CN109558216A - It is a kind of that optimization method and its system are virtualized based on the single I/O migrated online - Google Patents
It is a kind of that optimization method and its system are virtualized based on the single I/O migrated online Download PDFInfo
- Publication number
- CN109558216A CN109558216A CN201811507379.XA CN201811507379A CN109558216A CN 109558216 A CN109558216 A CN 109558216A CN 201811507379 A CN201811507379 A CN 201811507379A CN 109558216 A CN109558216 A CN 109558216A
- Authority
- CN
- China
- Prior art keywords
- virtual machine
- network
- monitoring module
- data
- virtual
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/4557—Distribution of virtual machine instances; Migration and load balancing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/45579—I/O management, e.g. providing access to device drivers or storage
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/45595—Network integration; Enabling network access in virtual machine instances
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
It is the present invention relates to field of cloud computer technology, in particular to a kind of based on the single I/O migrated online virtualization optimization method and its system；The network load situation of change of elder generation's dynamic sensing virtual machine of the present invention and the network data information for acquiring virtual machine；Again after network monitoring module receiving network data information, is classified to virtual machine and distribute VF；Then network monitoring module is monitored the network state of virtual machine, if network monitoring module is more than the threshold value of setting to the information data that virtual machine monitoring arrives, the virtual machine of monitoring is migrated；The present invention is capable of the network load of dynamic sensing virtual machine, classification is carried out to virtual machine and the network performance of virtual machine is promoted to the limited VF of virtual machine quantity allotted, to the network performance of lifting system entirety, and using the network pressure of the method alleviation host migrated online to virtual machine.
Description
Technical field
It is the present invention relates to field of cloud computer technology, in particular to a kind of based on the single I/O migrated online virtualization optimization
Method and its system.
Background technique
With the fast development of cloud computing technology, modern data center nearly all deploys virtualization technology, and virtual machine is soft
The separation of part and hardware preferably can carry out the operation such as software management, fault detection and system maintenance in data center.Pass through
The modes such as data isolation run more virtual servers on a physical server, and virtualization technology can increase substantially service
The physical resource utilization rate of device, and therefore the performance of virtual server can't be much affected.For at present, CPU
Comparatively comparative maturity, I/O virtualization technology have become the crucial institute of system for restricting performance with the virtualization technology of memory
?.
Internet resources are one kind particularly important in I/O resource, simultaneously because the scarcity of data center network resource and
High competition, the promotion of rational management and data center network bandwidth availability ratio to Internet resources in virtualized environment are anxious
A critical issue to be solved.
Network I/O virtualization is to solve the problems, such as this key technology, but in several virtualization sides I/O of current mainstream
In formula, none of these methods can meet again the performance needs of virtual machine while guaranteeing scalability.Software-based void
Quasi-ization has a good scalability, but due to monitor of virtual machine can the path I/O to virtual machine intervened and make its property
It can be limited.The virtualization mode either Pass-through technology of hardware auxiliary or single I/O virtualization technology all bypass
The intervention to the path virtual machine I/O of monitor of virtual machine, can reach the I/O performance close to physical machine, but it is destroyed
The service transparency of virtualization brings difficulty to the online migration of virtual machine.
Summary of the invention
Optimization side is virtualized based on the single I/O migrated online the invention mainly solves the technical problem of providing a kind of
Method is capable of the network load of dynamic sensing virtual machine, and the VF for classifying and being limited to virtual machine quantity allotted is carried out to virtual machine
To promote the network performance of virtual machine, thus the network performance of lifting system entirety, and using being migrated online to virtual machine
Method alleviates the network pressure of host, the case where delay machine occurs in order to avoid mainframe network overload, negative in mainframe network
Carrying also will not migrate virtual machine beyond its tolerance range；The present invention also provides a kind of based on the single I/ migrated online
O virtualizes optimization system.
In order to solve the above technical problems, one technical scheme adopted by the invention is that: it provides a kind of based on online migrating
Single I/O virtualizes optimization method, wherein includes the following steps:
Step S1, the network load situation of change of dynamic sensing virtual machine and acquire the network data information of virtual machine, will
The network data information is sent to network monitoring module；
Step S2, after network monitoring module receiving network data information, classified to virtual machine and distribute VF；
Step S3, network monitoring module is monitored the network state of virtual machine, if network monitoring module is to virtual
The information data that machine monitoring arrives is more than the threshold value of setting, then migrates the virtual machine of monitoring.
As an improvement of the present invention, step S1 includes:
Step S11, the flow information that virtual machine will acquire at the end of each sampling period is sent to network monitoring module；
Step S12, virtual machine reads its memory information currently in use, will acquire at the end of each sampling period in
It deposits information and is sent to host.
As a further improvement of the present invention, in step S2, network monitoring module network data information based on the received
Virtual machine is categorized into network-intensive virtual machine and non-network intensive virtual machine, distributes VF to network-intensive virtual machine,
And non-network intensive virtual machine is allowed to work using paravirtualized network interface card.
As a further improvement of the present invention, in step S2, the activation backup mode of Bonding actuation techniques is used
By single I/O virtualization Network card setup be main equipment and paravirtualized Network card setup is from equipment, network monitoring module creation
Two queues are managed virtual machine, have just started for all virtual machines to be put into generic queue, terminate when each sampling period
When by virtual machine according to this flow and descending sequence, choose a certain number of virtual machines and be put into Priority Queues, it is remaining
Virtual machine is put into generic queue, then distributes VF to the virtual machine in Priority Queues, and the virtual machine in generic queue makes
It is worked with paravirtualized network interface card.
As a further improvement of the present invention, step S3 includes:
Step S31, network monitoring module is monitored the network state of virtual machine, while the network for analyzing virtual machine is close
Collection degree；If network monitoring module is more than the threshold value of setting to the information data that virtual machine monitoring arrives, enter next step；
Step S32, predict the next cycle of the network state of the virtual machine information data whether be more than setting threshold
Value, if it exceeds the threshold value of setting, then enter next step；
Step S33, network monitoring module selectes the virtual machine for the threshold value that its information data is more than setting and by the virtual machine
It migrates to migration management module；
Step S34, the smallest host of network-intensive state in selection cluster is received the void of the migration by migration management module
Quasi- machine；
If step S35, the information data of the virtual machine of the migration is more than the threshold value of setting, repeatedly step S34.
As a further improvement of the present invention, in step S31, information data is stored in file by network monitoring module
Netinfo and the network-intensive degree for analyzing virtual machine simultaneously；The method that the network-intensive degree of analysis virtual machine uses is institute
It states network monitoring module and reads n sampled data in the past from file, if wherein there is threshold value of the k number evidence all beyond setting,
S33 is then entered step, deposit data is otherwise constantly repeated, read data and analyzes the process of data.
As a further improvement of the present invention, in step S32, the next cycle of the network state of virtual machine is predicted
The method of information data be to be predicted using the Self-regression Forecast Model based on time series.
As a further improvement of the present invention, in step S33, will be migrated on network monitoring module selection host
Virtual machine, and the network data information and memory information of virtual machine in file are read, based on making per unit byte transmission maximum
The IP address of the virtual machine is simultaneously sent to migration management module by the principle of load, the selected virtual machine that will be migrated.
It is a kind of that optimization system is virtualized based on the single I/O migrated online, wherein to include:
Several virtual machines；
Network monitoring module, the network data letter sent for each virtual machine being monitored and being received each virtual machine
Breath；
Migration management module when for receiving migration trigger signal, reads the network data letter of each virtual machine in file
The network bandwidth utilization factor of total flow and the method calculating main frame divided by network interface card bandwidth that breath sends and receives according to host, will lead
The sequence sequence that the bandwidth availability ratio of machine successively decreases from big to small, selects wherein the smallest host, by virtual machine (vm) migration to the host.
As an improvement of the present invention, further includes:
VF scheduler module, for according to the descending sequence of network flow in this sampling period, choosing virtual machine
The a certain number of virtual machines in front are put into Priority Queues, and remaining virtual machine is put into generic queue.
The beneficial effects of the present invention are: compared with prior art, the present invention is capable of the network load of dynamic sensing virtual machine,
Classification is carried out to virtual machine and promotes the network performance of virtual machine to the limited VF of virtual machine quantity allotted, thus lifting system
Whole network performance, and using the network pressure of the method alleviation host migrated online to virtual machine, in order to avoid host
Network load is overweight and the case where delay machine occurs, will not also carry out to virtual machine beyond its tolerance range in mainframe network load
Migration.
Detailed description of the invention
Fig. 1 is the step block diagram of the invention that optimization method is virtualized based on the single I/O migrated online；
Fig. 2 is the internal step block diagram of step S1 in Fig. 1；
Fig. 3 is the internal step block diagram of step S3 in Fig. 1；
Fig. 4 is the scheduling process schematic diagram of distribution VF of the invention；
Fig. 5 is the scheduling process schematic diagram of online migration virtual machine of the invention；
Fig. 6 is the scheduling process frame diagram of distribution VF of the invention；
Fig. 7 is the scheduling process frame diagram of online migration virtual machine of the invention；
Fig. 8 is the block diagram of the invention that optimization system is virtualized based on the single I/O migrated online.
Specific embodiment
Current virtual machine distribution and scheduling strategy are mainly the economy and height that care data center's computing resource utilizes
Effect property, and for there is also a degree of missings in terms of network resource scheduling.For calculating and storage resource, network
Resource still belongs to high competition and rare resource in data center.If rational management cannot be carried out to Internet resources, it will shadow
Ring the performance of whole system.It may be had simultaneously especially after data center deployment virtualization technology, on same host more
A virtual machine issues Internet resources and requests, this will cause the competition to Internet resources fiercer.Data center is often
Optimize the network performance of virtual machine by using single I/O virtualization technology.
In data center, single I/O virtualization technology in fact has become the standard technique of network I/O virtualization.Phase
Than in half virtualization technology, the communication of virtual machine can be greatly lowered in single I/O virtualization technology by way of monopolizing VF
Delay, but if each virtual machine distribution VF will be unable to satisfy the demand of network-intensive virtual machine communication, because often
Effective bandwidth acquired in a VF is inversely proportional to the quantity for configuring VF in total, and increasing for VF quantity will lead under single VF bandwidth
Drop.Further, since the changeability of application services in the process of running, system after running for a period of time may be negative in network
Become no longer balanced in load, lead to some hosts because network load is overweight and performance declines, and some hosts are because network is born
Load is inadequate and resource utilization is too low, if not advanced optimizing the scalability that necessarily will affect system to single I/O virtualization
And network performance.But using single I/O virtualization technology, there is also deficiencies, show two aspects: (1) when single I/O is empty
When the VF quantity of quasi-ization network interface card increases, effective bandwidth acquired in each VF is inversely proportional to the VF quantity configured in total, causes single
The performance of VF declines；(2) virtual machine of single I/O virtualization network interface card migrates limited online, thus when different main in group system
Load balancing cannot be reached when loading variation between machine, influence the overall performance of system.For this purpose, the invention proposes one kind to be based on
The single I/O of line migration virtualizes network optimized approach, the network load state of dynamic sensing virtual machine, by the number for limiting VF
Amount is to guarantee the performance of I/O intensity virtual machine, to guarantee the network performance of system entirety.And work as mainframe network overload
When being unable to satisfy the network demand of virtual machine, in such a way that single I/O virtualization and half virtualization combine, migration portion online
Virtual machine is divided to run to other more idle hosts, the network load between balancing cluster system, to optimize data center
Overall network performance.
As shown in Figure 1 to Figure 3, a kind of that optimization method is virtualized based on the single I/O migrated online, include the following steps:
Step S1, the network load situation of change of dynamic sensing virtual machine and acquire the network data information of virtual machine, will
The network data information is sent to network monitoring module；
Step S2, after network monitoring module receiving network data information, classified to virtual machine and distribute VF；
Step S3, network monitoring module is monitored the network state of virtual machine, if network monitoring module is to virtual
The information data that machine monitoring arrives is more than the threshold value of setting, then migrates the virtual machine of monitoring.
As shown in Fig. 2, step S1 includes:
Step S11, the flow information that virtual machine will acquire at the end of each sampling period is sent to network monitoring module；
Step S12, virtual machine reads its memory information currently in use, will acquire at the end of each sampling period in
It deposits information and is sent to host.
Specifically, the network data for acquiring virtual machine first is used for the network load situation of change of dynamic sensing virtual machine,
Virtual machine reads the network information that interface document/proc/net/dev that linux system provides obtains virtual machine, in each sampling
The flow information that will acquire when end cycle is sent to the network monitoring module of host；Virtual machine is by reading file/proc/ again
Meminfo obtains virtual machine memory size currently in use, and the data that will acquire at the end of each sampling period are sent to master
Machine.
Wherein, in step S2, virtual machine is categorized into network to network monitoring module by network data information based on the received
Intensive virtual machine and non-network intensive virtual machine, distribute VF to network-intensive virtual machine, and allow non-network intensive empty
Quasi- machine is worked using paravirtualized network interface card.
Further, in step S2, single I/O is virtualized into net using the activation backup mode of Bonding actuation techniques
Card be set as main equipment and paravirtualized Network card setup be from equipment, network monitoring module create two queues to virtual machine into
Row management, has just started for all virtual machines to be put into generic queue, by virtual machine according to this at the end of each sampling period
Flow and descending sequence choose a certain number of virtual machines and are put into Priority Queues, and remaining virtual machine is put into generic queue,
Then VF is distributed to the virtual machine in Priority Queues, and the virtual machine in generic queue uses paravirtualized network interface card work
Make；Specifically, after the network monitoring module of host receives the flow information that virtual machine is sent to, it is close that virtual machine is divided into network
Collection type virtual machine and non-network intensive virtual machine, distribute VF to network-intensive virtual machine, and allow non-network intensive virtual
Machine is worked using paravirtualized network interface card.In this process, if being originally the virtual machine of non-network intensity at this
Network-intensive virtual machine is classified as in sampling period, then it is virtual half virtualization network interface card of virtual machine to be switched to single I/O
The VF network interface card of change；If being originally the virtual machine of network-intensive and being classified as non-network intensity in this sampling period
Virtual machine, then depriving its VF makes it work using paravirtualized network interface card.Virtual machine is divided into network-intensive virtual machine and non-
The method of two class of network-intensive virtual machine is: being managed using two queues to virtual machine, what Priority Queues was put into is net
Network intensity virtual machine, what generic queue was put into is non-network-intensive virtual machine.Network monitoring module is according to this sampling week
The descending of virtual machine traffic is ranked up virtual machine in phase, chooses a certain number of virtual machines in front and is put into preferential team
Column, remaining virtual machine are put into generic queue.The method of single I/O virtualization network interface card and half virtualization network interface card switching at runtime is:
It is main equipment, paravirtualized net that single I/O, which is virtualized Network card setup, using the activation backup mode of Bonding actuation techniques
Card is arranged to from equipment, is directly realized by single I/O virtualization network interface card and half to the virtual hot plug of single I/O virtualization network interface card
Virtualize the switching at runtime of network interface card.
As shown in figure 3, step S3 includes:
Step S31, network monitoring module is monitored the network state of virtual machine, while the network for analyzing virtual machine is close
Collection degree；If network monitoring module is more than the threshold value of setting to the information data that virtual machine monitoring arrives, enter next step；
Step S32, predict the next cycle of the network state of the virtual machine information data whether be more than setting threshold
Value, if it exceeds the threshold value of setting, then enter next step；
Step S33, network monitoring module selectes the virtual machine for the threshold value that its information data is more than setting and by the virtual machine
It migrates to migration management module；
Step S34, the smallest host of network-intensive state in selection cluster is received the void of the migration by migration management module
Quasi- machine；
If step S35, the information data of the virtual machine of the migration is more than the threshold value of setting, repeatedly step S34.
Wherein, in step S31, information data is stored in file netinfo to network monitoring module and analysis simultaneously is virtual
The network-intensive degree of machine；The method that the network-intensive degree of analysis virtual machine uses is the network monitoring module from file
N sampled data in the past is read, if wherein there is k number according to the threshold value all beyond setting, enters step S33, otherwise constantly
Deposit data are repeated, data are read and analyze the process of data.
Further, in step S32, the method for predicting the information data of the next cycle of the network state of virtual machine is
It is predicted using the Self-regression Forecast Model based on time series.
Further, in step S33, network monitoring module selects the virtual machine that will be migrated on host, and reads
The network data information and memory information of virtual machine in file are selected based on the principle of per unit byte transmission maximum load is allowed
The IP address of the virtual machine is simultaneously sent to migration management module by the virtual machine that will be migrated.
Specifically, first, the network monitoring module of host is monitored the network state of virtual machine, works as virtual machine every time
Data on flows reach when, network monitoring module by data be stored in file netinfo；Network monitoring module was read from file
The n sampled data gone enters next step if wherein there is k number according to the threshold value all beyond setting, otherwise constantly repeats
Data are stored in, data are read and analyze the process of data；Second, the network monitoring module of host to virtual machine network state not
Carry out trend to be predicted, if the result of prediction next cycle also will exceed the threshold value of setting, triggers migration；Network monitoring
The selected virtual machine that will be migrated of principle of the module based on per unit byte transmission maximum load, by the IP address of the virtual machine and
Triggering migration signal is sent to the migration management module of migration control host, into next step；Wherein, the next period is predicted
The method of data is: being predicted using the Self-regression Forecast Model based on time series, the autoregressive prediction mode refers to
With the process for itself doing regression variable, using first order autoregressive model AR (1), time series x1, x2 ..., xp are considered, prediction the
(p+1) value in a sampling period estimates the value of (p+1) as follows:Wherein,
μ is the desired value of preceding p time series, and parameter phi is used for the variation of capture time sequence, and δ is that white Gaussian noise is decided by currently
The degree of fluctuation of sequence；The migration management module of third, migration control host selects the smallest master of network-intensive degree in cluster
Machine receives the virtual machine being migrated, and starts to migrate virtual machine, if what is migrated is the void of half virtualization either Full-virtualization
Quasi- machine, Direct Transfer；And if single I/O virtual equipment is used in selected virtual machine, it is inserted in conjunction with ACPI heat
Activation-backup mode of technology and Bonding actuation techniques is pulled out, the VF of the single I/O virtualization of switching at runtime is paravirtualized
Microsoft Loopback Adapter.If migrating network load or overweight after virtual machine, that is, migration signal is also triggered, then repeatedly one step of front
Suddenly.The method that virtual machine is migrated in selection host is: based on the principle of per unit byte transmission maximum load is allowed, according to following public affairs
Formula calculates virtual machine (vm) migration and determines the factor:
Wherein, Traffic_data indicates the size for sending and receiving data packet in a sampling period in total, and Num is indicated
Sampling number in total, mem are the sizes of virutal machine memory.The sequence row that the factor is successively decreased from big to small is determined according to virtual machine (vm) migration
Sequence selects maximum virtual machine to be migrated.
Wherein, the virtual machine of single I/O virtual equipment migrates that the specific method is as follows online:
(I) main equipment is set by single I/O virtual equipment using the activation backup mode of Bonding actuation techniques,
Paravirtualized Network card setup is from equipment；(II) when triggering migration, the migration management module of migration control host notifies source
The virtual thermal that the virtual hot plug controller of host makes it trigger the selected single I/O virtual equipment of virtual machine removes event；
(III) virtual machine closes the driving of single I/O virtual equipment, and stops using single I/O virtual equipment virtual to respond
Hot removal event, it is virtual that the movable network interface card based on Bonding actuation techniques is then automatically converted to half by single I/O virtualization network interface card
The network interface card of change；(IV) at this moment virtual machine is no longer hardware-related, can safety transfer to destination host；(V) after the completion of migration, mesh
Host virtual hot plug controller triggering the single I/O virtual equipment of virtual machine virtual thermal add event；(VI) virtual
Machine loads single I/O virtualization trawl performance and new virtual functions is selected to start to work.
As shown in figure 8, a kind of virtualize optimization system based on the single I/O migrated online, comprising:
Several virtual machines；
Network monitoring module, the network data letter sent for each virtual machine being monitored and being received each virtual machine
Breath；
Migration management module when for receiving migration trigger signal, reads the network data letter of each virtual machine in file
The network bandwidth utilization factor of total flow and the method calculating main frame divided by network interface card bandwidth that breath sends and receives according to host, will lead
The sequence sequence that the bandwidth availability ratio of machine successively decreases from big to small, selects wherein the smallest host, by virtual machine (vm) migration to the host.
Of the invention is a kind of based on the single I/O migrated online virtualization optimization system, further includes:
VF scheduler module, for according to the descending sequence of network flow in this sampling period, choosing virtual machine
The a certain number of virtual machines in front are put into Priority Queues, and remaining virtual machine is put into generic queue.
Wherein, network monitoring module is provided with the analytical unit of the network-intensive degree for analyzing virtual machine.
As shown in Figure 4 and Figure 5, detailed process is as follows by the VF of the single I/O virtualization network interface card of dynamic dispatching of the invention:
Step 1: all data in acquisition virtual machine, network data and memory information including virtual machine, this step 1
Include step performed below:
(1.1) network data for accurately collecting virtual machine first is used for the network load condition of dynamic sensing virtual machine, empty
Intend the network information that the machine-readable kernel interface file for taking Linux/proc/net/dev obtains virtual machine, by existing transmission and connects
Receive total flow subtract last time send and receive total flow be the total flow that sends and receives in this sampling period and, this is adopted
The total flow and be exactly the bandwidth availability ratio that this samples the virtual machine divided by network interface card bandwidth that the sample period sends and receives.Virtual machine
In network monitoring module collect the network data of virtual machine according to the method described above, network monitoring module is tied in each sampling period
By total flow that this sampling period sends and receives and the network monitoring module for being sent to host, the network monitoring of host when beam
The data of all virtual machines of module collection be exactly the total flow that sends and receives of the host and.
(1.2) virtual machine reads file/proc/meminfo and obtains virtual machine memory size currently in use, is adopting every time
The memory information of acquisition is sent to host when sample end cycle and is stored in file so as to subsequent processing, after triggering migration
For determining which virtual machine will be migrated.
Step 2: the VF resource of single I/O virtualization network interface card is scheduled
It the use of the activation backup mode of Bonding actuation techniques is first main equipment by single I/O virtualization Network card setup,
Paravirtualized network interface card is arranged to from equipment, and it is empty to be directly realized by single I/O to the virtual hot plug of single I/O virtualization network interface card
The switching at runtime that quasi-ization network interface card virtualizes network interface card with half；With aft engine network monitoring module create two queues to virtual machine into
Row management, has just started for all virtual machines to be put into generic queue, by virtual machine according to this at the end of each sampling period
Flow and descending sequence choose maximum a certain number of virtual machines and are put into Priority Queues, and remaining virtual machine is put into general
Then queue distributes VF to the virtual machine in Priority Queues, and the virtual machine in generic queue uses paravirtualized net
Card work.In this process, if being originally that the virtual machine of non-network intensity is classified as network in this sampling period
Half virtualization network interface card is then switched to the VF network interface card of single I/O virtualization by intensive virtual machine, if being originally network-intensive
Virtual machine be classified as non-network intensive virtual machine in this sampling period, then depriving its VF, to use it paravirtualized
Network interface card work.
Step 3: the network monitoring module of host carries out unified management and analysis to virtual machine network data, and analysis is empty
The network-intensive degree of quasi- machine, this step 3 include step performed below:
(3.1) when the network data of each virtual machine reaches, data are stored in text by the network monitoring module of host
Part netinfo, while analyzing the network-intensive degree of virtual machine；Wherein, the side that the network-intensive degree of virtual machine uses is analyzed
Method is: network monitoring module reads n sampled data in the past from file, i.e., the total flow that sends and receives of the host with such as
Fruit wherein has k number according to the threshold value all beyond setting, then enters step (3.2), otherwise constantly repeats deposit data, reads data
With the process of analysis data.
(3.2) future trend for predicting virtual machine network state, sets if the result of the next cycle of prediction also exceeds
Fixed threshold value, then triggering migrates into step 3, while network monitoring module is sent to migration control host for signal is migrated
Migration management module；Wherein, the method for predicting the data in next period is using the Self-regression Forecast Model based on time series
It is predicted, which is to be examined with the process for itself doing regression variable using first order autoregressive model AR (1)
Consider time series x1, x2 ..., xp, predict the value in (p+1) a sampling period, estimates the value of (p+1) as follows:
Wherein, μ is the desired value of preceding p time series, and parameter phi is used for the variation of capture time sequence, and δ is white Gaussian
Noise is decided by the degree of fluctuation of current sequence.
Step 4: after triggering migration, network monitoring module can select the virtual machine that will be migrated on the host, read text
The network and memory information of virtual machine in part, it is selected to will be migrated based on allowing the principle of per unit byte transmission maximum load
Virtual machine and the migration management module that the IP address of the virtual machine is sent to migration control host；Wherein, quilt in host is selected
The method of the virtual machine of migration is to calculate virtual machine according to following formula based on the principle of per unit byte transmission maximum load is allowed
Migration determines the factor:
Wherein, Traffic_data indicates the size for sending and receiving data packet in a sampling period in total, and Num is indicated
Sampling number in total, mem are the sizes of virutal machine memory.The sequence row that the factor is successively decreased from big to small is determined according to virtual machine (vm) migration
Sequence selects maximum virtual machine to be migrated.
Step 5: the smallest host of network-intensive state receives in the migration management module selection cluster of migration control host
The virtual machine that will be migrated, the method for choosing the smallest host of network-intensive state is: the total stream sent and received according to host
The network bandwidth utilization factor of amount and the method calculating main frame divided by network interface card bandwidth, successively decreases from big to small by the bandwidth availability ratio of host
Sequence sequence, selection wherein the smallest host.Start to migrate virtual machine, if network load still mistake after migration virtual machine
Weight, that is, also trigger migration signal, then repeatedly step 4.
Wherein, the online moving method of virtual machine of single I/O virtualization includes step performed below:
(5.1) using the activation backup mode of Bonding actuation techniques by single I/O virtual equipment setting based on set
Standby, paravirtualized network interface card is arranged to from equipment；Master is used when main equipment, that is, single I/O virtual equipment is active
Equipment work, only when main equipment be in unactivated state, just use works from equipment, that is, paravirtualized network interface card.
(5.2) when triggering migration, the virtual hot plug control of the migration management module notice source host of migration control host
The virtual thermal that device processed triggers the selected single I/O virtual equipment of virtual machine removes event.
(5.3) virtual machine closes the driving of the single I/O virtual equipment of virtual machine and stops using the single I/O of virtual machine empty
Quasi-ization equipment removes event to respond virtual thermal, and the activation network interface card of Bonding actuation techniques is automatic by single I/O virtualization network interface card
Be converted to paravirtualized network interface card.
(5.4) at this moment virtual machine is no longer hardware-related, can safety move to destination host.
(5.5) after the completion of migration, the single I/O virtualization of virtual hot plug controller triggering virtual machine of destination host is set
Standby virtual thermal adds event.
(5.6) virtual machine loads single I/O virtualization trawl performance and new virtual functions is selected to start to work.
As shown in Figure 4 and Figure 6, the host of VF scheduling includes n virtual machine and a privileged domain virtual machine, by privileged domain void
Quasi- machine is managed to the single I/O network interface card virtualized and the dispatching distribution of VF.Network monitoring module in privileged domain virtual machine
The network traffic information that each virtual machine is sent is received, VF scheduler module is by virtual machine according to the network flow in this sampling period
Descending sequence is measured, a certain number of virtual machines in front is chosen and is put into Priority Queues, remaining virtual machine is put into generic queue.
What preferential elder generation's queue was put into is network-intensive virtual machine, and what generic queue was put into is non-network-intensive virtual machine, initial
All virtual machines of state are all placed into generic queue, during each scheduling that the virtual machine insertion of network-intensive is excellent
First queue.VF scheduler module distributes VF to the virtual machine in Priority Queues, and the virtual machine in generic queue is virtualized using half
Network interface card work.In this scheduling process, a typical queue variation is as follows: if when scheduling Priority Queues certain
One virtual machine network flow is less than a certain virtual machine of generic queue, and VF scheduler module will move this virtual machine of Priority Queues
To generic queue and deprive its VF, and the maximum virtual machine of generic queue network flow is inserted into Priority Queues and distribute VF to
This virtual machine.
As shown in figure 5 and figure 7, network resource scheduling is made of n host and a migration control host.Internet resources tune
The host of degree system includes k virtual machine and a network monitoring module, and network monitoring module reads the kernel interface text of Linux
Part/proc/net/dev is used with the network information and file/proc/meminfo that obtain virtual machine to obtain virtual machine
Memory size, acquisition data deposit file in so as to subsequent processing.Network monitoring at the end of each sampling period
Module reads past n sampled data in file, i.e., the total flow that sends and receives of the host with if wherein there is k number
According to the threshold value all beyond setting, then the data for the next cycle predicted are touched if prediction result also exceeds the threshold value of setting
Deportation shifting signal.Network monitoring module will calculate virtual machine (vm) migration after triggering migrates signal and determine the factor, move according to virtual machine
The sequence sequence for determining that the factor is successively decreased from big to small is moved, maximum virtual machine is selected to be migrated.Shown migration control master
Include migration management module in machine, when migration management module receives migration trigger signal, reads the net of each host in file
Network information, the total flow sent and received according to host and the network bandwidth divided by the method calculating main frame of network interface card bandwidth utilize
Rate, the sequence that the bandwidth availability ratio of host is successively decreased from big to small sort, and select wherein the smallest host, virtual machine (vm) migration is arrived
The host.
In the present invention, the network based on the single I/O virtualization optimization method dynamic sensing virtual machine migrated online is negative
It carries, virtual machine is divided into network-intensive virtual machine and non-network intensive virtual machine, is distributed to the virtual machine of network-intensive
The VF of limited amount is to promote the network performance of network-intensive virtual machine, thus the network performance of lifting system entirety.This is excellent
Change method uses the network pressure for alleviating host to the method that virtual machine migrates online, in order to avoid mainframe network is overweight and occurs
The case where delay machine, also will not migrate virtual machine beyond its tolerance range in mainframe network load.Present invention use pair
The method of network resource threshold detection triggers migration, in order to eliminate unnecessary transient peak interference to ensure host in future
Lasting network load is had, as long as there are k in n sampled data beyond threshold value, is just judged as effectively triggering, simultaneously
The variation tendency of future network load is predicted to be preferably scheduled to Internet resources.
The present invention has the advantage that
1, method combines half virtualization and single I/O to virtualize respective advantage, can promote the network performance of virtual machine
It can guarantee the scalability of virtual machine, the network load state of dynamic sensing virtual machine, by limiting the quantity of VF to protect again
The performance for demonstrate,proving I/O intensity virtual machine, to guarantee the network performance of system entirety.
2, the optimization method migrates partial virtual when single I/O virtualization network interface card is unable to satisfy mainframe network performance requirement
Machine is run to other more idle hosts, the network load between balancing cluster system, to optimize the integral net of data center
Network performance.
3, the optimization method will not modify the scheduler program of monitor of virtual machine itself, only dispatch to monitor of virtual machine
The supplement of system can be applied to any virtual platform, such as Xen and KVM.
4, the activation backup mode of the optimization method combination ACPI hot plug technology and Bonding actuation techniques, dynamically cuts
The VF and paravirtualized virtual network interface card that change single I/O virtualization can solve the virtual machine of single I/O virtual equipment very well
The problem of online migration.
Mode the above is only the implementation of the present invention is not intended to limit the scope of the invention, all to utilize this
Equivalent structure or equivalent flow shift made by description of the invention and accompanying drawing content, it is relevant to be applied directly or indirectly in other
Technical field is included within the scope of the present invention.
Claims (10)
1. a kind of virtualize optimization method based on the single I/O migrated online, which comprises the steps of:
Step S1, the network load situation of change of dynamic sensing virtual machine and acquire the network data information of virtual machine, by the net
Network data information is sent to network monitoring module；
Step S2, after network monitoring module receiving network data information, classified to virtual machine and distribute VF；
Step S3, network monitoring module is monitored the network state of virtual machine, if network monitoring module supervises virtual machine
The information data controlled is more than the threshold value of setting, then migrates the virtual machine of monitoring.
2. according to claim 1 a kind of based on the single I/O migrated online virtualization optimization method, which is characterized in that
Step S1 includes:
Step S11, the flow information that virtual machine will acquire at the end of each sampling period is sent to network monitoring module；
Step S12, virtual machine reads its memory information currently in use, the memory letter that will acquire at the end of each sampling period
Breath is sent to host.
3. according to claim 1 a kind of based on the single I/O migrated online virtualization optimization method, which is characterized in that
In step S2, network monitoring module based on the received network data information by virtual machine be categorized into network-intensive virtual machine and
Non-network intensity virtual machine, distributes VF to network-intensive virtual machine, and makes non-network intensive virtual machine virtual using half
The network interface card of change works.
4. according to claim 3 a kind of based on the single I/O migrated online virtualization optimization method, which is characterized in that
In step S2, using Bonding actuation techniques activation backup mode by single I/O virtualization Network card setup be main equipment and
Paravirtualized Network card setup is from equipment, and network monitoring module creates two queues and is managed to virtual machine, just started by
All virtual machines are put into generic queue, by virtual machine according to this flow and descending row at the end of each sampling period
Sequence chooses a certain number of virtual machines and is put into Priority Queues, and remaining virtual machine is put into generic queue, then in Priority Queues
In virtual machine distribute VF, and the virtual machine in generic queue is worked using paravirtualized network interface card.
5. according to claim 1 a kind of based on the single I/O migrated online virtualization optimization method, which is characterized in that
Step S3 includes:
Step S31, network monitoring module is monitored the network state of virtual machine, while analyzing the network-intensive journey of virtual machine
Degree；If network monitoring module is more than the threshold value of setting to the information data that virtual machine monitoring arrives, enter next step；
Step S32, predict the next cycle of the network state of the virtual machine information data whether be more than setting threshold value, such as
Fruit is more than the threshold value of setting, then enters next step；
Step S33, network monitoring module selectes the virtual machine for the threshold value that its information data is more than setting and by the virtual machine (vm) migration
To migration management module；
Step S34, the smallest host of network-intensive state in selection cluster is received the virtual machine of the migration by migration management module；
If step S35, the information data of the virtual machine of the migration is more than the threshold value of setting, repeatedly step S34.
6. according to claim 5 a kind of based on the single I/O migrated online virtualization optimization method, which is characterized in that
In step S31, information data is stored in file netinfo and analyzes the network-intensive journey of virtual machine simultaneously by network monitoring module
Degree；The method that the network-intensive degree of analysis virtual machine uses reads the past n from file for the network monitoring module and adopts
Sample data, if wherein have k number according to all beyond setting threshold value, enter step S33, otherwise constantly repeat deposit data,
It reads data and analyzes the process of data.
7. according to claim 5 a kind of based on the single I/O migrated online virtualization optimization method, which is characterized in that
In step S32, the method for predicting the information data of the next cycle of the network state of virtual machine is using based on time sequence
The Self-regression Forecast Model of column is predicted.
8. according to claim 5 a kind of based on the single I/O migrated online virtualization optimization method, which is characterized in that
In step S33, network monitoring module selects the virtual machine that will be migrated on host, and reads the network of virtual machine in file
Data information and memory information, based on the principle of per unit byte transmission maximum load is allowed, the selected virtual machine that will be migrated is simultaneously
The IP address of the virtual machine is sent to migration management module.
9. a kind of virtualize optimization system based on the single I/O migrated online characterized by comprising
Several virtual machines；
Network monitoring module, the network data information sent for each virtual machine being monitored and being received each virtual machine；
Migration management module, when for receiving migration trigger signal, the network data information for reading each virtual machine in file is pressed
The network bandwidth utilization factor of total flow and the method calculating main frame divided by network interface card bandwidth that license-master's machine sends and receives, by host
The sequence sequence that bandwidth availability ratio successively decreases from big to small, selects wherein the smallest host, by virtual machine (vm) migration to the host.
10. according to claim 9 a kind of based on the single I/O migrated online virtualization optimization system, which is characterized in that
Further include:
VF scheduler module, for virtual machine according to the descending sequence of network flow in this sampling period, to be chosen front
A certain number of virtual machines are put into Priority Queues, and remaining virtual machine is put into generic queue.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811507379.XA CN109558216B (en) | 2018-12-11 | 2018-12-11 | Single root I/O virtualization optimization method and system based on online migration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811507379.XA CN109558216B (en) | 2018-12-11 | 2018-12-11 | Single root I/O virtualization optimization method and system based on online migration |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109558216A true CN109558216A (en) | 2019-04-02 |
| CN109558216B CN109558216B (en) | 2020-10-27 |
Family
ID=65869385
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811507379.XA Active CN109558216B (en) | 2018-12-11 | 2018-12-11 | Single root I/O virtualization optimization method and system based on online migration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109558216B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110113184A (en) * | 2019-04-17 | 2019-08-09 | 中国科学院深圳先进技术研究院 | KVM virtual machine network optimization method and device under SR-IOV environment |
| CN115484212A (en) * | 2022-07-26 | 2022-12-16 | 浪潮通信技术有限公司 | VF QoS adjustment method, device and electronic equipment |
| CN115576732A (en) * | 2022-11-15 | 2023-01-06 | 阿里云计算有限公司 | Root cause positioning method and system |
| WO2023124477A1 (en) * | 2021-12-31 | 2023-07-06 | 阿里巴巴（中国）有限公司 | Fault recovery method and apparatus for virtualization device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106250211A (en) * | 2016-08-05 | 2016-12-21 | 浪潮(北京)电子信息产业有限公司 | A kind of virtualization implementation method based on SR_IOV |
| CN107133093A (en) * | 2017-06-02 | 2017-09-05 | 郑州云海信息技术有限公司 | The online moving method and device of a kind of virtual machine |
| CN107562511A (en) * | 2016-06-30 | 2018-01-09 | 中兴通讯股份有限公司 | Moving method, physical server, virtual server and the system of physical to virtual |
-
2018
- 2018-12-11 CN CN201811507379.XA patent/CN109558216B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107562511A (en) * | 2016-06-30 | 2018-01-09 | 中兴通讯股份有限公司 | Moving method, physical server, virtual server and the system of physical to virtual |
| CN106250211A (en) * | 2016-08-05 | 2016-12-21 | 浪潮(北京)电子信息产业有限公司 | A kind of virtualization implementation method based on SR_IOV |
| CN107133093A (en) * | 2017-06-02 | 2017-09-05 | 郑州云海信息技术有限公司 | The online moving method and device of a kind of virtual machine |
Non-Patent Citations (1)
| Title |
|---|
| LINGFANG ZENG等: ""Raccoon: A Novel Network I/O Allocation Framework for Workload-Aware VM Scheduling in Virtual Environments"", 《IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110113184A (en) * | 2019-04-17 | 2019-08-09 | 中国科学院深圳先进技术研究院 | KVM virtual machine network optimization method and device under SR-IOV environment |
| CN110113184B (en) * | 2019-04-17 | 2021-10-26 | 杭州中科先进技术研究院有限公司 | KVM virtual machine network optimization method and device under SR-IOV environment |
| WO2023124477A1 (en) * | 2021-12-31 | 2023-07-06 | 阿里巴巴（中国）有限公司 | Fault recovery method and apparatus for virtualization device |
| CN115484212A (en) * | 2022-07-26 | 2022-12-16 | 浪潮通信技术有限公司 | VF QoS adjustment method, device and electronic equipment |
| CN115484212B (en) * | 2022-07-26 | 2023-10-20 | 浪潮通信技术有限公司 | QoS adjustment method and device for VF and electronic equipment |
| CN115576732A (en) * | 2022-11-15 | 2023-01-06 | 阿里云计算有限公司 | Root cause positioning method and system |
| CN115576732B (en) * | 2022-11-15 | 2023-03-10 | 阿里云计算有限公司 | Root cause positioning method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109558216B (en) | 2020-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109558216A (en) | It is a kind of that optimization method and its system are virtualized based on the single I/O migrated online | |
| EP3847549B1 (en) | Minimizing impact of migrating virtual services | |
| CN102279771B (en) | Method and system for adaptively allocating resources as required in virtualization environment | |
| US20200167258A1 (en) | Resource allocation based on applicable service level agreement | |
| CN101436966B (en) | Network monitoring and analysis system under virtual machine circumstance | |
| US9378032B2 (en) | Information processing method, information processing apparatus, recording medium, and system | |
| Mei et al. | Performance measurements and analysis of network i/o applications in virtualized cloud | |
| US8510747B2 (en) | Method and device for implementing load balance of data center resources | |
| CN101465863B (en) | Method for implementing high-efficiency network I/O in kernel virtual machine circumstance | |
| CN103810016B (en) | Realize method, device and the group system of virtual machine (vm) migration | |
| CN104102543B (en) | The method and apparatus of adjustment of load in a kind of cloud computing environment | |
| CN107992362B (en) | Method, device and system for automatic performance test | |
| WO2014090070A1 (en) | Method and system for managing servers | |
| CN103605574A (en) | Virtual machine resource scheduling method and system for server clusters | |
| CN109343931B (en) | Load balancing oriented application perception virtual machine scheduling method in IaaS environment | |
| CN104077212A (en) | Pressure test system and method | |
| CN108920153A (en) | A kind of Docker container dynamic dispatching method based on load estimation | |
| CN105302641B (en) | The method and device of node scheduling is carried out in virtual cluster | |
| WO2023109068A1 (en) | Automatic virtual machine migration decision-making method based on user experience in multi-cloud environment | |
| CN107370783B (en) | Scheduling method and device for cloud computing cluster resources | |
| CN104866370A (en) | Dynamic time slice dispatching method and system for parallel application under cloud computing environment | |
| CN114356714A (en) | Resource integration monitoring and scheduling device based on Kubernetes intelligent board card cluster | |
| CN109739634A (en) | A kind of atomic task execution method and device | |
| Quaresma et al. | Controlling garbage collection and request admission to improve performance of faas applications | |
| CN110865871A (en) | Resource rationalization application-based virtualized cluster resource scheduling method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |