode.
Migration computing module 300, the globally optimal solution of the virtual machine that the PSO Algorithm for using based on extremal optimization need move; And
Virtual machine (vm) migration module 300 also for, according to described optimum solution result, adjustment virtual machine.
Adjustment cycle module 400, in predetermined time window, repeats above-mentioned steps every the predetermined cycle thus completes the dynamic conditioning of virtual machine.
Concrete, for the virtual machine that user newly applies for, described system also comprises: assignment module 500, for traveling through Servers-all main frame, user newly being applied for the virtual machine increased is assigned to and can meet SLA rate of violation and save in the server host of energy consumption most.
In a particular embodiment of the present invention, described migration computing module 300 specifically for: in described particle cluster algorithm iteration renewal process, incorporate extremal optimization local search algorithm with probabilistic manner.
More specifically, described migration computing module 300 specifically for: in described extremal optimization algorithm, each constituent element is corresponding with the virtual machine that need move; For each constituent element gives corresponding fitness, and the constituent element selecting fitness minimum and neighbours thereof make a variation.
Described migration computing module is specifically for the fitness being calculated described constituent element by following formula:
Wherein, λ
ifor the destination server main frame that the fitness of i constituent element, h (i) distribute for virtual machine i; The processor utilization that Utlz (h (i)) is destination server main frame h (i); Utlz (i) is the processor utilization of virtual machine i; α and β is weight parameter; C1 and c2 is weight primary constants.As detailed above.
Embodiment 1:
Emulation experiment:
Cloud computing emulation platform CloudSimtoolkit (being derived from document HPCS2009, ISBN:978-1-4244-49071) is adopted to emulate.
CloudSim is the cloud computing emulation platform that the grid experiment room of Univ Melbourne Australia and Gridbus project were released in 2009, it inherits the programming model of GridSim, support the research and development of cloud computing, provide the characteristic of cloud computing, support resource management and the dispatching simulation of cloud computing.The CIS (CloudInformationService) of CloudSim and DataCenterBroker realizes resource discovering and information interaction, is the core of operation simulation.The independently developed dispatching algorithm of user can realize in the method for DataCenterBroker, thus realizes the simulation of dispatching algorithm.The simulated experiment of current latest edition back-level server main frame Energy-aware, this programme has adopted CloudSim platform to test algorithm.
The emulation cloud platform configuration of test is 1000 station server main frames, comprise 500 ProLiantDL360G4p main frames (being configured to 3400MHz*2core, 6GB internal memory, the 1GB network bandwidth), 500 ProLiantML110G3 main frames (being configured to 3000MHz*2core, 4GB internal memory, the 1GB network bandwidth).The energy ezpenditure of every station server calculates according to formula (3).
According to SPECpower2010 benchmark test fourth quarter statistical average result, by E
maxvalue be set to 259Wh.
The virtual machine task load of this emulation experiment is data from CoMon project.This project belongs to a part for PlanetLab monitoring facilities.Data mainly comprise 500 the different local upper cpu busy percentages more than 1000 virtual machines that spread all over the world, and these data detected every 5 minutes and obtain.
Select on March 3rd, 2011 measured data as final test figure.Adjustment cycle is 5 minutes.During test, new VM application is used as in each VM load on test set, needs heart Resources allocation operation service in the data.Scheduling Framework of the present invention is adopted to carry out intelligent scheduling to all VM loads.The cycle VM dynamic conditioning performed at regular intervals, during each scheduling, continuous G=250 the iteration of optimum solution is not improved, and think and reach the condition of convergence, algorithm exits.Often kind of scheduling scheme reruns and gets the performance of its mean value as evaluation algorithms for 10 times.
Mainly compare from data center SLA rate of violation (SLAV) and energy ezpenditure (Energy) angle.Experimental result is as shown in Table 1:
Several scheduling strategy results contrast of table 1
Wherein, first is classified as target SLAV, and experiment tests the performance of SLAV value from 1% to 5% (× E-3) all kinds of algorithm respectively.Secondary series LTH is the minimum utilization factor of server host, and often group SLAV value all comprises the respective hosts utilization factor from 0.1 to 0.5.
For in the algorithm that compares, the first NPA is non-energy perception scheduling strategy, allows Servers-all main frame run under ceiling capacity consumption patterns under this strategy; The second is DVFS pattern, does not carry out any replacement integrate process under this pattern to VM; The third strategy is the corresponding adjustable strategies of virtual machine dynamic adjusting method of the present invention.
Often kind of algorithm is tested under various different SLAV and LTH combines, and draws the value of energy ezpenditure (kWh) and virtual machine (vm) migration number (Migr.).
As shown in Table 1, Energy-aware strategy is adopted significantly can to reduce the energy ezpenditure of cloud data center.When adopting VM allocation strategy of the present invention, energy ezpenditure only accounts for about 3% of NPA strategy, saves the consumption of energy greatly.
Meanwhile, in this experiment test, as the minimum 86kWh of energy consumption, still can keep the SLA rate of violation of 5% × E-3, namely system still can keep extraordinary service quality while conserve energy.Therefore, the particle cluster algorithm based on extremal optimization of the present invention, utilizes the good local searching strategy of extremal optimization technology to can further improve the dispatching efficiency of population, realizes better energy-conservation object when virtual machine (vm) migration number is more or less the same.Such as when SLAV=5% × E-3, LTH=0.3, population extremal optimization scheme only expends the lowest power consumption of 86kWh.
On the other hand, as shown in Table 1, LTH usually can not be too little, though too little can the avoiding of LTH cuts dormant state server host too much, also there is the problem that energy consumption is too high; Meanwhile, LTH value can not be too large, and LTH too conference causes server host frequently to switch between dormancy and wake-up states, causes extra energy ezpenditure, if LTH is too large until approaching UpperThreshold, even can cause the problem that virtual machine cannot distribute.The comparatively ideal value of usual LTH should be set between 0.2 ~ 0.4.
Document " BeloglazovA., BuyyaR..OptimalOnlineDeterministicAlgorithmsandAdaptiveH euristicsforEnergyandPerformanceEfficientDynamicConsolid ationofVirtualMachinesinCloudDataCenters [J] .ConcurrencyandComputation:PracticeandExperience, 2011, 0:1-24 " describe a kind of Energy-aware algorithm to process the scheduling problem of cloud data center, this algorithm adopts the strategies such as minimum transition time (MMT) to select to need the virtual machine of migration, and adopt MBFD algorithm to process the Placement Problems of virtual machine.
This algorithm is adopting under optimum MMT strategy, and under CloudSim platform, carry out analogue simulation to certain incoming task, obtaining least energy consumption is 87.67kWh, and the SLAV drawn is 4.65% × E-3.
The contrast of above-mentioned algorithm and algorithm of the present invention:
Under same test condition (test platform is identical with task initial conditions), algorithm target SLAV is set to 4.65% × E-3 and tests.As shown in Figure 4, under identical SLAV level, adopt algorithm of the present invention only to need the energy ezpenditure of 82.35, the object of better saving energy consumption can be reached.
Be understandable that, for those of ordinary skills, can be equal to according to technical scheme of the present invention and the present invention's design and replace or change, and all these change or replace the protection domain that all should belong to the claim appended by the present invention.
Claims (10)
1. a dynamic adjusting method for virtual machine, is characterized in that, described method comprises:
Obtain the server host being in first and second ruuning situation;
Described first ruuning situation is: the Overloaded Servers main frame that can not meet SLA rate of violation;
Described second ruuning situation is: load utilization is lower than the unloaded server host of predetermined threshold value;
Move some virtual machines on described Overloaded Servers main frame in other server hosts;
By all virtual machine (vm) migrations on described unloaded server host in other servers, and described unloaded server host is switched to energy saver mode;
The globally optimal solution of the virtual machine using the PSO Algorithm based on extremal optimization to move; And
According to described optimum solution result, adjustment virtual machine;
In predetermined time window, repeat above-mentioned steps every the predetermined cycle thus complete the dynamic conditioning of virtual machine.
2. the dynamic adjusting method of virtual machine according to claim 1, is characterized in that, the described particle cluster algorithm based on extremal optimization is specially:
In described particle cluster algorithm iteration renewal process, incorporate extremal optimization local search algorithm with probabilistic manner.
3. the dynamic adjusting method of virtual machine according to claim 2, is characterized in that, in described extremal optimization algorithm, each constituent element is corresponding with the virtual machine that need move;
For each constituent element gives corresponding fitness, and
The minimum constituent element of fitness and neighbours thereof are selected to make a variation.
4. the dynamic adjusting method of virtual machine according to claim 3, is characterized in that, the fitness of described constituent element calculates especially by following formula:
λ
ifor the destination server main frame that the fitness of i constituent element, h (i) distribute for virtual machine i; The processor utilization that Utlz (h (i)) is destination server main frame h (i); Utlz (i) is the processor utilization of virtual machine i; α and β is weight parameter; C1 and c2 is weight primary constants.
5. the dynamic adjusting method of virtual machine according to claim 1, is characterized in that, described method also comprises:
Traversal Servers-all main frame, is assigned to the virtual machine newly increased and can meets SLA rate of violation and save in the server host of energy consumption most.
6. a dynamic debugging system for virtual machine, is characterized in that, described system comprises:
Server host ruuning situation acquisition module, for obtaining the server host being in first and second ruuning situation;
Described first ruuning situation is: the Overloaded Servers main frame that can not meet SLA rate of violation;
Described second ruuning situation is: load utilization is lower than the unloaded server host of predetermined threshold value; Virtual machine (vm) migration module, for moving some virtual machines on described Overloaded Servers main frame in other servers to other server hosts and by all virtual machine (vm) migrations on described unloaded server host, and described unloaded server host is switched to energy saver mode;
Migration computing module, the globally optimal solution of the virtual machine that the PSO Algorithm for using based on extremal optimization need move; And
Virtual machine (vm) migration module also for, according to described optimum solution result, adjustment virtual machine;
Adjustment cycle module, in predetermined time window, repeats above-mentioned steps every the predetermined cycle thus completes the dynamic conditioning of virtual machine.
7. the dynamic debugging system of virtual machine according to claim 6, is characterized in that, described migration computing module specifically for:
In described particle cluster algorithm iteration renewal process, incorporate extremal optimization local search algorithm with probabilistic manner.
8. the dynamic debugging system of virtual machine according to claim 7, is characterized in that, described migration computing module specifically for:
In described extremal optimization algorithm, each constituent element is corresponding with the virtual machine that need move;
For each constituent element gives corresponding fitness, and
The minimum constituent element of fitness and neighbours thereof are selected to make a variation.
9. the dynamic debugging system of virtual machine according to claim 8, is characterized in that, described migration computing module is specifically for the fitness being calculated described constituent element by following formula:
λ
ifor the destination server main frame that the fitness of i constituent element, h (i) distribute for virtual machine i; The processor utilization that Utlz (h (i)) is destination server main frame h (i); Utlz (i) is the processor utilization of virtual machine i; α and β is weight parameter; C1 and c2 is weight primary constants.
10. the dynamic debugging system of virtual machine according to claim 6, it is characterized in that, described system also comprises: assignment module, for traveling through Servers-all main frame, user newly being applied for the virtual machine increased is assigned to and can meet SLA rate of violation and save in the server host of energy consumption most.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510703546.8A CN105302630B (en) | 2015-10-26 | 2015-10-26 | A kind of dynamic adjusting method and its system of virtual machine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510703546.8A CN105302630B (en) | 2015-10-26 | 2015-10-26 | A kind of dynamic adjusting method and its system of virtual machine |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105302630A true CN105302630A (en) | 2016-02-03 |
CN105302630B CN105302630B (en) | 2018-07-13 |
Family
ID=55199924
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510703546.8A Active CN105302630B (en) | 2015-10-26 | 2015-10-26 | A kind of dynamic adjusting method and its system of virtual machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105302630B (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106936905A (en) * | 2017-03-07 | 2017-07-07 | 中国联合网络通信集团有限公司 | The dispatching method and its scheduling system of the Nova component virtual machines based on openstack |
CN107861802A (en) * | 2017-10-24 | 2018-03-30 | 郑州云海信息技术有限公司 | Method for managing resource and device in cloud data system |
CN108306780A (en) * | 2017-09-07 | 2018-07-20 | 上海金融期货信息技术有限公司 | A kind of system and method for the virtual machine communication quality self-optimizing based on cloud environment |
CN108469983A (en) * | 2018-04-02 | 2018-08-31 | 西南交通大学 | A kind of virtual machine deployment method based on particle cluster algorithm under cloud environment |
CN109491760A (en) * | 2018-10-29 | 2019-03-19 | 中国科学院重庆绿色智能技术研究院 | A kind of high-effect data center's Cloud Server resource autonomous management method and system |
CN109766190A (en) * | 2019-01-15 | 2019-05-17 | 无锡华云数据技术服务有限公司 | Cloud resource dispatching method, device, equipment and storage medium |
CN109901932A (en) * | 2019-03-12 | 2019-06-18 | 东北大学 | A kind of Server Consolidation method based on virtual machine |
CN109960568A (en) * | 2019-02-18 | 2019-07-02 | 深圳大学 | A kind of dispatching method and electronic equipment |
CN110362388A (en) * | 2018-04-11 | 2019-10-22 | 中移(苏州)软件技术有限公司 | A kind of resource regulating method and device |
CN112395161A (en) * | 2020-11-26 | 2021-02-23 | 国网天津市电力公司 | Big data center energy consumption analysis method and computing equipment |
CN113075995A (en) * | 2021-04-26 | 2021-07-06 | 华南理工大学 | Virtual machine energy-saving integration method, system and storage medium based on mixed group intelligence |
CN113448718A (en) * | 2020-03-26 | 2021-09-28 | 安徽寒武纪信息科技有限公司 | Method, apparatus and computer-readable storage medium for frequency modulation of a chip |
CN113821313A (en) * | 2020-12-28 | 2021-12-21 | 京东科技控股股份有限公司 | Task scheduling method and device and electronic equipment |
CN116881085A (en) * | 2023-09-05 | 2023-10-13 | 北京华鲲振宇智能科技有限责任公司 | Method for optimizing energy consumption of server |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050108712A1 (en) * | 2003-11-14 | 2005-05-19 | Pawan Goyal | System and method for providing a scalable on demand hosting system |
CN1764127A (en) * | 2004-10-18 | 2006-04-26 | 国际商业机器公司 | Method for distributing resource for high-grade user and data processor |
US20080243300A1 (en) * | 2007-03-26 | 2008-10-02 | Jack Liu | Object relocation guided by data cache miss profile |
CN101504620A (en) * | 2009-03-03 | 2009-08-12 | 华为技术有限公司 | Load balancing method, apparatus and system of virtual cluster system |
CN103488539A (en) * | 2013-09-23 | 2014-01-01 | 北京交通大学 | Data center energy saving method based on central processing unit (CPU) dynamic frequency modulation technology |
CN103576829A (en) * | 2012-08-01 | 2014-02-12 | 复旦大学 | Hybrid genetic algorithm based dynamic cloud-computing virtual machine scheduling method |
CN103699446A (en) * | 2013-12-31 | 2014-04-02 | 南京信息工程大学 | Quantum-behaved particle swarm optimization (QPSO) algorithm based multi-objective dynamic workflow scheduling method |
US20150074222A1 (en) * | 2013-09-12 | 2015-03-12 | Guanfeng Liang | Method and apparatus for load balancing and dynamic scaling for low delay two-tier distributed cache storage system |
CN104679594A (en) * | 2015-03-19 | 2015-06-03 | 成都艺辰德迅科技有限公司 | Middleware distributed calculating method |
CN104793993A (en) * | 2015-04-24 | 2015-07-22 | 江南大学 | Cloud computing task scheduling method of artificial bee colony particle swarm algorithm based on Levy flight |
-
2015
- 2015-10-26 CN CN201510703546.8A patent/CN105302630B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050108712A1 (en) * | 2003-11-14 | 2005-05-19 | Pawan Goyal | System and method for providing a scalable on demand hosting system |
CN1764127A (en) * | 2004-10-18 | 2006-04-26 | 国际商业机器公司 | Method for distributing resource for high-grade user and data processor |
US20080243300A1 (en) * | 2007-03-26 | 2008-10-02 | Jack Liu | Object relocation guided by data cache miss profile |
CN101504620A (en) * | 2009-03-03 | 2009-08-12 | 华为技术有限公司 | Load balancing method, apparatus and system of virtual cluster system |
CN103576829A (en) * | 2012-08-01 | 2014-02-12 | 复旦大学 | Hybrid genetic algorithm based dynamic cloud-computing virtual machine scheduling method |
US20150074222A1 (en) * | 2013-09-12 | 2015-03-12 | Guanfeng Liang | Method and apparatus for load balancing and dynamic scaling for low delay two-tier distributed cache storage system |
CN103488539A (en) * | 2013-09-23 | 2014-01-01 | 北京交通大学 | Data center energy saving method based on central processing unit (CPU) dynamic frequency modulation technology |
CN103699446A (en) * | 2013-12-31 | 2014-04-02 | 南京信息工程大学 | Quantum-behaved particle swarm optimization (QPSO) algorithm based multi-objective dynamic workflow scheduling method |
CN104679594A (en) * | 2015-03-19 | 2015-06-03 | 成都艺辰德迅科技有限公司 | Middleware distributed calculating method |
CN104793993A (en) * | 2015-04-24 | 2015-07-22 | 江南大学 | Cloud computing task scheduling method of artificial bee colony particle swarm algorithm based on Levy flight |
Non-Patent Citations (1)
Title |
---|
张燕 等: ""一种求解云计算资源优化的改进蝙蝠算法"", 《科技通报》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106936905A (en) * | 2017-03-07 | 2017-07-07 | 中国联合网络通信集团有限公司 | The dispatching method and its scheduling system of the Nova component virtual machines based on openstack |
CN108306780A (en) * | 2017-09-07 | 2018-07-20 | 上海金融期货信息技术有限公司 | A kind of system and method for the virtual machine communication quality self-optimizing based on cloud environment |
CN108306780B (en) * | 2017-09-07 | 2021-07-20 | 上海金融期货信息技术有限公司 | Cloud environment-based virtual machine communication quality self-optimization system and method |
CN107861802A (en) * | 2017-10-24 | 2018-03-30 | 郑州云海信息技术有限公司 | Method for managing resource and device in cloud data system |
CN108469983A (en) * | 2018-04-02 | 2018-08-31 | 西南交通大学 | A kind of virtual machine deployment method based on particle cluster algorithm under cloud environment |
CN110362388A (en) * | 2018-04-11 | 2019-10-22 | 中移(苏州)软件技术有限公司 | A kind of resource regulating method and device |
CN109491760A (en) * | 2018-10-29 | 2019-03-19 | 中国科学院重庆绿色智能技术研究院 | A kind of high-effect data center's Cloud Server resource autonomous management method and system |
CN109491760B (en) * | 2018-10-29 | 2021-10-19 | 中国科学院重庆绿色智能技术研究院 | High-performance data center cloud server resource autonomous management method |
CN109766190A (en) * | 2019-01-15 | 2019-05-17 | 无锡华云数据技术服务有限公司 | Cloud resource dispatching method, device, equipment and storage medium |
CN109960568A (en) * | 2019-02-18 | 2019-07-02 | 深圳大学 | A kind of dispatching method and electronic equipment |
CN109901932A (en) * | 2019-03-12 | 2019-06-18 | 东北大学 | A kind of Server Consolidation method based on virtual machine |
CN109901932B (en) * | 2019-03-12 | 2023-04-07 | 东北大学 | Server integration method based on virtual machine |
CN113448718A (en) * | 2020-03-26 | 2021-09-28 | 安徽寒武纪信息科技有限公司 | Method, apparatus and computer-readable storage medium for frequency modulation of a chip |
CN112395161A (en) * | 2020-11-26 | 2021-02-23 | 国网天津市电力公司 | Big data center energy consumption analysis method and computing equipment |
CN113821313A (en) * | 2020-12-28 | 2021-12-21 | 京东科技控股股份有限公司 | Task scheduling method and device and electronic equipment |
CN113075995A (en) * | 2021-04-26 | 2021-07-06 | 华南理工大学 | Virtual machine energy-saving integration method, system and storage medium based on mixed group intelligence |
CN113075995B (en) * | 2021-04-26 | 2024-02-27 | 华南理工大学 | Virtual machine energy-saving integration method, system and storage medium based on hybrid group intelligence |
CN116881085A (en) * | 2023-09-05 | 2023-10-13 | 北京华鲲振宇智能科技有限责任公司 | Method for optimizing energy consumption of server |
Also Published As
Publication number | Publication date |
---|---|
CN105302630B (en) | 2018-07-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105302630A (en) | Dynamic adjustment method and system for virtual machine | |
Luo et al. | Hybrid shuffled frog leaping algorithm for energy-efficient dynamic consolidation of virtual machines in cloud data centers | |
Zhong et al. | A cost-efficient container orchestration strategy in kubernetes-based cloud computing infrastructures with heterogeneous resources | |
Ranjbari et al. | A learning automata-based algorithm for energy and SLA efficient consolidation of virtual machines in cloud data centers | |
Li et al. | Energy-aware and multi-resource overload probability constraint-based virtual machine dynamic consolidation method | |
Kaur et al. | Energy efficiency techniques in cloud computing: A survey and taxonomy | |
Quang-Hung et al. | A genetic algorithm for power-aware virtual machine allocation in private cloud | |
Liao et al. | Towards a green cluster through dynamic remapping of virtual machines | |
Liang et al. | The placement method of resources and applications based on request prediction in cloud data center | |
Luo et al. | Simulation of power consumption of cloud data centers | |
Wang et al. | A green-aware virtual machine migration strategy for sustainable datacenter powered by renewable energy | |
Al-Dulaimy et al. | Type-aware virtual machine management for energy efficient cloud data centers | |
Kim et al. | Free cooling-aware dynamic power management for green datacenters | |
CN106951059A (en) | Based on DVS and the cloud data center power-economizing method for improving ant group algorithm | |
CN102662750A (en) | Virtual machine resource optimal control method and control system based on elastic virtual machine pool | |
Xu et al. | Resource pre-allocation algorithms for low-energy task scheduling of cloud computing | |
CN103823718A (en) | Resource allocation method oriented to green cloud computing | |
Zakarya et al. | Managing energy, performance and cost in large scale heterogeneous datacenters using migrations | |
Liang et al. | Memory-aware resource management algorithm for low-energy cloud data centers | |
Jin et al. | Energy-efficient task scheduling for CPU-intensive streaming jobs on Hadoop | |
Zhang et al. | An energy-aware host resource management framework for two-tier virtualized cloud data centers | |
Ghose et al. | Urgent point aware energy-efficient scheduling of tasks with hard deadline on virtualized cloud system | |
Yuan et al. | Energy aware resource scheduling algorithm for data center using reinforcement learning | |
Cui et al. | Shadows on the Cloud: An Energy-aware, Profit Maximizing Resilience Framework for Cloud Computing. | |
Zhu et al. | Energy-aware rolling-horizon scheduling for real-time tasks in virtualized cloud 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 |