CN109144664B - Dynamic migration method of virtual machine based on user service quality demand difference - Google Patents

Dynamic migration method of virtual machine based on user service quality demand difference Download PDF

Info

Publication number
CN109144664B
CN109144664B CN201810836808.1A CN201810836808A CN109144664B CN 109144664 B CN109144664 B CN 109144664B CN 201810836808 A CN201810836808 A CN 201810836808A CN 109144664 B CN109144664 B CN 109144664B
Authority
CN
China
Prior art keywords
virtual machine
host
migration
current
physical host
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810836808.1A
Other languages
Chinese (zh)
Other versions
CN109144664A (en
Inventor
孙红光
曹昊
盛敏
史琰
李建东
张琰
文娟
刘俊宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN201810836808.1A priority Critical patent/CN109144664B/en
Publication of CN109144664A publication Critical patent/CN109144664A/en
Application granted granted Critical
Publication of CN109144664B publication Critical patent/CN109144664B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • 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 invention belongs to the technical field of devices for executing special programs, and discloses a dynamic migration method of a virtual machine based on user service quality demand difference, which changes the distribution in a physical host according to the difference of QoS demands requested by different users in a system; determining an adaptive virtual machine overload migration threshold together according to the load condition of the current system; and finding out the physical host with the current too low load by using a light-load host detection algorithm, and migrating and merging the virtual machine instances. The invention can optimize the overall energy consumption of the cloud data center on the basis of guaranteeing the QoS requirements of users and obtain better performance than the existing algorithm. The invention achieves the purposes of improving the CPU utilization rate of the physical hosts by changing the difference of the QoS requirements of the requests in each physical host, thereby reducing the number of the physical hosts required by the bearing requests and reducing the energy consumption of the data center.

Description

Dynamic migration method of virtual machine based on user service quality demand difference
Technical Field
The invention belongs to the technical field of devices for executing special programs, and particularly relates to a dynamic migration method of a virtual machine based on user service quality demand difference.
Background
Currently, the current state of the art commonly used in the industry is such that: with the rapid development of the internet and the rapid growth of user data, cloud computing is widely used for real-time processing of large-scale data. The cloud data center can allocate virtualized resources to the users according to the needs of the users as required so as to provide services. Although the advent of cloud computing can effectively meet users' large-scale computing needs, it also results in significant energy consumption. Due to the adoption of the virtualization technology, a plurality of virtual machines can be operated on one physical machine, and the application program can be directly operated on the virtual machines, so that the number of physical machines of the cloud data center is reduced, and the energy consumption is reduced. The existing energy consumption-based virtual machine scheduling algorithm mainly comprises two stages: static placement and live migration. Static placement is typically used for initial virtual machine placement, and once a virtual machine is assigned to a corresponding host, this correspondence will not change. The dynamic migration can realize the real-time migration of the virtual machines among the physical hosts, so that the virtual machines are concentrated in a part of hosts and the rest idle hosts are closed, and the purpose of reducing energy consumption is achieved. However, if the virtual machine dynamically migrates too aggressively, some part of the physical hosts may be overloaded, resulting in difficulty in meeting the quality of service (QoS) requirements of the users. Currently, in the research on the dynamic migration of the virtual machine, the influence of the difference of the QoS requirements of the users on the allocation of the server resources is not considered. The QoS requirements of the user are reflected in the CPU utilization of the physical host carrying its request, and if the CPU utilization of the current physical host exceeds a specified percentage, it is considered that the QoS requirements of the user cannot be met. Since a physical host often carries multiple virtual machine instances, each virtual machine instance corresponds to a user request. Because the QoS requirements of each user are different, the CPU utilization of the physical host will be limited to the most QoS-critical virtual machine instance in the set of virtual machines that it carries. Therefore, the CPU utilization rate of the physical server is lower, more physical hosts are needed to provide services for the same number of requests, and the overall energy consumption of the system is improved.
In summary, the problems of the prior art are as follows: at present, the influence of the difference of the QoS requirements of users on the allocation of server resources is not considered in the dynamic migration of virtual machines, so that the server resources are wasted to a certain extent. The CPU utilization of each server will be limited by the strictest QoS requirements among all user requests in the host, so when the QoS difference of the internal request of one host is too large, the CPU of the host is severely limited, resulting in that resources cannot be fully utilized, and thus more physical hosts are required to carry requests of the same scale, which also results in the increase of the overall energy consumption of the system.
The difficulty and significance for solving the technical problems are as follows: the problem of adjusting the difference in user QoS requirements within each physical host in the system is faced: when the system starts to serve, the number of internal user requests is small, the difference of the user QoS inside the host is unstable along with the arrival of the user requests, and because the system load is low, the influence of optimizing the difference of the QoS on energy consumption is small, and even the performance may be deteriorated. When the load of the system is close to saturation, because the remaining resources of each host in the system are less and additional resources are needed in the dynamic migration process of the virtual machine, there may not be enough resources for the algorithm to perform the dynamic migration process of the virtual machine. Therefore, the invention considers the difference of the QoS requirement of the user, and normalizes the current load condition of the system by sensing the load condition of the current system and recording the historical load data of the system. And determining a dynamic migration threshold of the virtual machine by using the ratio of the variance of all requested QoS requirements in the system to the normalized current load condition, thereby dynamically adjusting the difference of the QoS requirements requested in each host according to the load condition of the system. And when the system load is lower, the overload migration threshold is relaxed, and the threshold is gradually tightened along with the continuous rise of the system load, so that the difference of QoS (quality of service) requirements in each physical host is reduced, the QoS requirements of users in each physical host are close to each other as much as possible, the utilization rate of server resources is improved, the number of required hosts is reduced, and finally, the idle physical hosts are closed to achieve the purpose of reducing the overall energy consumption of the system.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a dynamic migration method of a virtual machine based on the difference of user service quality requirements.
The invention is realized in such a way that a virtual machine dynamic migration method based on user service quality demand difference changes the distribution in a physical host according to the difference of QoS demands requested by different users in a system; determining an adaptive virtual machine overload migration threshold together according to the load condition of the current system; and finding out the physical host with the current too low load by using a light-load host detection algorithm, and migrating and merging the virtual machine instances.
Further, the method for dynamically migrating the virtual machine based on the difference of the user service quality demands comprises the following steps:
step one, when a request comes, performing initial host allocation for the request; for each incoming request vm, selecting a proper physical host from a physical host list of the current system to allocate to the request vm;
step two, executing an overload host detection algorithm according to time intervals, and judging whether each physical host is overloaded or not through a self-adaptive overload migration threshold;
step three, the execution time of the overload host detection is the same, and for each physical host, whether the virtual machines in the physical host can be migrated and merged through dynamic migration of the virtual machines is judged through a light load host detection algorithm;
and step four, after the overload host and the light load set in the system are obtained, executing a virtual machine selection algorithm. For a lightly loaded host, virtual machine selection need not be performed;
step five, after the virtual machine set to be migrated is obtained, executing the migration algorithm of the virtual machine of the overload host;
and step six, the execution time of the overload host virtual machine migration is the same, and after the virtual machine set to be migrated is obtained, the light-load host virtual machine migration algorithm is executed.
Further, the first step specifically includes:
(1) acquiring all PM sets PMS in the system;
(2) for each physical host pm in the set PMS, judging whether the current residual CPU and bandwidth resources meet the requirements of vm, and if not, removing the physical host pm from the set PMS;
(3) predicting whether pm can violate the requested QoS requirement on the assumption that the current vm is loaded, and if pm can violate the QoS requirement of vm, removing pm from the PMS set;
(4) and allocating pm with the highest CPU utilization rate in the PMS to vm.
Further, if the host computer in the second step is overloaded, the virtual machine migration is performed, and the calculation process of the overload migration threshold includes:
(1) recording the historical minimum load miLoad of the system in the running process of the system;
(2) acquiring a current load currload of a system;
(3) calculating the variance sysVar of all requested QoS requirements in the current system;
(4) calculating the variance hostVar of all requested QoS requirements in the current physical host;
(5) if hostVar > sysVar x (mload/currload), the current physical host is considered to be overloaded, and the virtual machine in the host is migrated; otherwise the physical host is considered not overloaded.
Further, the calculation process of the light load detection threshold in the third step specifically includes:
(1) for each virtual machine instance in the current physical host, calculating the percentage vmCpu of the CPU resource occupied by the virtual machine instance in the current PM and acquiring the maximum value vmCpu in the percentage vmCpumax；
(2) Acquiring the CPU utilization rate hostCpu of the current host;
(3) if hostCpu-vmCpumaxAnd if the load is less than eta, the current PM is considered to be a light load host, and dynamic migration of the virtual machine is carried out. Wherein eta is the CPU power consumption model parameter.
Further, the fourth step specifically includes:
(1) obtaining a virtual machine set VMS in a physical hostiCalculating the current VMSiThe variance of the QoS requirement of (preVar);
(2) for VMSiPer virtual machine instance vm, compute assumptions slave VMSiIn-after vm removal VMSiThe variance of the QoS requirement of (postVar);
(3) get each get per virtual machine, get VMS after removing itiThe decrease in variance of the QoS requirements of (d) decVar (preVar-postVar);
(4) mixing VMSiAll the virtual machines in (1) are sorted in descending order according to the value of decVar;
(5) in turn from the VMSiRemove the VM and add it to the to-be-migrated list until the current physical host is no longer overloaded.
Further, the fifth step specifically includes:
(1) acquiring VMS (virtual machine set to be migrated) from overloaded hostmigAnd sorting in ascending order according to QoS requirements;
(2) for each one belonging to the VMSmigTraversing a PM set PMS in the system, and for each physical host PM, calculating a variance preVar of the QoS requirement of the virtual machine set in the current PM and a variance postVar of the QoS requirement of the virtual machine set in the PM after the vm is supposed to be migrated to the PM;
(3) calculating decVar ═ postVar/preVar, and filtering out all PMs with decVar being more than or equal to 1; acquiring the CPU utilization rate cpuNext of pm after vm is supposed to migrate into pm; calculating temp ═ decVar/cpuNext;
(4) recording the minimum value of temp and the corresponding pm, and if the pm exists, migrating the current virtual machine vm to the pm; otherwise, removing vm from the list to be migrated, and not migrating.
Further, the sixth step specifically includes:
(1) obtaining from lightVirtual machine set VMS to be migrated of load hostmigSorting the VMs in a descending order according to the requirements of the CPU resources;
(2) for VMSmigEach virtual machine instance vm iniThe source host is pmjTraversing the set of physical hosts, for each physical host pm thereinkCalculate the sum vmiSlave pmjMigration to pmkReduction of post-system energy consumption:
Esave(vmi,pmj,pmk,t)＝Edif(vmi,pmj,pmk,t)-Emigr(vmi,pmj,pmk,t)；
wherein Edif(vmi,pmj,pmkAnd t) denotes time t due to vmiAre respectively at pmjAnd pmkThe difference in energy consumption caused, and Emigr(vmi,pmj,pmkAnd t) represents vm at time tiSlave pmjMigration to pmkThe energy consumption brought is respectively expressed as:
Edif(vmi,pmj,pmk,t)＝
(tre(vmi)-tmigr(vmi,pmj,pmk,t))×(Pcpudec(pmj,vmi,t)-Pcpuinc(pmk,vmi,t))；
Emigr(vmi,pmj,pmk,t)＝
tmigr(vmi,pmj,pmk,t)×(Pcpuinc(pmk,vmi,t)+Pnet(pmj,t)+Pnet(pmk,t))；
wherein t isre(vmi) Is vmiRemaining service time of Pcpudec(pmj,vmiAnd t) represents that vmiSlave pmjPost removal pmjReduction of CPU power consumption, Pcpuinc(pmk,vmiAnd t) represents that vmiMigration into pmkRear pmkIncrease in CPU Power consumption, Pnet(pmjT) and Pnet(pmkT) each represents pmjAnd pmkIncreased power consumption of network communication due to migration, tmigr(vmi,pmj,pmkT) is time vmiSlave pmjMigration to pmkThe migration time required, is expressed as:
Figure BDA0001744655550000061
wherein the RAM (vm)i) Represents vmiMemory size, BW ofre(pmjT) represents time pmjAvailable bandwidth of;
(3) record Esave(vmi,pmj,pmkT) maximum value of and compares the current vmiMigrate to corresponding pmkIn (1).
The invention also aims to provide a virtual machine scheduling system applying the dynamic virtual machine migration method based on the user service quality demand difference.
The invention also aims to provide the virtual machine applying the dynamic migration method of the virtual machine based on the user service quality demand difference.
In summary, the advantages and positive effects of the invention are: the dynamic migration method of the virtual machine, which is adopted by the invention, can optimize the overall energy consumption of the cloud data center on the basis of guaranteeing the QoS (quality of service) requirements of users and obtain better performance than the existing algorithm. The dynamic migration method of the virtual machine, which is adopted by the invention, can dynamically adjust the dynamic migration threshold of the virtual machine by sensing the current system load of the cloud data center and the difference of the QoS (quality of service) requests of users in the current data center and combining the historical data of the system, and can improve the CPU (Central processing Unit) utilization rate of the physical hosts by changing the difference of the QoS (quality of service) requirements of the requests in each physical host, thereby reducing the number of the physical machines required by the bearing requests and reducing the energy consumption of the data center. As can be seen from the simulation results of fig. 4 to fig. 6, the virtual machine dynamic migration method adopted in the present invention can effectively reduce the overall energy consumption of the data center network, wherein, in a scenario where the request arrival rate λ is 8, the overall energy consumption is reduced by 12.88% compared with the existing power consumption-aware greedy migration algorithm.
Drawings
Fig. 1 is a flowchart of a method for dynamically migrating a virtual machine based on a difference in user qos requirements according to an embodiment of the present invention.
Fig. 2 is a diagram of a scenario for allocating and scheduling virtual machines in a data center according to an embodiment of the present invention.
Fig. 3 is a flowchart of an implementation of a method for dynamically migrating a virtual machine based on a difference in user qos requirements according to an embodiment of the present invention.
FIG. 4 is a schematic diagram of the system energy consumption index of the present invention compared with other algorithms over simulation time according to an embodiment of the present invention.
Fig. 5 is a schematic diagram illustrating a comparison between the system energy consumption index of the present invention and several other algorithms when the λ parameter of the request poisson arrival process is increased according to the embodiment of the present invention.
FIG. 6 is a schematic diagram of the energy consumption index of the system of the present invention compared with other algorithms when the amount of PM provided by the embodiment of the present invention is increased.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention aims to solve the problem that the influence of the difference of QoS (quality of service) requirements of users on server resource allocation is not considered in the dynamic migration of the virtual machines at present. Compared with the prior art, the self-adaptive virtual machine dynamic migration method based on the user QoS requirement and the system load perception can obviously reduce the system energy consumption on the premise of guaranteeing the user QoS.
The following detailed description of the principles of the invention is provided in connection with the accompanying drawings.
As shown in fig. 1, the method for dynamically migrating a virtual machine based on a difference in user quality of service requirements according to an embodiment of the present invention includes the following steps:
s101: initial host allocation: when a request comes, performing initial virtual machine allocation; for each incoming request, selecting a proper physical host from a physical host list of the current system to distribute to the request;
s102: and (3) overload host detection: executing according to a certain time interval, and judging whether each physical host is overloaded or not through a self-adaptive overload migration threshold;
s103: and (3) detecting a light-load host: the execution time of the overload host detection is the same, and for each physical host, whether the virtual machines in the physical host can be migrated and merged through the dynamic migration of the virtual machines is judged through a light load host detection algorithm, so that the energy consumption is reduced;
s104: calculating a reduction value of QoS variance requested in the host after migration of each virtual machine instance in the host to serve as a standard for virtual machine selection, sorting the virtual machines in a descending order according to the index, and sequentially migrating the virtual machines until the host is not overloaded any more;
s105: and (3) migration of the overload host virtual machine: for the virtual machine to be migrated from the overloaded host, other hosts in the system are traversed, the reduction amount of the QoS variance requested by the hosts after the virtual machine instance is migrated to each host is calculated, and the host with the most reduced variance is selected as a new host of the virtual machine instance;
s106: light load host virtual machine migration: when the virtual machine from the light-load host is migrated, the difference between the energy consumption which will be brought by the virtual machine instance when the virtual machine runs on different physical hosts and the energy consumption which will be brought by the virtual machine instance when the virtual machine runs on the current host is calculated by acquiring the residual service time of the virtual machine, the energy consumption of migration itself is calculated by combining a virtual machine dynamic migration energy consumption model, finally, the reduction of system energy consumption which can be brought by the current virtual machine instance after migration for each physical host is obtained, and the reduction is used as a greedy strategy to allocate a new host.
The application of the principles of the present invention will now be described in further detail with reference to the accompanying drawings.
A CPU: a Central Processing Unit,
QoS: quality of service,
VM: virtual Machine,
PM: physical Machine: physical host
Referring to fig. 1, different users in the simulation scenario of the present invention may correspond to one or more user requests, the total number of the user requests is N, after reaching the global resource controller, the data center schedules virtualized resources for the user requests and creates VM instances, and finally the VM instances are allocated to a certain PM in a physical host set inside the system. When a running VM instance exists in a PM, the instance is considered to be in a working state, otherwise, the PM can be shut down to save energy consumption.
Referring to fig. 2, the method for dynamically migrating a virtual machine of the present invention includes the following steps:
step 1: when a request comes, an initial host allocation is made for the request. For each incoming request vm, the appropriate physical host is selected from the list of physical hosts of the current system to be allocated to the request.
1. And acquiring all PM sets PMS in the system.
2. And for each physical host pm in the set PMS, judging whether the current residual CPU and bandwidth resources meet the requirements of vm, and if not, removing the physical host pm from the set PMS.
3. The pm is predicted to remove pm from the set PMS assuming that the current vm carrying the bearer would violate the requested QoS requirements, and if so, the pm.
4. And allocating pm with the highest CPU utilization rate in the PMS to vm.
Step 2: and executing an overload host detection algorithm according to a certain time interval, and judging whether each physical host is overloaded or not through a self-adaptive overload migration threshold. And if the host is overloaded, carrying out virtual machine migration. The calculation process of the overload migration threshold is as follows:
1. recording the historical minimum load mload of the system during the operation of the system
2. Obtaining system current load currload
3. Calculate the variance of all requested QoS requirements in the current system, sysVar
4. Calculate variance hostVar of all requested QoS requirements in current physical host
5. If hostVar > sysVar x (mload/currload), the virtual machine therein may be migrated assuming that the current physical host is overloaded. Otherwise the physical host is considered not overloaded.
And step 3: the execution time of the overload host detection is the same, and for each physical host, whether the virtual machines in the physical host can be migrated and merged through the dynamic migration of the virtual machines is judged through a light load host detection algorithm, so that the energy consumption is reduced. The calculation process of the light load detection threshold is as follows:
1. for each virtual machine instance in the current physical host, calculating the percentage vmCpu of the CPU resource occupied by the virtual machine instance in the current PM and acquiring the maximum value vmCpu in the percentage vmCpumax
2. Obtaining CPU utilization rate hostCpu of current host
3. If hostCpu-vmCpumaxAnd < eta, considering that the current PM is a light load host machine, and performing dynamic migration on the virtual machine. Wherein eta is the CPU power consumption model parameter.
And 4, step 4: after acquiring the overloaded host and the light load set in the system, executing a virtual machine selection algorithm. For lightly loaded hosts, virtual machine selection need not be performed. For each overloaded host, the following virtual machine selection algorithm is performed:
1. obtaining a virtual machine set VMS in a physical hostiCalculating the current VMSiVariance of QoS requirement of (1) preVar
2. For VMSiPer virtual machine instance vm, compute assumptions slave VMSiIn-after vm removal VMSiVariance of QoS requirement of (postVar)
3. Get each get per virtual machine, get VMS after removing itiThe decrease in variance of the QoS requirement of (1) decVar ═ preVar-postVar
4. Mixing VMSiAll virtual machines in (1) are sorted in descending order of the value of decVar
5. In turn from the VMSiRemove the VM and add it to the to-be-migrated list until the current physical host is no longer overloaded.
And 5: after the virtual machine set to be migrated is obtained, executing an overload host virtual machine migration algorithm: for the virtual machine to be migrated from the overloaded host, because the purpose of reducing the energy consumption can not be achieved directly through dynamic migration, the invention improves the distribution of the QoS requirements requested in the system through migration to improve the CPU utilization rate of the host to achieve the purpose of reducing the energy consumption of the system indirectly by using less PM to carry the same request quantity.
1. Acquiring VMS (virtual machine set to be migrated) from overloaded hostmigAnd sorted in ascending order according to QoS requirements.
2. For each one belonging to the VMSmigAnd traversing a PM set PMS in the system, and for each physical host PM, calculating the variance preVar of the QoS requirement of the virtual machine set in the current PM and the variance postVar of the QoS requirement of the virtual machine set in the PM after the vm is supposed to be migrated into the PM.
3. DecVar is calculated as postVar/preVar, and all PMs with decVar ≧ 1 are filtered out. And acquiring the CPU utilization rate cpuNext of pm after vm is supposed to migrate into pm. Calculating temp ═ decVar/cpuNext
4. And recording the minimum value of temp and the corresponding pm, and migrating the current virtual machine vm to the pm if the pm exists. Otherwise, the vm is removed from the list to be migrated, and the vm is not migrated.
Step 6: the execution time of the overload host virtual machine migration is the same as that of the overload host virtual machine migration, and after the virtual machine set to be migrated is obtained, a light-load host virtual machine migration algorithm is executed: since the power consumption of the CPU increases at a higher rate with load at low load than at high load, it should be avoided to operate the physical host at low load as much as possible. The virtual machine instances in the part of hosts can be migrated and merged to other hosts in the running process through the migration of the virtual machines of the light-load hosts, so that the energy consumption of the system is reduced.
1. Acquiring VMS (virtual machine set) to be migrated from light-load hostmigAnd sorting the VMs in descending order according to the requirement of the CPU resource.
2. For VMSmigEach virtual machine instance vm iniThe source host is pmjTraversing the set of physical hosts, for each physical host pm thereinkCalculate the sum vmiSlave pmjMigration to pmkReduction of post-system energy consumption:
Esave(vmi,pmj,pmk,t)＝Edif(vmi,pmj,pmk,t)-Emigr(vmi,pmj,pmk,t)
wherein Edif(vmi,pmj,pmkAnd t) denotes time t due to vmiAre respectively at pmjAnd pmkThe difference in energy consumption caused, and Emigr(vmi,pmj,pmkAnd t) represents vm at time tiSlave pmjMigration to pmkThe energy consumption brought by the method can be respectively expressed as:
Edif(vmi,pmj,pmk,t)＝
(tre(vmi)-tmigr(vmi,pmj,pmk,t))×(Pcpudec(pmj,vmi,t)-Pcpuinc(pmk,vmi,t))
Emigr(vmi,pmj,pmk,t)＝
tmigr(vmi,pmj,pmk,t)×(Pcpuinc(pmk,vmi,t)+Pnet(pmj,t)+Pnet(pmk,t))
wherein t isre(vmi) Is vmiRemaining service time of Pcpudec(pmj,vmiAnd t) represents that vmiSlave pmjPost removal pmjReduction of CPU power consumptionAmount, Pcpuinc(pmk,vmiAnd t) represents that vmiMigration into pmkRear pmkIncrease in CPU Power consumption, Pnet(pmjT) and Pnet(pmkT) each represents pmjAnd pmkIncreased power consumption of network communication due to migration, tmigr(vmi,pmj,pmkT) is time vmiSlave pmjMigration to pmkThe required migration time can be expressed as:
Figure BDA0001744655550000111
wherein the RAM (vm)i) Represents vmiMemory size, BW ofre(pmjT) represents time pmjThe available bandwidth of (a).
3. Record Esave(vmi,pmj,pmkT) maximum value of and compares the current vmiMigrate to corresponding pmkIn (1).
The application effect of the present invention will be described in detail with reference to the simulation.
1. Simulation scenario
Simulation scenarios of the invention as shown in fig. 1, consider the scenario of a single data center. In the simulation, in order to ensure the credibility of the simulation result, the simulation method carries out 100 times of simulation on each scene and averages the result, and because of the problem of simulation time, the simulation method carries out equal proportion change on parameters influencing the simulation time. The number of user requests is set to be 1000, the arrival of the user requests obeys a Poisson process with a parameter of lambda, the QoS requirements of the users are uniformly and randomly distributed between 0.5 and 1, and the number of instructions required to be executed by the requests obeys uniform and random distribution between 20000 and 320000 MI. The CPU occupancy and bandwidth utilization of the VMs are subject to a uniform random distribution between 2.5% and 12.5%. The CPU maximum processing capacity of PM is 64000MIPS, and the bandwidth is 30 MB/s.
2. Simulation content and results
In simulation, the invention respectively compares four algorithms, and the self-adaptive virtual machine dynamic migration method provided by the invention is called as a self-adaptive migration algorithm; the energy consumption perception greedy migration is similar to the energy consumption perception greedy migration, and only when a virtual machine in a light-load host machine is migrated, the prediction of future energy consumption is used as a target of a greedy algorithm instead of the power consumption of the host machine in the current state; the greedy algorithm which does not consider the dynamic migration process of the virtual machine and takes the power consumption of the host as an optimization target is called as the greedy algorithm without migration.
2a) Keeping the number of the PMs in the system and the parameter lambda of the request arrival process unchanged, setting the number of the PMs as 100 and lambda as 8, observing the change trend of the overall energy consumption of the system along with time, and calculating the energy consumption index corresponding to the system.
2b) Keeping the number of the PMs in the system unchanged, setting the number of the PMs to be 100, increasing an arrival process parameter lambda of the request, observing the variation trend of the energy consumption of the system along with the increase of the lambda, and calculating an index corresponding to the energy consumption of the system.
2c) And keeping the arrival process parameter lambda of the request unchanged, setting lambda as 10, changing the number of the PMs in the data center network, observing the variation trend of the system energy consumption along with the increase of the number of the PMs, and calculating the index of the corresponding system energy consumption.
3. Analysis of simulation results
3a) As can be seen from fig. 3, the dynamic migration process of the virtual machine is not considered in the greedy algorithm without migration, and the initial allocation is optimized and cannot be adjusted according to the change of the system state, so that the energy consumption is the highest among the four algorithms. Compared with the power consumption sensing greedy migration algorithm which only considers the power consumption condition of the current physical host, the energy consumption sensing greedy migration algorithm can select a more appropriate migration target for the virtual machine to be migrated in the light-load host, so that the system energy consumption of the energy consumption sensing greedy migration algorithm is slightly lower than that of the power consumption sensing greedy migration algorithm, but is not obvious. The self-adaptive migration algorithm provided by the invention is based on the idea of improving the utilization rate of the CPU of the host by changing the difference of the user QoS requirements in the system on the basis of the energy consumption perception greedy migration, and compared with other algorithms, the self-adaptive migration algorithm can bear the requests of the same scale while using fewer physical hosts, thereby reducing the overall energy consumption of the system. According to the simulation data of the invention, compared with the existing power consumption perception greedy algorithm, the dynamic migration method of the self-adaptive virtual machine based on the user QoS demand and the system load perception provided by the invention under the scene has the advantage that the system energy consumption is reduced by 12.88%.
3b) As can be seen from fig. 4, keeping the total number of requests at 1000 and the number of PMs in the system at 100, the energy consumption of each algorithm tends to increase first and then decrease as the parameter λ of the arrival of the request poisson increases. This is because, in the first half of the curve, the larger λ, the more frequently requests come, so at the same time, the larger the number of virtual machine instances in the system, and thus the more physical hosts are needed to carry the requests, so the power consumption increases with the increase of λ. Whereas in the second half of the curve, a certain critical point is reached due to lambda. The available physical host capacity in the system cannot bear all requests in the corresponding scene, so a part of the requests can be rejected, and the energy consumption is reduced on the curve. This is also shown in figure 3. In addition, the invention can see that the algorithm energy consumption index of the invention is better than other algorithms when λ is less than or equal to 10, and when λ is between 9 and 10, the energy consumption increase rate of the other three algorithms is obviously reduced, which indicates that for the three algorithms, the system is already in a saturated state when λ is 9, and the algorithm of the invention can still accommodate more requests, which indicates that for the same number of requests, the algorithm of the invention can use fewer physical hosts for carrying, thereby reducing the energy consumption. Whereas when λ > 10, a part of the requests is rejected because the system is already saturated, whereas the algorithm of the invention can accommodate more requests and therefore the energy consumption is higher instead. However, in a real data center, there is almost no system saturation.
3c) As can be seen from fig. 5, keeping λ equal to 10, the energy consumption of each algorithm tends to increase first and then to be smooth as the number of PMs in the system increases. It can be seen that when the number of PMs is less than 100, the performance of the "power consumption aware greedy migration" and "power consumption aware greedy migration" algorithms is instead due to the algorithm of the present invention, as mentioned above, because when λ is 10 and the number of PMs is 100, the system cannot carry all requests, and the accommodating capability of the algorithm of the present invention is due to other algorithms, which also brings higher power consumption. When the number of the PMs reaches 120, since the capacity of the system can accommodate all the requests, the energy consumption of each algorithm also tends to be smooth, and it can be seen that the energy consumption of the algorithm provided by the invention is obviously superior to that of other algorithms.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (8)

1. A dynamic migration method of virtual machines based on user service quality demand difference is characterized in that the dynamic migration method of virtual machines based on user service quality demand difference changes the distribution in a physical host according to the difference of QoS demands requested by different users in a system; determining an adaptive virtual machine overload migration threshold together according to the load condition of the current system; finding out a physical host with over-low current load by using a light-load host detection algorithm, and migrating and merging the virtual machine instances;
the dynamic virtual machine migration method based on the user service quality demand difference comprises the following steps:
step one, when a request comes, performing initial host allocation for the request; for each incoming request vm, selecting a proper physical host from a physical host list of the current system to allocate to the request vm;
step two, executing an overload host detection algorithm according to time intervals, and judging whether each physical host is overloaded or not through a self-adaptive overload migration threshold; if the host computer is overloaded, carrying out virtual machine migration, wherein the calculation process of the overload migration threshold comprises the following steps:
(1) recording the historical minimum load miLoad of the system in the running process of the system;
(2) acquiring a current load currload of a system;
(3) calculating the variance sysVar of all requested QoS requirements in the current system;
(4) calculating the variance hostVar of all requested QoS requirements in the current physical host;
(5) if hostVar > sysVar x (mload/currload), the current physical host is considered to be overloaded, and the virtual machine in the host is migrated; otherwise, the physical host is not overloaded;
step three, the execution time of the overload host detection is the same, and for each physical host, whether the virtual machines in the physical host can be migrated and merged through dynamic migration of the virtual machines is judged through a light load host detection algorithm;
step four, after acquiring an overloaded host and a light load set in the system, executing a virtual machine selection algorithm; for a lightly loaded host, virtual machine selection need not be performed;
step five, after the virtual machine set to be migrated is obtained, executing the migration algorithm of the virtual machine of the overload host;
and step six, the execution time of the overload host virtual machine migration is the same, and after the virtual machine set to be migrated is obtained, the light-load host virtual machine migration algorithm is executed.
2. The method for dynamically migrating virtual machines based on user qos requirement differences according to claim 1, wherein the first step specifically includes:
(1) acquiring all PM sets PMS in the system;
(2) for each physical host pm in the set PMS, judging whether the current residual CPU and bandwidth resources meet the requirements of vm, and if not, removing the physical host pm from the set PMS;
(3) predicting whether pm can violate the requested QoS requirement on the assumption that the current vm is loaded, and if pm can violate the QoS requirement of vm, removing pm from the PMS set;
(4) and allocating pm with the highest CPU utilization rate in the PMS to vm.
3. The method for dynamically migrating virtual machines based on user qos requirement differences according to claim 1, wherein the step of determining whether the virtual machines can be migrated through dynamic migration of virtual machines by using a light-load host detection algorithm specifically includes:
(1) for each virtual machine instance in the current physical host, calculating the percentage vmCpu of the CPU resource occupied by the virtual machine instance in the current PM and acquiring the maximum value vmCpu in the percentage vmCpumax；
(2) Acquiring the CPU utilization rate hostCpu of the current host;
(3) if hostCpu-vmCpumaxIf the PM is less than eta, the current PM is considered to be a light-load host, and dynamic migration of the virtual machine is carried out; wherein eta is the CPU power consumption model parameter.
4. The method for dynamically migrating virtual machines based on user qos requirement differences according to claim 1, wherein the fourth step specifically includes:
(1) obtaining a virtual machine set VMS in a physical hostiCalculating the current VMSiThe variance of the QoS requirement of (preVar);
(2) for VMSiPer virtual machine instance vm, compute assumptions slave VMSiIn-after vm removal VMSiThe variance of the QoS requirement of (postVar);
(3) get each get per virtual machine, get VMS after removing itiThe decrease in variance of the QoS requirements of (a) Δ decVar ═ preVar-postVar;
(4) mixing VMSiAll the virtual machines in (1) are sorted in descending order according to the value of delta decVar;
(5) in turn from the VMSiRemove the VM and add it to the to-be-migrated list until the current physical host is no longer overloaded.
5. The method for dynamically migrating virtual machines based on user qos requirement differences according to claim 1, wherein the step five specifically includes:
(1) acquiring VMS (virtual machine set to be migrated) from overloaded hostmigAnd sorting in ascending order according to QoS requirements;
(2) for each one belonging to the VMSmigTraversing a PM set PMS in the system, and for each physical host PM, calculating a variance preVar of the QoS requirement of the virtual machine set in the current PM and a variance postVar of the QoS requirement of the virtual machine set in the PM after the vm is supposed to be migrated to the PM;
(3) calculating decVar ═ postVar/preVar, and filtering out all PMs with decVar being more than or equal to 1; acquiring the CPU utilization rate cpuNext of pm after vm is supposed to migrate into pm; calculating temp ═ decVar/cpuNext;
(4) recording the minimum value of temp and the corresponding pm, and if the pm exists, migrating the current virtual machine vm to the pm; otherwise, removing vm from the list to be migrated, and not migrating.
6. The method for dynamically migrating virtual machines based on user qos requirement differences according to claim 1, wherein the sixth step specifically includes:
(1) acquiring VMS (virtual machine set) to be migrated from light-load hostmigSorting the VMs in a descending order according to the requirements of the CPU resources;
(2) for VMSmigEach virtual machine instance vm iniThe source host is pmjTraversing the set of physical hosts, for each physical host pm thereinkCalculate the sum vmiSlave pmjMigration to pmkReduction of post-system energy consumption:
Esave(vmi,pmj,pmk,t)＝Edif(vmi,pmj,pmk,t)-Emigr(vmi,pmj,pmk,t)；
wherein Edif(vmi,pmj,pmkAnd t) denotes time t due to vmiAre respectively at pmjAnd pmkThe difference in energy consumption caused, and Emigr(vmi,pmj,pmkAnd t) represents vm at time tiSlave pmjMigration to pmkThe energy consumption brought is respectively expressed as:
Figure FDA0003192950020000031
Figure FDA0003192950020000041
wherein t isre(vmi) Is vmiRemaining service time of Pcpudec(pmj,vmiAnd t) represents that vmiSlave pmjPost removal pmjReduction of CPU power consumption, Pcpuinc(pmk,vmiAnd t) represents that vmiMigration into pmkRear pmkIncrease in CPU Power consumption, Pnet(pmjT) and Pnet(pmkT) each represents pmjAnd pmkIncreased power consumption of network communication due to migration, tmigr(vmi,pmj,pmkT) is time vmiSlave pmjMigration to pmkThe migration time required, is expressed as:
Figure FDA0003192950020000042
wherein the RAM (vm)i) Represents vmiMemory size, BW ofre(pmjT) represents time pmjAvailable bandwidth of;
(3) record Esave(vmi,pmj,pmkT) maximum value of and compares the current vmiMigrate to corresponding pmkIn (1).
7. A virtual machine scheduling system applying the method for dynamically migrating virtual machines based on the difference of user service quality demands according to any one of claims 1 to 6.
8. A virtual machine applying the method for dynamically migrating the virtual machine based on the difference of the user service quality demands according to any one of claims 1 to 6.
CN201810836808.1A 2018-07-26 2018-07-26 Dynamic migration method of virtual machine based on user service quality demand difference Active CN109144664B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810836808.1A CN109144664B (en) 2018-07-26 2018-07-26 Dynamic migration method of virtual machine based on user service quality demand difference

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810836808.1A CN109144664B (en) 2018-07-26 2018-07-26 Dynamic migration method of virtual machine based on user service quality demand difference

Publications (2)

Publication Number Publication Date
CN109144664A CN109144664A (en) 2019-01-04
CN109144664B true CN109144664B (en) 2021-09-14

Family

ID=64798043

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810836808.1A Active CN109144664B (en) 2018-07-26 2018-07-26 Dynamic migration method of virtual machine based on user service quality demand difference

Country Status (1)

Country Link
CN (1) CN109144664B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110083435A (en) * 2019-05-08 2019-08-02 扬州大学 Virtual machine (vm) migration design method based on coalition formation
CN115562812A (en) * 2022-10-23 2023-01-03 国网江苏省电力有限公司信息通信分公司 Distributed virtual machine scheduling method, device and system for machine learning training

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105159751A (en) * 2015-09-17 2015-12-16 河海大学常州校区 Energy-efficient virtual machine migration method in cloud data center
CN106227600A (en) * 2016-07-25 2016-12-14 湖南大学 A kind of multidimensional virtual resource allocation method based on Energy-aware
CN108170522A (en) * 2017-12-06 2018-06-15 南京邮电大学 A kind of cloud computing virtual machine (vm) migration control method based on dynamic threshold

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8627328B2 (en) * 2008-11-14 2014-01-07 Oracle International Corporation Operation control for deploying and managing software service in a virtual environment

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105159751A (en) * 2015-09-17 2015-12-16 河海大学常州校区 Energy-efficient virtual machine migration method in cloud data center
CN106227600A (en) * 2016-07-25 2016-12-14 湖南大学 A kind of multidimensional virtual resource allocation method based on Energy-aware
CN108170522A (en) * 2017-12-06 2018-06-15 南京邮电大学 A kind of cloud computing virtual machine (vm) migration control method based on dynamic threshold

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
Dynamic Virtual Machine Migration Algorithms Using Enhanced Energy Consumption Model for Green Cloud Data Centers;Jing Huang等;《2014 International Conference on High Performance Computing & Simulation》;20140922;第902-910页 *
Optimal online deterministic algorithms and adaptive heuristics for energy and performance efficient dynamic consolidation of virtual machines in cloud data centers;Anton Beloglazov等;《Concurrency & Computation Practice & Experience》;20120910;第24卷(第13期);第1-24页 *
Server Consolidation with Migration Control for Virtualized Data Centers;Tiago C. Ferreto等;《Future Generation Computer Systems》;20111031;第27卷(第8期);第1027-1034页 *
基于能耗降低的虚拟机动态迁移算法;李飞标等;《华东理工大学学报（自然科学版）》;20171030;第43卷(第5期);第692-697页 *
数据中心虚拟机节能管理机制;朱德剑等;《计算机科学》;20171015;第44卷(第10期);第19-25页 *

Also Published As

Publication number Publication date
CN109144664A (en) 2019-01-04

Similar Documents

Publication Publication Date Title
CN108509276B (en) Video task dynamic migration method in edge computing environment
US10185592B2 (en) Network storage device using dynamic weights based on resource utilization
Hieu et al. Virtual machine consolidation with multiple usage prediction for energy-efficient cloud data centers
CN106790726B (en) Priority queue dynamic feedback load balancing resource scheduling method based on Docker cloud platform
CN105491138B (en) Distributed load scheduling method based on load rate graded triggering
Amini et al. A Dynamic SLA Aware Heuristic Solution For IaaS Cloud Placement Problem Without Migration
Vignesh et al. Resource management and scheduling in cloud environment
CN107122233B (en) TSN service-oriented multi-VCPU self-adaptive real-time scheduling method
CN104142860A (en) Resource adjusting method and device of application service system
KR20130011491A (en) Apparatus or task assignment, method for task assignment and a computer-readable storage medium
Mosa et al. Dynamic virtual machine placement considering CPU and memory resource requirements
CN107220108B (en) Method and system for realizing load balance of cloud data center
CN112905326B (en) Task processing method and device
CN109144664B (en) Dynamic migration method of virtual machine based on user service quality demand difference
US8312463B2 (en) Resource management in computing scenarios
CN114048006A (en) Virtual machine dynamic migration method and device and storage medium
CN107370783B (en) Scheduling method and device for cloud computing cluster resources
CN113672391B (en) Parallel computing task scheduling method and system based on Kubernetes
More et al. Energy-aware VM migration using dragonfly–crow optimization and support vector regression model in Cloud
CN110795323A (en) Load statistical method, device, storage medium and electronic equipment
Ajayi et al. Multi-Class load balancing scheme for QoS and energy conservation in cloud computing
Gouasmi et al. Cost-efficient distributed MapReduce job scheduling across cloud federation
CN110069319B (en) Multi-target virtual machine scheduling method and system for cloud resource management
CN115878260A (en) Low-carbon self-adaptive cloud host task scheduling system
US20120042322A1 (en) Hybrid Program Balancing

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