CN108429815A - Dynamic resource scheduling method based on OpenStack - Google Patents
Dynamic resource scheduling method based on OpenStack Download PDFInfo
- Publication number
- CN108429815A CN108429815A CN201810245365.9A CN201810245365A CN108429815A CN 108429815 A CN108429815 A CN 108429815A CN 201810245365 A CN201810245365 A CN 201810245365A CN 108429815 A CN108429815 A CN 108429815A
- Authority
- CN
- China
- Prior art keywords
- node
- migration
- virtual machine
- cpu
- load
- 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.)
- Pending
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
- H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
-
- 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
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The present invention relates to a kind of dynamic resource scheduling methods based on OpenStack, belong to resource scheduling algorithm field.Algorithm begins through setting virtual machine (vm) migration trigger policy：The migration trigger condition of virtual machine is largely divided into two types, i.e. upper limit threshold triggering migration and lower threshold triggering migration.For both trigger conditions mainly due to the considerations of two different aspects, the wherein setting of upper limit threshold cannot be satisfied user demand, or operation is broken down because load too high leads to node primarily to avoid node load excessively high.Then the Ceilometer modules for utilizing OpenStack, obtain the every load attribute value and its consumption rate of all physical servers.The present invention reduces the quantity of operation node, has saved electric energy, avoids the incorrect migration generated by the concussion of moment.
Description
Technical field
The invention belongs to resource scheduling algorithm field, it is related to the dynamic resource scheduling method based on OpenStack.
Background technology
OpenStack is a cloud computing service platform increased income, for public cloud and private clound etc. provide one it is expansible,
Telescopic cloud computing service.Any enterprise or individual can build privately owned cloud platform using OpenStack to provide service,
Publicly-owned cloud platform can also be built to provide external cloud service.
With the large scale deployment of OpenStack platforms, cloud data center deploys thousands of physical node, at this
Different services and applications are run on a little physical nodes, to have higher requirement to OpenStack platforms.Separately
Outside, an important research direction in the fields OpenStack is the scheduling of resource of OpenStack, by OpenStack environment
The resources such as calculating, network, storage Optimized Operation, the service efficiency of OpenStack resources can be improved, and make be entirely
System is optimal performance according to demand.In OpenStack platforms, using resource dispatching strategy appropriate, it can rationally adjust and be
Various resources in system, and whole system is made to be optimal performance according to demand.With the increasing of task quantity and customer flow
Add, load will largely be added on individual server, if merely expanding cluster, can not make full use of resource, task
It distributes unbalanced.Therefore, it is necessary to use load-balancing technique that the request of user is evenly distributed on multiple servers.Load
Balancing technique is task based access control Thread Scheduling Algorithms, and the node being evenly distributed in cluster can improve the task of system
Processing capacity.Currently, aspect there are two main in the load-balancing technique research of OpenStack cloud platforms：First, by appropriate
Scheduling strategy select suitable calculate node to dispose virtual machine, second is that distributed software task requests are distributed to suitably
On virtual machine.Scheduling strategy in OpenStack environment mainly studies the selection of load information.There are document CPU, memory and net
Network flow proposes that the Virtual Machine Manager multiple-objection optimization virtual machine dynamic based on load manages as load information.There is document work
Person selects CPU and disk as load information, proposes to create resources of virtual machine pond, is increased or decreased according to loading condition virtual
Machine.In some documents, select CPU as load information, it is proposed that adaptive load balancing algorithm.According to loading condition, soon
The whole relevant parameter of velocity modulation dispatches the server resource in cluster.Also document is accessed using CPU and Web service as load information.
By analyzing these documents, the resource scheduling algorithm in OpenStack is studied, and uses CPU, memory and hard disk
The utilization rate of storage is as load information.
Invention content
In view of this, the purpose of the present invention is to provide a kind of dynamic resource scheduling method based on OpenStack, protecting
Card original system operates in the load balancing realized on the basis of kilter in cloud platform system operation, lifting system
Operational efficiency.
In order to achieve the above objectives, the present invention provides the following technical solutions：
Dynamic resource scheduling method based on OpenStack, this approach includes the following steps：
S1：Defined parameters；
CPUavgIndicate the average value of the consumption rate of the CPU attributes of all physical servers in dispatching zone；Dispatching zone n refers to
The number of all loaded physical servers, CPUiRefer to the CPU usage of some physical server in dispatching zone；
MEMavgIndicate the average value of the consumption rate of the memory attribute of all physical servers in dispatching zone；MEMiRefer to and adjusts
Spend the memory usage of some physical server in domain；
DISKavgIndicate the average value of the hard disk consumption of all physical servers in dispatching zone；DISKiRefer in dispatching zone
The hard disk consumption rate of some server；
wcpu, wmem, wdiskCPU, memory, the weighted value of each attribute of hard-disc storage are indicated respectively；PMcpu, PMmem, PMdiskPoint
The CPU of other representative server, memory, the capacity of hard disk, VMcpu, VMmem, VMdiskVirtual machine required CPU is respectively represented, it is interior
It deposits, the size of hard disk；
S2：Algorithm starts, and using the Ceilometer modules of OpenStack, the items for obtaining all physical servers are negative
Carry attribute value and its consumption rate；
S3：It checks whether upper limit early warning queue is empty, if upper limit early warning queue is sky, thens follow the steps S4 operations；If depositing
S6 operations are thened follow the steps in upper limit early warning node；
S4：It checks whether lower limit early warning queue is empty, if lower limit early warning queue is sky, thens follow the steps S5 operations；If depositing
S7 operations are thened follow the steps in lower limit early warning node；
S5：Check whether request queue is empty, if request queue is sky, algorithm terminates；It is held if there are request task
The S8 operations of row step；
S6：Upper limit early warning node is obtained, best migration node is checked for, if executing migration in the presence of if, if not depositing
It is then opening a new idle node and suitable virtual machine is selected to be migrated, then executing step S4 operations；
S7：Lower limit early warning node is obtained, best migration node is checked for, migration is executed if having, if being not present
Lower threshold detection is no longer then carried out by warning vertex ticks and in time t, then executes step S5 operations；
S8：All physics server lists are obtained, is filtered, is filtered out in available dispatching zone according to filtering setting
Physical server list, check for it is best create node, if in the presence of if by virtual machine creating on this node, otherwise will
Task creation is in a new idle node and the node detects in time t without lower threshold, and algorithm terminates.
Further, the selection of the best migration node is divided into three steps：
1) selection of source node：Directly transition condition is arranged in the factor in need for migrating virtual machine on decision node, will
After the load attribute information of all operation nodes is collected in system, the load information of each operation node is checked, wherein wrapping
Include CPU usage, memory consumption rate and hard disk utilization rate；When a certain or multinomial load attribute value in node is met or exceeded
The node is then added to upper limit early warning queue by limit threshold value, then will when every load attribute value in node is below lower threshold
The node is added to lower limit early warning queue；
2) on source node virtual machine to be migrated selection：A. it is the node higher than upper limit threshold if it is upper limit early warning queue
In virtual machine to be migrated selection, then select to need the virtual machine that migrates by the resource utilization of the server；Meeting
The virtual machine for selecting LF smaller in the virtual machine of condition is migrated, and the migration cost of virtual machine smaller LF is also smaller；
LvmRepresent the load capacity of virtual machine, LpmThe load capacity of physical server is represented, LF represents virtual machine to physics
The load effect factor of server；
Lpm=PMcpu×wcpu+PMmem×wmem+PMdisk×wdisk
Lvm=VMcpu×wcpu+VMmem×wmem+VMdisk×wdisk
B. namely it is less than the virtual machine in the node of lower threshold if it is lower limit early warning queue according to empty on server
The size of quasi- machine LF determines that the priority of migration sequence, virtual machine bigger LF first migrate, and it is i.e. determining all first to carry out simulation migration
Virtual machine unifies migration again after can moving to as desired on other nodes, if cannot if be considered as without best migration node i.e.
All virtual machines do not migrate and the node no longer carries out lower threshold inspection in time t；
3) selection of destination node：In migration strategy, the selection of destination node meets need by selection scheduling domain interior energy
Create or migrate all physical servers of virtual machine requirement and the value no more than setting；In the physical server for meeting condition
Middle selection LPMMinimum server.
Further, the migration trigger condition of the virtual machine is divided into two types：Upper limit threshold triggering migration and lower limit threshold
Value triggering migration；Wherein the setting of upper limit threshold cannot be satisfied user demand, or because negative for avoiding node load excessively high
Carry it is excessively high cause node operation break down；Setting lower threshold is used to reduce the quantity of operation node；The load information of node
The consumption rate of utilization rate and hard disk including CPU and memory, describes the behaviour in service of resource.
Further, the condition of the upper limit threshold triggering migration includes that CPU usage, memory consumption rate and hard disk utilize
Rate；As long as wherein there are one factors will be migrated more than threshold value；The condition of upper limit threshold triggering migration also needs to consider real
Shi Gengxin node load information is susceptible to instantaneous peak phenomenon, in order to avoid the incorrect migration generated due to the concussion because of moment,
Using the strategy of Time-delayed trigger, once some attribute value that is, in load information meets or exceeds defined threshold value in advance, just open
Beginning counts the consumption rate of subsequent load information, and node load attribute value is sampled and set according to certain time interval
Maximum times of collection Mmax is added to pre- police if statistical value all in the data collected at Mmax times is above threshold value
Row, otherwise will abandon monitoring.
Further, the condition of the lower threshold triggering migration is：In lower limit trigger condition, only item is triggered in the upper limit
Every attribute value of load information in part is worked at the same time and could be migrated under a low load；Once the every of load information belongs to
Property value is below lower threshold, just starts to count subsequent load information utilization rate, by node load information according to it is certain when
Between interval sampled and set maximum times of collection Nmax, if all statistical value is below in the data of collection at Nmax time
Lower threshold is then added to warning queue.
The beneficial effects of the present invention are：The present invention reduces the quantity of operation node, has saved electric energy, avoids because of wink
Between concussion and the incorrect migration that generates.
Description of the drawings
In order to keep the purpose of the present invention, technical solution and advantageous effect clearer, the present invention provides following attached drawing and carries out
Explanation：
Fig. 1 is flow chart of the present invention；
Fig. 2 is upper bound condition triggering migration flow chart；
Fig. 3 is lower limit condition triggering migration flow chart.
Specific implementation mode
Below in conjunction with attached drawing, the preferred embodiment of the present invention is described in detail.
Defined parameters
CPUavgIndicate the average value of the consumption rate of the CPU attributes of all physical servers in dispatching zone.Dispatching zone n is
Refer to the number of all loaded physical servers, CPUiRefer to the CPU usage of some physical server in dispatching zone.
MEMavgIndicate the average value of the consumption rate of the memory attribute of all physical servers in dispatching zone.MEMiRefer to and adjusts
Spend the memory usage of some physical server in domain.
DISKavgIndicate the average value of the hard disk consumption of all physical servers in dispatching zone.DISKiRefer in dispatching zone
The hard disk consumption rate of some server.
wcpu, wmem, wdiskCPU, memory, the weighted value of each attribute of hard-disc storage are indicated respectively.PMcpu, PMmem, PMdiskPoint
The CPU of other representative server, memory, the capacity of hard disk.VMcpu, VMmem, VMdiskVirtual machine required CPU is respectively represented, it is interior
It deposits, the size of hard disk.
Key step：
(1) algorithm starts, and using the Ceilometer modules of OpenStack, the items for obtaining all physical servers are negative
Carry attribute value and its consumption rate.
(2) it checks whether upper limit early warning queue is empty, if upper limit early warning queue is sky, executes the operation of (3) step；If depositing
The operation of (5) step is then executed in upper limit early warning node.
(3) it checks whether lower limit early warning queue is empty, if lower limit early warning queue is sky, executes the operation of (4) step；If depositing
The operation of (6) step is then executed in lower limit early warning node.
(4) check whether request queue is empty, if request queue is sky, algorithm terminates；It is held if there are request task
Row (7) step operates.
(5) upper limit early warning node is obtained, best migration node is checked for, if executing migration in the presence of if, if not depositing
It is then opening a new idle node and suitable virtual machine is selected to be migrated, then executing the operation of (3) step.
(6) lower limit early warning node is obtained, best migration node is checked for, migration is executed if having, if being not present
Lower threshold detection is no longer then carried out by warning vertex ticks and in time t, then executes the operation of (4) step.
(7) all physics server lists are obtained, is filtered, is filtered out in available dispatching zone according to filtering setting
Physical server list, check for it is best create node, if in the presence of if by virtual machine creating on this node, otherwise will
Task creation is in a new idle node and the node detects in time t without lower threshold, and algorithm terminates.
Flow of the present invention is as shown in Figure 1.
The management and running of virtual machine are divided into two aspects of initial deployment and dynamic migration of virtual machine, the initial portion of virtual machine
Administration refers to selecting most suitable physical server when creating virtual machine；The dynamic migration of virtual machine selects most suitable physics to take
It is engaged in device server as a purpose, virtual machine to be migrated is moved into destination server from source physical server.
Best migration scheme mentioned in flow, it is therefore intended that make the consumption rate of each physical server to a certain extent
Equilibrium, while also solving the selection of source node and source node virtual machine to be migrated and the selection of destination node.
The selection of best migration node is generally divided into three steps：
1) selection of source node
Directly the factor in need for migrating virtual machine is setting transition condition on decision node, by all operations in system
After the load attribute information of node is collected, the load information of each operation node is checked, wherein including mainly that CPU is used
Rate, memory consumption rate and hard disk utilization rate.When a certain or multinomial load attribute value in node meets or exceeds upper limit threshold then
The node is added to upper limit early warning queue, then adds the node when every load attribute value in node is below lower threshold
Enter to lower limit early warning queue.
2) on source node virtual machine to be migrated selection
A. the selection for being namely higher than virtual machine to be migrated in the node of upper limit threshold if it is upper limit early warning queue is then led to
The resource utilization of the server is crossed to select the virtual machine for needing to migrate, for example be higher than the clothes of upper limit threshold in CPU consumption rates
Selection CPU usage is higher on business device and value as defined in migrating before the consumption rate of CPU after the virtual machine will be less than.Meeting
Select the smaller virtual machines of LF to be migrated in the virtual machine of these conditions because the migration cost of virtual machine smaller LF also compared with
It is small.
LvmRepresent the load capacity of virtual machine, LpmRepresent the load capacity of physical server.LF represents virtual machine to physics
The load effect factor of server.
Lpm=PMcpu×wcpu+PMmem×wmem+PMdisk×wdisk
Lvm=VMcpu×wcpu+VMmem×wmem+VMdisk×wdisk
B. namely it is less than the virtual machine in the node of lower threshold if it is lower limit early warning queue according to empty on server
The size of quasi- machine LF determines that the priority of migration sequence, virtual machine bigger LF first migrate, and it is i.e. determining all first to carry out simulation migration
Virtual machine unifies migration again after can moving to as desired on other nodes, if cannot if be considered as without best migration node i.e.
All virtual machines do not migrate and the node no longer carries out lower threshold inspection in time t.
3) selection of destination node
In migration strategy, the selection of destination node is especially important, it directly influences the load of the operation node of system
Equilibrium situation.If the node chosen is unreasonable, unnecessary migration or secondary migration may be caused, it is negative to increase operation
Load and system energy consumption.The selection of destination node then by selection scheduling domain interior energy meet need create or migrate virtual machine require and
No more than all physical servers of the value of setting.L is selected in the physical server for meeting conditionPMMinimum server.
Virtual machine (vm) migration trigger policy：
The migration trigger policy of virtual machine is which type of carries out virtual machine (vm) migration operation under the conditions of for determining.Virtually
The migration trigger condition of machine is largely divided into two types, i.e. upper limit threshold triggering migration and lower threshold triggering migration.Both
Trigger condition is mainly due to the considerations of two different aspects, and wherein the setting of upper limit threshold is primarily to avoid node load
It is excessively high, user demand is cannot be satisfied, or operation is broken down because load too high leads to node.Setting lower threshold is mainly
In order to reduce the quantity of operation node to the greatest extent, in order to save electric energy.The load information of node includes the use of CPU and memory
The consumption rate of rate and hard disk, can more accurately describe the behaviour in service of resource, and single CPU, memory or hard disk utilize
Rate cannot really reflect the service condition of resource.
The setting of upper limit trigger condition
CPU usage, memory consumption rate and hard disk utilization are the factors considered required for upper limit trigger condition.If its
In there are one factor be more than threshold value will be migrated.The problem that the upper limit trigger condition of migration also needs to consider is real-time
Update node load information is susceptible to instantaneous peak phenomenon, but can lower rapidly in a short time.If load information
Just migrated up to or over defined threshold value, then will carry out when encounter instantaneous peak value the case where it is many unnecessarily
Migration, to waste the computing resource of system and increase the resource consumption of system.In order to avoid being generated due to the concussion because of moment
Incorrect migration, we use Time-delayed trigger strategy, get over once some attribute value that is, in load information is met or exceeded
Threshold value as defined in elder generation just starts to count the consumption rate of subsequent load information, by node load attribute value according to the regular hour
Interval is sampled and is set maximum times of collection Mmax, if statistical value all in the data collected at Mmax times is above
Threshold value is then added to early warning queue, otherwise will abandon monitoring.Upper bound condition triggering migration flow is as shown in Figure 2.
The setting of lower limit trigger condition：
The attribute value of load information in upper limit trigger condition, as long as wherein a certain item has met or exceeded threshold value and just needed
It is migrated.That is the triggering migration at this time in order to avoid influencing the work quality of node because of the inadequate resource in terms of certain
Just it can effectively ensure the running quality of all virtual machines on node.But different is in lower limit trigger condition, only
Every attribute of load is worked at the same time and could be migrated under a low load.Because of the type meeting of resource needed for different virtual machines
It makes a big difference, the utilization rate of certain resources can be relatively high on working node in some cases, but some other resource
May be relatively idle, the running quality of virtual machine may be made if carrying out virtual machine (vm) migration in this case
At prodigious influence, and the load of other working nodes is virtually increased, or even triggers the migration task of chain type.
Reducing the quantity of operation node to the greatest extent, and it is minimum to ensure that the scheduling of virtual machine influences all operation nodes
Under the premise of, it proposes to trigger migration strategy using the lower limit of lower threshold combination Time-delayed trigger, once the i.e. items of load information
Attribute value is below lower threshold, just starts to count subsequent load information utilization rate, by node load information according to certain
Time interval is sampled and is set maximum times of collection Nmax, if statistical value all in the data collected at Nmax times is all
Warning queue is then added to less than lower threshold.Lower limit condition triggering migration flow is as shown in Figure 3.
(1) algorithm begins through setting virtual machine (vm) migration trigger policy：The migration trigger condition of virtual machine is largely divided into two
Type, i.e. upper limit threshold triggering migration and lower threshold triggering migration.Both trigger conditions are mainly due to two differences
The considerations of aspect, the wherein setting of upper limit threshold, primarily to avoid node load excessively high, cannot be satisfied user demand, or
Because load too high leads to node, operation is broken down.Setting lower threshold is primarily to reduce the number of operation node to the greatest extent
Amount, in order to save electric energy.Then the Ceilometer modules for utilizing OpenStack, obtain the items of all physical servers
Load attribute value and its consumption rate.
(2) once with reference to Fig. 2 it is found that some attribute value in load information met or exceeded in advance as defined in threshold value,
Just the consumption rate for starting to count subsequent load information, by node load attribute value according to certain time interval carry out sampling and
The maximum times of collection Mmax of setting, is added to if statistical value all in the data collected at Mmax times has been above threshold value
Otherwise upper limit early warning queue will be abandoned monitoring.
(3) once with reference to Fig. 3 it is found that every attribute value of load information is below lower threshold, it is subsequent just to start statistics
Load information utilization rate, node load information is sampled according to certain time interval and sets maximum times of collection
Nmax is added to lower limit early warning queue if statistical value all in the data collected at Nmax times is below lower threshold,
Otherwise monitoring will be abandoned and do not carrying out lower threshold detection in time t.
(4) referring to Fig.1 it is found that check whether early warning queue is empty, if not having early warning task, that is, next step is executed；If having
Early warning node then is obtained from upper limit early warning queue, checks for best migration node, it is suitable empty if being selected in the presence of if
Quasi- machine is migrated, and if one new idle node of opening there is no and if suitable virtual machine is selected to be migrated.It is to be migrated
The selection of virtual machine then selects the virtual machine for needing to migrate by the resource utilization of the server, such as in CPU consumption rates
Higher than selecting CPU usage higher on the server of upper limit threshold and migrate the consumption rate of CPU after the virtual machine to will be less than it
Value as defined in preceding.The virtual machine for selecting LF smaller in the virtual machine for meeting these conditions is migrated, because of void smaller LF
The migration cost of quasi- machine is also smaller.
The selection of destination node then needs to migrate virtual machine requirement and no more than setting by selection scheduling domain interior energy satisfaction
Value all physical servers.The server of Lpm minimums is selected in the physical server for meeting condition.Then it executes next
Step
(5) check whether lower limit early warning queue is empty, and next step is executed if without task；If any then from the pre- police of lower limit
Early warning node is obtained in row, checks for best migration node, virtual machine to be migrated is according to virtual machine on server
The size of Lvm determines that the priority of migration sequence, virtual machine bigger Lvm first migrate, and first carries out simulation migration and determines all void
Quasi- machine unifies migration again after can moving to as desired on other nodes, if executing migration in the presence of if, if there is no will
Early warning vertex ticks and lower threshold detection is no longer carried out in time t, then execute next step.
(6) check whether request queue is whether empty i.e. user applies for establishment virtual machine, if without if algorithm terminate；If having
All physics server lists in dispatching zone are then obtained, is filtered, is filtered out in available dispatching zone according to filtering setting
Physical server list, check for it is best create node, if in the presence of if by virtual machine creating on this node, otherwise create
It build in a new idle node and is detected without lower threshold in time t, algorithm terminates.
Finally illustrate, preferred embodiment above is merely illustrative of the technical solution of the present invention and unrestricted, although logical
It crosses above preferred embodiment the present invention is described in detail, however, those skilled in the art should understand that, can be
Various changes are made to it in form and in details, without departing from claims of the present invention limited range.
Claims (5)
1. the dynamic resource scheduling method based on OpenStack, it is characterised in that：This approach includes the following steps：
S1：Defined parameters；
CPUavgIndicate the average value of the consumption rate of the CPU attributes of all physical servers in dispatching zone；Dispatching zone n refers to all
The number of loaded physical server, CPUiRefer to the CPU usage of some physical server in dispatching zone；
MEMavgIndicate the average value of the consumption rate of the memory attribute of all physical servers in dispatching zone；MEMiRefer in dispatching zone
The memory usage of some physical server；
DISKavgIndicate the average value of the hard disk consumption of all physical servers in dispatching zone；DISKiRefer to some in dispatching zone to take
The hard disk consumption rate of business device；
wcpu, wmem, wdiskCPU, memory, the weighted value of each attribute of hard-disc storage are indicated respectively；PMcpu, PMmem, PMdiskGeneration respectively
The CPU of list server, memory, the capacity of hard disk, VMcpu, VMmem, VMdiskRespectively represent the required CPU of virtual machine, memory,
The size of hard disk；
S2：Algorithm starts, and using the Ceilometer modules of OpenStack, the every load for obtaining all physical servers belongs to
Property value and its consumption rate；
S3：It checks whether upper limit early warning queue is empty, if upper limit early warning queue is sky, thens follow the steps S4 operations；If there are upper
Limit early warning node thens follow the steps S6 operations；
S4：It checks whether lower limit early warning queue is empty, if lower limit early warning queue is sky, thens follow the steps S5 operations；In the presence of if
Limit early warning node thens follow the steps S7 operations；
S5：Check whether request queue is empty, if request queue is sky, algorithm terminates；Step is executed if there are request task
Rapid S8 operations；
S6：Upper limit early warning node is obtained, best migration node is checked for, if executing migration in the presence of if, if being not present
It opens a new idle node and suitable virtual machine is selected to be migrated, then execute step S4 operations；
S7：Lower limit early warning node is obtained, best migration node is checked for, migration is executed if having, if there is no will
It alerts vertex ticks and no longer carries out lower threshold detection in time t, then execute step S5 operations；
S8：All physics server lists are obtained, is filtered according to filtering setting, filters out the physics in available dispatching zone
Server list, check for it is best create node, if in the presence of if by virtual machine creating on this node, otherwise by task
It creates in a new idle node and the node detects in time t without lower threshold, algorithm terminates.
2. the dynamic resource scheduling method according to claim 1 based on OpenStack, it is characterised in that：It is described best
The selection of migration node is divided into three steps：
1) selection of source node：Directly transition condition is arranged in the factor in need for migrating virtual machine on decision node, by system
In all operation nodes load attribute information collect after, the load information of each operation node is checked, including CPU
Utilization rate, memory consumption rate and hard disk utilization rate；When a certain or multinomial load attribute value in node meets or exceeds upper limit threshold
The node is then added to upper limit early warning queue by value, when every load attribute value in node is below lower threshold then by the section
Point is added to lower limit early warning queue；
2) on source node virtual machine to be migrated selection：A. it is to be waited in the node higher than upper limit threshold if it is upper limit early warning queue
The selection of virtual machine is migrated, then the virtual machine for needing to migrate is selected by the resource utilization of the server；Meeting condition
Virtual machine in the smaller virtual machines of selection LF migrated, the migration cost of virtual machine smaller LF is also smaller；
LvmRepresent the load capacity of virtual machine, LpmThe load capacity of physical server is represented, LF represents virtual machine to physical services
The load effect factor of device；
Lpm=PMcpu×wcpu+PMmem×wmem+PMdisk×wdisk
Lvm=VMcpu×wcpu+VMmem×wmem+VMdisk×wdisk
B. namely it is less than the virtual machine in the node of lower threshold if it is lower limit early warning queue according to virtual machine on server
The size of LF determines that the priority of migration sequence, virtual machine bigger LF first migrate, and it is i.e. determining all virtual first to carry out simulation migration
Machine unifies migration again after can moving to as desired on other nodes, if cannot if be considered as and own without best migration node
Virtual machine do not migrate and the node no longer carries out lower threshold inspection in time t；
3) selection of destination node：In migration strategy, the selection of destination node needs to create by selection scheduling domain interior energy satisfaction
Build or migrate all physical servers of virtual machine requirement and the value no more than setting；It is selected in the physical server for meeting condition
Select LPMMinimum server.
3. the dynamic resource scheduling method according to claim 1 or 2 based on OpenStack, it is characterised in that：The void
The migration trigger condition of quasi- machine is divided into two types：Upper limit threshold triggering migration and lower threshold triggering migration；Wherein upper limit threshold
The setting of value cannot be satisfied user demand, or because load too high causes node to run out for avoiding node load excessively high
Existing failure；Setting lower threshold is used to reduce the quantity of operation node；The load information of node includes the utilization rate of CPU and memory
And the consumption rate of hard disk.
4. the dynamic resource scheduling method according to claim 3 based on OpenStack, it is characterised in that：The upper limit
The condition of threshold triggers migration includes CPU usage, memory consumption rate and hard disk utilization；As long as being wherein more than there are one factor
Threshold value will be migrated；The condition of upper limit threshold triggering migration also needs to consider that real-time update node load information is susceptible to
Instantaneous peak phenomenon, it is using the strategy of Time-delayed trigger, i.e., once negative in order to avoid the incorrect migration generated due to the concussion because of moment
Some attribute value in information carrying breath meets or exceeds defined threshold value in advance, just starts to count the consumption of subsequent load information
Node load attribute value is sampled according to certain time interval and sets maximum times of collection Mmax, if in Mmax by rate
All statistical values are above threshold value in the data of secondary collection, then are added to upper limit early warning queue, otherwise will abandon monitoring.
5. the dynamic resource scheduling method according to claim 3 based on OpenStack, it is characterised in that：The lower limit
Threshold triggers migration condition be：In lower limit trigger condition, the every of load information only in upper limit trigger condition belongs to
Property value is worked at the same time and could be migrated under a low load；Once every attribute value of load information is below lower threshold, just
Start to count subsequent load information utilization rate, node load information is sampled according to certain time interval and is set most
Big times of collection Nmax is added to lower limit police if statistical value all in the data collected at Nmax times is below lower threshold
Accuse queue.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810245365.9A CN108429815A (en) | 2018-03-23 | 2018-03-23 | Dynamic resource scheduling method based on OpenStack |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810245365.9A CN108429815A (en) | 2018-03-23 | 2018-03-23 | Dynamic resource scheduling method based on OpenStack |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108429815A true CN108429815A (en) | 2018-08-21 |
Family
ID=63158941
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810245365.9A Pending CN108429815A (en) | 2018-03-23 | 2018-03-23 | Dynamic resource scheduling method based on OpenStack |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108429815A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109445931A (en) * | 2018-08-31 | 2019-03-08 | 安徽四创电子股份有限公司 | A kind of big data resource scheduling system and method |
| CN110134495A (en) * | 2019-05-21 | 2019-08-16 | 山东大学 | A kind of container is across the online moving method of host, storage medium and terminal device |
| CN110348681A (en) * | 2019-06-04 | 2019-10-18 | 国网浙江省电力有限公司衢州供电公司 | A kind of electric power CPS dynamic load distribution method |
| CN110399214A (en) * | 2019-05-30 | 2019-11-01 | 腾讯科技（深圳）有限公司 | A kind of method, apparatus and computer equipment of the load of optimization video card |
| CN110865872A (en) * | 2019-11-14 | 2020-03-06 | 北京京航计算通讯研究所 | Virtualized cluster resource scheduling system based on resource rationalization application |
| CN110995863A (en) * | 2019-12-19 | 2020-04-10 | 上海交通大学 | Data center load distribution method and system based on load demand characteristics |
| CN112395045A (en) * | 2020-11-13 | 2021-02-23 | 深圳力维智联技术有限公司 | Virtual machine recovery and resource adjustment method thereof |
| CN112416530A (en) * | 2020-12-08 | 2021-02-26 | 西藏宁算科技集团有限公司 | Method and device for flexibly managing cluster physical machine nodes and electronic equipment |
| CN112433858A (en) * | 2020-12-17 | 2021-03-02 | 济南浪潮数据技术有限公司 | Load distribution method, device, equipment and readable storage medium |
| CN113568746A (en) * | 2021-07-27 | 2021-10-29 | 北京达佳互联信息技术有限公司 | Load balancing method and device, electronic equipment and storage medium |
| CN115277710A (en) * | 2022-05-31 | 2022-11-01 | 马上消费金融股份有限公司 | Resource management method, device, electronic equipment and storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102314391A (en) * | 2011-05-27 | 2012-01-11 | 运软网络科技(上海)有限公司 | Architecture and method for monitoring performance of virtualization system based on hardware basis tools |
| CN102932418A (en) * | 2012-09-27 | 2013-02-13 | 东软集团股份有限公司 | Cloud application deploying method and device |
| CN103607459A (en) * | 2013-11-21 | 2014-02-26 | 东北大学 | Dynamic resource monitoring and scheduling method of cloud computing platform IaaS layer |
| US20140207944A1 (en) * | 2013-01-24 | 2014-07-24 | Hitachi, Ltd. | Method and system for managing cloud computing environment |
| US20150039764A1 (en) * | 2013-07-31 | 2015-02-05 | Anton Beloglazov | System, Method and Computer Program Product for Energy-Efficient and Service Level Agreement (SLA)-Based Management of Data Centers for Cloud Computing |
| CN104572307A (en) * | 2015-01-30 | 2015-04-29 | 无锡华云数据技术服务有限公司 | Method for flexibly scheduling virtual resources |
| CN104636197A (en) * | 2015-01-29 | 2015-05-20 | 东北大学 | Evaluation method for data center virtual machine migration scheduling strategies |
| CN105049536A (en) * | 2015-09-08 | 2015-11-11 | 南京大学 | Load balancing system and load balancing method in IaaS (Infrastructure As A Service) cloud environment |
| CN106970831A (en) * | 2017-05-15 | 2017-07-21 | 金航数码科技有限责任公司 | The resources of virtual machine dynamic scheduling system and method for a kind of facing cloud platform |
| CN107203255A (en) * | 2016-03-20 | 2017-09-26 | 田文洪 | Power-economizing method and device are migrated in a kind of network function virtualized environment |
-
2018
- 2018-03-23 CN CN201810245365.9A patent/CN108429815A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102314391A (en) * | 2011-05-27 | 2012-01-11 | 运软网络科技(上海)有限公司 | Architecture and method for monitoring performance of virtualization system based on hardware basis tools |
| CN102932418A (en) * | 2012-09-27 | 2013-02-13 | 东软集团股份有限公司 | Cloud application deploying method and device |
| US20140207944A1 (en) * | 2013-01-24 | 2014-07-24 | Hitachi, Ltd. | Method and system for managing cloud computing environment |
| US20150039764A1 (en) * | 2013-07-31 | 2015-02-05 | Anton Beloglazov | System, Method and Computer Program Product for Energy-Efficient and Service Level Agreement (SLA)-Based Management of Data Centers for Cloud Computing |
| CN103607459A (en) * | 2013-11-21 | 2014-02-26 | 东北大学 | Dynamic resource monitoring and scheduling method of cloud computing platform IaaS layer |
| CN104636197A (en) * | 2015-01-29 | 2015-05-20 | 东北大学 | Evaluation method for data center virtual machine migration scheduling strategies |
| CN104572307A (en) * | 2015-01-30 | 2015-04-29 | 无锡华云数据技术服务有限公司 | Method for flexibly scheduling virtual resources |
| CN105049536A (en) * | 2015-09-08 | 2015-11-11 | 南京大学 | Load balancing system and load balancing method in IaaS (Infrastructure As A Service) cloud environment |
| CN107203255A (en) * | 2016-03-20 | 2017-09-26 | 田文洪 | Power-economizing method and device are migrated in a kind of network function virtualized environment |
| CN106970831A (en) * | 2017-05-15 | 2017-07-21 | 金航数码科技有限责任公司 | The resources of virtual machine dynamic scheduling system and method for a kind of facing cloud platform |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109445931A (en) * | 2018-08-31 | 2019-03-08 | 安徽四创电子股份有限公司 | A kind of big data resource scheduling system and method |
| CN110134495A (en) * | 2019-05-21 | 2019-08-16 | 山东大学 | A kind of container is across the online moving method of host, storage medium and terminal device |
| CN110399214A (en) * | 2019-05-30 | 2019-11-01 | 腾讯科技（深圳）有限公司 | A kind of method, apparatus and computer equipment of the load of optimization video card |
| CN110399214B (en) * | 2019-05-30 | 2024-03-22 | 腾讯科技（深圳）有限公司 | Method and device for optimizing display card load and computer equipment |
| CN110348681B (en) * | 2019-06-04 | 2022-02-18 | 国网浙江省电力有限公司衢州供电公司 | Power CPS dynamic load distribution method |
| CN110348681A (en) * | 2019-06-04 | 2019-10-18 | 国网浙江省电力有限公司衢州供电公司 | A kind of electric power CPS dynamic load distribution method |
| CN110865872A (en) * | 2019-11-14 | 2020-03-06 | 北京京航计算通讯研究所 | Virtualized cluster resource scheduling system based on resource rationalization application |
| CN110995863A (en) * | 2019-12-19 | 2020-04-10 | 上海交通大学 | Data center load distribution method and system based on load demand characteristics |
| CN112395045A (en) * | 2020-11-13 | 2021-02-23 | 深圳力维智联技术有限公司 | Virtual machine recovery and resource adjustment method thereof |
| CN112416530B (en) * | 2020-12-08 | 2023-12-22 | 西藏宁算科技集团有限公司 | Method and device for elastically managing cluster physical machine nodes and electronic equipment |
| CN112416530A (en) * | 2020-12-08 | 2021-02-26 | 西藏宁算科技集团有限公司 | Method and device for flexibly managing cluster physical machine nodes and electronic equipment |
| CN112433858A (en) * | 2020-12-17 | 2021-03-02 | 济南浪潮数据技术有限公司 | Load distribution method, device, equipment and readable storage medium |
| CN113568746A (en) * | 2021-07-27 | 2021-10-29 | 北京达佳互联信息技术有限公司 | Load balancing method and device, electronic equipment and storage medium |
| CN115277710A (en) * | 2022-05-31 | 2022-11-01 | 马上消费金融股份有限公司 | Resource management method, device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108429815A (en) | Dynamic resource scheduling method based on OpenStack | |
| CN108182105B (en) | Local dynamic migration method and control system based on Docker container technology | |
| CN105159751B (en) | The virtual machine migration method of energy efficient in a kind of cloud data center | |
| CN102981910B (en) | The implementation method of scheduling virtual machine and device | |
| CN104102543B (en) | The method and apparatus of adjustment of load in a kind of cloud computing environment | |
| CN107197053A (en) | A kind of load-balancing method and device | |
| CN111796908B (en) | System and method for automatic elastic expansion and contraction of resources and cloud platform | |
| CN106095531B (en) | A kind of dispatching method of virtual machine loaded based on grade and physical machine in cloud platform | |
| CN110058932A (en) | A kind of storage method and storage system calculated for data flow driven | |
| WO2011110026A1 (en) | Method and apparatus for realizing load balance of resources in data center | |
| CN105893113A (en) | Management system and management method of virtual machine | |
| CN102981890A (en) | Computing task and virtual machine deploying method within a virtual data center | |
| Lai et al. | Sol: Fast distributed computation over slow networks | |
| WO2023109068A1 (en) | Automatic virtual machine migration decision-making method based on user experience in multi-cloud environment | |
| CN109271257A (en) | A kind of method and apparatus of virtual machine (vm) migration deployment | |
| CN105868004A (en) | Cloud computing based business system scheduling method and apparatus | |
| CN104199724B (en) | A kind of virtual resources method for optimizing scheduling based on cost performance | |
| CN107872517A (en) | A kind of data processing method and device | |
| CN112380005A (en) | Data center energy consumption management method and system | |
| Li et al. | A dynamic I/O sensing scheduling scheme in Kubernetes | |
| CN107479974A (en) | A kind of dispatching method of virtual machine and device | |
| CN113672391A (en) | Parallel computing task scheduling method and system based on Kubernetes | |
| Lin et al. | A workload-driven approach to dynamic data balancing in MongoDB | |
| CN109144664A (en) | A kind of dynamic migration of virtual machine method based on user's QoS requirement difference | |
| CN106713051A (en) | Network management system |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180821 |
|
| RJ01 | Rejection of invention patent application after publication |