CN115718644A - Computing task cross-region migration method and system for cloud data center - Google Patents
Computing task cross-region migration method and system for cloud data center Download PDFInfo
- Publication number
- CN115718644A CN115718644A CN202211496048.7A CN202211496048A CN115718644A CN 115718644 A CN115718644 A CN 115718644A CN 202211496048 A CN202211496048 A CN 202211496048A CN 115718644 A CN115718644 A CN 115718644A
- Authority
- CN
- China
- Prior art keywords
- migration
- host
- energy consumption
- load
- data center
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- 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
A computing task cross-region migration method and system for a cloud data center are disclosed, wherein a computing task cross-region migration process comprises three steps of migration environment sensing, migration decision and migration synchronization, the migration environment sensing step dynamically monitors the state of the cloud data center, and information such as cloud data center load and power consumption is acquired in real time; in the migration decision step, a self-adaptive dynamic threshold method based on the energy consumption level is adopted to determine the migration trigger time, optimization targets such as low energy consumption and load balance are comprehensively considered, and an optimal migration task set and a migration target host set are given; the migration synchronization step realizes the consistency of data and states before and after the migration and the dynamic adaptation of the network resource environment. The cloud data center energy consumption optimizing method and the cloud data center energy consumption optimizing system can reduce the energy consumption of the cloud data center system, simultaneously meet the requirements of system load balancing and application performance, help to promote the development of the cloud data center energy consumption management technology, and form a reliable operation dynamic cross-region migration scheme in the aspects of cloud data center energy consumption optimization and operation migration.
Description
Technical Field
The invention belongs to the technical field of cloud computing, and particularly relates to a computing task cross-region migration method and system for a cloud data center.
Background
In recent years, cloud computing technology is receiving more and more attention and application, the cloud service scale is continuously enlarged, the number and scale of cloud data centers are also remarkably increased to support a large amount of data storage and user computing requests, and with the continuous development of the technology, information transmission, processing and the like of the cloud data centers consume more electric energy, so that how to use electric power more energy-saving and green is one of the most concerned problems of data center enterprises and the whole society. According to statistics, the energy consumption of the server in the idle state is 50% -70% of the energy consumption of the server in the full load state, that is, the idle static server does not contribute to the calculation task but consumes a large part of energy consumption, so that the energy consumption management scheme of some data centers can monitor the utilization rate of the server to trigger the migration of virtual machines and containers to integrate underutilized servers. In addition to the energy consumption optimization scenario, the migration technology of virtual machines and containers is also used in the load balancing, server upgrading and machine shutdown maintenance scenarios. For example, virtual machines and containers on overloaded nodes are migrated, so that the situation that the application performance on the nodes is reduced due to the lack of node resources is prevented, and the goals of improving cluster load balance and cluster resource utilization rate are achieved.
Selecting a proper virtual machine and container to transfer under the target constraint and finding the most proper migration target node is a key problem of computing task migration. Currently, the migration scheduling process of virtual machines and containers is generally divided into three aspects of determining the migration opportunity (When), the migrated virtual machine, the selection of the container (whish), and the selection of the migration target physical host node (Where).
At present, the migration opportunity is mainly determined by setting a migration trigger threshold, overload migration is triggered when the host load exceeds an overload threshold, and underload migration is triggered when the host load is lower than an underload threshold. The migration trigger threshold is divided into a static threshold and a dynamic threshold. The static threshold value directly sets the overload threshold value and the low-load threshold value manually, the implementation is simple, but the threshold value cannot be changed in the whole migration process and is not flexible enough; the dynamic threshold value can dynamically change according to time, resource conditions and cluster load conditions, and has better adaptability and flexibility, however, in most of the current researches, a threshold value dynamic change rule is only formulated according to the resource and load conditions, the influence of the cluster energy consumption level on the threshold value is not considered, the reaction speed of system energy consumption optimization can be accelerated by formulating the dynamic threshold value according to the cluster energy consumption level, the system energy consumption optimization effect is improved, and therefore, the influence of the cluster energy consumption level on the threshold value is considered to be significant.
The selection of the migrated virtual machine and the migrated container mainly includes maximum correlation selection, minimum migration cost selection, minimum performance loss selection and other mode selections, which respectively consider one or more migration indexes such as resource utilization rate, load balance, migration time, migration data volume, application performance, service quality, energy consumption and the like, but lack a method for meeting comprehensive requirements, and lack countermeasures when a specific reduction amount of host energy consumption is required.
Currently, heuristic algorithms, meta-heuristic algorithms and machine learning algorithms are mainly used to determine the target physical host. The heuristic algorithm has the advantages of simple realization, high convergence speed and the like, but is easy to fall into a local minimum value, and cannot achieve the optimal overall placement effect. Relatively, the meta-heuristic algorithm can better find the global optimal solution, but the convergence speed is low, the algorithm parameters are too many, the reusability of the calculation result is poor, and the parameter tuning cannot be quickly and effectively performed. The machine learning algorithm has the characteristics of self-learning, self-adaption and the like, has better global search capability, but has poorer intelligibility and higher dependence on training data. Most of the related researches for determining the target physical host only consider finding the target host in the data center, and lack of consideration on cross-data center migration, when the energy consumption level of the data center is high and the overall resource utilization is saturated, the cross-data center finding of the target host is necessary.
Disclosure of Invention
The invention aims to solve the problems that: dynamic thresholds have better adaptability and flexibility than static thresholds, how can threshold dynamic change rules be formulated for cloud data center energy consumption levels? How to comprehensively and comprehensively consider multiple migration indexes such as energy consumption, resource utilization rate, load balancing, and service quality? How to configure countermeasures when there is a demand for a specific reduction amount of host power consumption? How to search for a target host across data centers for cross-regional migration of computing tasks when the energy consumption level of a cloud data center is high and the overall resource utilization is saturated? These problems remain to be solved.
The technical scheme of the invention is as follows: a computing task cross-region migration method facing a cloud data center considers data management requirements and dynamic load balancing requirements of the cloud data center, aims at reducing system energy consumption, and conducts computing task cross-region migration, and comprises the following steps:
step one, migration environment perception: the method comprises the steps of dynamically monitoring the state of a cloud data center, and acquiring the load and power consumption of the cloud data center and the characteristic attribute and operation information of a computing task in real time;
step two, migration decision: the method comprises the following steps of obtaining node information of a cloud data center through a migration environment sensing step, determining migration triggering time based on an energy consumption level self-adaptive dynamic threshold method, comprehensively considering optimization targets of low energy consumption, load balancing and QoS, determining a migration decision, and giving an optimal migration task set and a migration target host set, wherein the migration decision comprises the following steps:
step B1, determining the migration time: setting migration triggering thresholds of low load and overload, and migrating when the load of the physical host reaches the migration triggering thresholds;
step B2, giving an optimal emigration task set: when the low-load migration is triggered, all the computing tasks on the low-load host computer are migrated out; when overload migration is triggered, arranging migrated computing tasks according to load correlation between the virtual machines and the host, load correlation between the containers and the host and comprehensive load of the computing tasks, and if migration has a requirement on specific reduction of energy consumption of the host, solving an optimal migration task set for all computing tasks on the overload host by taking the minimum migration cost of the containers and the virtual machines as a target under the requirement on the reduction of the energy consumption;
step B3, giving an optimal migration destination host set: for the host with the task immigration condition, comprehensively considering the energy consumption of the host, load balance, resource utilization rate and user service quality, and selecting to obtain an optimal immigration host set by adopting a heuristic multi-target optimization method;
step three, migration synchronization: and data synchronization before and after the migration of the virtual machine and the container is realized by adopting an increment synchronization mode, and a network environment of a migration target host is dynamically adapted by adopting a mode of binding a virtual IP address by a task.
The migration environment sensing step is designed to dynamically monitor the states of the server host, the virtual machine, the container and the computing task and acquire the load and the power consumption of the data center and the characteristic attribute and the running information of the computing task in real time. The method comprises the steps of firstly respectively obtaining load information of a CPU, a memory, an IO and a network, calculating the resource utilization rate of nodes by utilizing the load information, then carrying out multilevel and multidimensional energy consumption quantitative analysis on a physical host, a virtual machine, a container and the like by adopting intelligent learning methods such as a decision tree, a random forest, a Support Vector Machine (SVM), an association rule, a deep neural network and the like and combining a multivariate polynomial regression method, establishing a refined model of power consumption measurement of the virtual machine, the container and the physical nodes, obtaining the power consumption of the virtual machine, the container and the physical nodes through the model and the load data, and calculating the energy consumption of a cloud data center.
In the migration decision step, a self-adaptive dynamic threshold method based on the energy consumption level is adopted to determine the migration trigger time, optimization targets such as low energy consumption and load balancing are comprehensively considered, and an optimal migration task set and a migration target host set are given. Defining an underload threshold and an overload threshold of a host, further dynamically and automatically adjusting a migration trigger threshold according to different energy consumption levels, monitoring the load of the host through a migration environment sensing step, and triggering underload migration or overload migration according to load data of the host within a period of time by combining the underload threshold and the overload threshold. When the low-load migration is triggered, all the computing tasks on the low-load host computer are migrated; when overload migration is triggered, a computing task with higher load correlation with a physical node can be selected for migration, when specific reduction of host energy consumption is required, a dynamic programming method can be used for providing a set of minimum container and virtual machine migration cost meeting the energy consumption reduction requirement, and the migration cost is a positive correlation function of migration data volume. The selection of the migration target host set is given by comprehensively considering host energy consumption, load balance, resource utilization rate, user service quality and the like by adopting a heuristic multi-objective optimization method.
In migration synchronization, the main purpose of incremental synchronization migration is to reduce downtime as much as possible, after the migration starts, a virtual machine and a container of a source host are not immediately terminated but continue to operate, memory data of a source node is sent to a target node in transmission iteration, all memory page data are sent in first iteration, only modified memory page data are sent in subsequent iteration, until the memory page modification rate is reduced to be below a threshold value or the iteration number upper limit is reached, the iteration process is ended, the virtual machine and the container of the source node are closed, the last round of dirty page data is transmitted, then the virtual machine and the container of the target node are recovered and started, and the whole incremental synchronization process is completed. In a stateful migration behavior, the running data and the related configuration environment in the memory need to be maintained in the migration process, and the service quality after the migration of the computing task is ensured not to be reduced, wherein the most important point is the network environment.
Based on the method, the invention further provides a cloud data center system, which comprises a plurality of data centers, wherein the cloud data center system is configured with a computer program, and the computer program is executed to realize the cross-region migration method of the computing task.
The invention has the following beneficial effects:
the invention researches and designs a cross-region migration strategy of a cloud data center computing task. An adaptive dynamic threshold method based on energy consumption level is provided, and a migration trigger threshold can be dynamically and automatically adjusted according to different energy consumption levels; comprehensively considering factors such as energy consumption, resource utilization rate, load balance, service quality and the like, selecting a migrated task set, and using a dynamic planning method to provide a minimum migrated operation set meeting the energy consumption reduction requirement when the specific reduction of the host energy consumption is required; selecting a host with the calculation capacity, the memory space and the network performance meeting the requirements as a migration candidate host set, then comprehensively considering the host energy consumption, the load balance, the resource utilization rate and the user service quality, adopting a heuristic multi-target optimization method to select an optimal migration host set, and when the cloud data center is high in energy consumption level and the overall resource utilization is saturated, searching a target host across the data center to perform cross-region migration of calculation tasks; the data synchronization and the state consistency before and after the migration are realized, and the task can be ensured to be dynamically adaptive to the resource environment of the migration target host. The research result of the invention can be directly applied to the cloud data center, which is beneficial to filling up the domestic technical blank and promoting the development of the energy consumption management technology of the cloud data center; in the aspects of energy consumption optimization and operation migration of the cloud data center, a reliable operation dynamic trans-regional migration technology is formed.
Drawings
Fig. 1 is a general architecture diagram of an embodiment of a cloud data center-oriented computing task cross-region migration method according to the present invention.
Fig. 2 is a flowchart of a computing task cross-region migration method facing a cloud data center.
Fig. 3 is a configuration diagram of a policy scheme for transferring computing tasks across zones in a cloud data center according to the present invention.
FIG. 4 is a diagram of the dynamic migration scheme architecture of the present invention based on migration synchronization techniques.
Detailed Description
The invention aims to provide a computing task cross-region migration technology facing a cloud data center, which can reduce the energy consumption of a cloud data center system and meet the requirements of system load balance and application performance at the same time.
The invention relates to a cloud data center-oriented computing task cross-region migration technology, aiming at reducing system energy consumption, wherein the overall architecture of the implementation scheme is shown in figure 1, and a cloud data center computing task cross-region migration core technology is researched based on the requirements of reducing the energy consumption cost of a cloud data center, uniformly optimizing and managing resources, balancing dynamic loads and the like. The specific process of the invention is as shown in fig. 2, firstly, a migration environment sensing step is carried out, the state of the cloud data center is dynamically monitored, the load and power consumption of the cloud data center and the characteristic attribute and running information of the computing task are obtained in real time, then, through a migration decision step, the migration trigger time is determined, the optimal migration task set and the migration target host set are given out, then, the data synchronization before and after the migration of the virtual machine and the container is realized, the data is dynamically adapted to the network environment of the target host, and finally, the computing task is started to run on the new host. The specific technical scheme adopted by the invention is as follows.
Step one, migration environment perception: the migration environment sensing step is designed for dynamically monitoring the states of the server host, the virtual machine, the container and the computing task and acquiring the load and the power consumption of the data center and the characteristic attribute and the running information of the computing task in real time.
As a preferred embodiment, in step one, A1: acquiring physical node load at regular intervals and calculating node resource utilization rate U = { U = CPU ,U MEM ,U IO ,U BW }，U CPU For CPU utilization, U MEM For memory utilization, U IO For IO utilization, U BW Is the network bandwidth utilization; a2, establishing a refined model of the power consumption measurement of the virtual machine, the container and the physical node, obtaining the power consumption of the virtual machine, the container and the physical node through the model and the load data, and calculating to obtain the energy consumption of the cloud data center.
In A1, CPU load information can be obtained through a top command, and a calculation method of the CPU utilization rate is as follows:
wherein the content of the first and second substances,
the percentage of free CPU that is directly output for the top command.
The memory load information can be obtained through the free command, and the calculation formula of the memory utilization rate is as follows:
wherein M is total Is the total amount of memory of the host, M free Is the size of the free memory.
Obtaining the I/O times U of a disk per second through an iostat command iops The calculation formula of the IO utilization rate is as follows:
wherein, M iops Is the maximum achievable IO count per unit time for the disk.
The network load information can be obtained through the ifconfig command, and the calculation formula of the network bandwidth utilization rate is as follows:
wherein F in Is the amount of local data flowing in per unit time, F out Is the amount of local outgoing data per unit time, B net Network card bandwidth for the host.
In the step A2, aiming at the characteristics of the cloud data center task request and resource use diversity, demand, dynamics and sharing, multi-level and multi-dimensional energy consumption quantitative analysis of a physical host, a virtual machine, a container and the like is carried out by combining intelligent learning methods such as a decision tree, a random forest, a Support Vector Machine (SVM), an association rule, a deep neural network and the like with a multivariate polynomial regression method.
Physical host computer: the physical host, i.e., the physical node, usually has a positive correlation between the power consumption of the physical host and its CPU utilization, memory utilization, bandwidth utilization, etc., and quantifies power consumption models generated for different load states (high/medium/low) under different hardware specification configuration conditions, such as CPU number, kernel number, memory number, GPU presence, etc., with the physical host as a modeling unit.
Virtual machine: and establishing energy consumption quantification models of the virtual machines with different configurations under various load states (high/medium/low) by taking the virtual host as a modeling unit.
A container: and establishing energy consumption quantification models of containers with different configurations under different working load states (high/medium/low) by taking the container as a modeling unit.
The cloud data center can express the sum of the energy consumptions of the hosts by using the energy consumptions of the physical hosts, wherein the energy consumption of the physical hosts = the static energy consumption of the hosts + the dynamic energy consumption generated by the virtual machines and containers running on the hosts; the energy consumption models of the virtual machines and the containers are used for measuring the energy consumption of the migrated objects (the virtual machines and the containers) in the subsequent migration decision. After power consumption/energy consumption models of the host, the virtual machine and the container are established, the energy consumption is quantitatively analyzed by adopting the existing data learning analysis methods such as the decision tree and the random forest.
The energy consumption of the cloud data center is equal to the sum of the energy consumptions of all physical nodes in the center, so that the energy consumption calculation formula of the center in the time interval from t1 to t2 is as follows:
h is the set of all physical nodes of the cloud data center, and P h (t) is the power consumption of the physical node h at the time t, given by the power consumption model of the physical host, and t is the element [ t1, t2 ]]。
Step two, migration decision: as shown in fig. 3, a migration decision-making specific scheme is that node information is obtained through a migration environment sensing step, a migration trigger time is determined based on an energy consumption level adaptive dynamic threshold method, optimization objectives such as energy consumption, load balancing, qoS and the like are comprehensively considered, and an optimal migration task set and an migration destination host set are given as follows:
step B1, determining the migration time: two migration trigger thresholds are configured, and a low load threshold Thr of a physical host in the cloud data center is defined low ＝{CPU low ,MEM low ,IO low ,BW low Therein, CPU low The component is the lower threshold of host CPU utilization, MEM low The component is the lower threshold of memory utilization, IO low The component is the lower limit threshold value of IO utilization ratio, BW low The components are a lower limit threshold of the network bandwidth utilization rate and an overload threshold Thr high ＝{CPU high ,MEM high ,IO high ,BW high }，CPU high Component is the upper threshold of host CPU utilization, MEM high The component is the upper threshold of memory utilization, IO high The component is the IO utilization upper threshold, BW high The component is the upper threshold of network bandwidth utilization, wherein the CPU high According to dynamic adjustment of energy consumption level, corresponding CPUs are configured for different energy consumption levels high . And monitoring the load calculation utilization rate of the host through a migration environment sensing step, determining a migration trigger threshold corresponding to the current energy consumption level, and realizing self-adaptive dynamic migration trigger. In time interval t, m groups of load data of one host computer are collected, and when at least n groups (n) of m groups of data are collected<= m) data less than Thr low All components of, the host triggers an underloaded migrationWhen at least n groups (n) of the m groups of data are present<= m) data greater than Thr high If any, then the host triggers overload migration. The self-adaptive dynamic threshold value method based on the energy consumption level dynamically and automatically adjusts the migration trigger threshold value according to different energy consumption levels.
As a preferred embodiment, in step B1, the invention describes the energy consumption level of the cloud data center by using a three-color energy consumption representation method, and sets the maximum tolerable energy consumption of the center as E max When the central energy consumption is 0.5E max The energy consumption status is blue at 0.5E max And 0.75E max In the middle, the state is green, at 0.75E max The above state is red. Overload threshold Thr high Of (2) MEM high 、IO high And BW high The component adopts a static threshold value, the value of the static threshold value is 0.8 no matter how the energy consumption level of the cloud data center changes high Is a CPU high The components adopt dynamic threshold values, and the specific calculation formula is as follows:
thus realizing Thr high Based on the adaptive dynamic adjustment of the power consumption level, wherein the CPU max The maximum amount of CPU resources owned by the physical host, step is (0, 0.1)]α is an adjustable parameter of (0, 1). When the energy consumption level of the cloud data center is in a blue state, the CPU of the overload threshold value high Keeping unchanged and taking the value of 0.9 multiplied by CPU max (ii) a When the energy consumption level is in green state, CPU high Entering a fixed step size reduction process, wherein the minimum value is 0.7 multiplied by CPU max (ii) a When the energy consumption level is in red state, CPU high Entering a constant coefficient reduction process with a minimum value of 0.6 × CPU max . The low-load threshold value adopts a static threshold value method, and the four components are kept unchanged and take a value of 0.2, namely Thr low ＝{0.2，0.2，0.2，0.2}。
B2, giving an optimal emigration task set: when the low-load migration is triggered, all the calculation tasks on the low-load host computer are migrated without involving the optimal selection; when overload migration is triggered, the invention provides two optimal migration task selection methods, and an administrator can independently select according to different requirements.
The first method is that the calculation tasks with high load correlation and small comprehensive load value are selected for migration, so that the host can be recovered to a normal load state relatively quickly, and the migration cost is relatively low.
The load dependency is first calculated, and the reason for considering the load dependency is that the higher the load dependency of a virtual machine or a container to a host, the higher the probability that the virtual machine or the container will overload the host. The measurement of the correlation mainly comprises a Pearson correlation coefficient method, an Euclidean distance method, an included angle cosine similarity method and the like. The computing task is packaged in a virtual machine or a container, namely the load correlation between the virtual machine and the host and between the container and the host are computed, and the historical load record of the virtual machine is recorded as { a 1 ,a 2 ,…,a k The historical load of the container is recorded as b 1 ,b 2 ,…,b k The historical load of the physical host is recorded as c 1 ,c 2 ,…,c k And then, the correlation coefficient calculation formula is as follows:
wherein r is ac Is the correlation coefficient of the virtual machine and the physical host, r bc Is the correlation coefficient of the container with the physical host,
the average values of the historical load data of the virtual machine, the container and the physical host respectively, and the value of the correlation coefficient r is [ -1,1]A larger absolute value of r indicates a higher degree of correlation. When it is 0.8<When | < r | > is less than or equal to 1, the virtual machine, the container and the host are in extremely strong correlationWhen it is 0.6<And l r l is less than or equal to 0.8, the virtual machines and the containers are in strong correlation with the host, and the virtual machines and the containers meeting the strong correlation and strong correlation are added into the candidate emigration set.
Then, calculating the comprehensive load value of the elements in the candidate migration set, wherein the calculation formula is as follows:
Load l ＝0.5×CPU l +0.3×MEM l +0.1×IO l +0.1×BW l
wherein, load l For the integrated load value of the l-th element in the set, the CPU l CPU load value of element l, MEM l Is the memory load value, IO, of element l l IO load value, BW, for element l l The network bandwidth load value of element l. And sorting the virtual machines and the containers in the candidate migration set in a descending order according to the comprehensive load value, and sequentially migrating until the overloaded host recovers the normal load level.
And secondly, when the specific reduction amount of the energy consumption of the host is required, providing a set of minimum migration operation meeting the specific requirements of energy consumption reduction. The invention uses a dynamic planning method to provide a set of minimum migration operation, and assumes that the specific reduction amount of the required energy consumption is E d Adding all the calculation tasks on the overloaded host computer into the candidate migration set, and obtaining the energy consumption metric value { e ] of the candidate migration object according to the virtual machine and container energy consumption fine model in the step A2 1 ,e 2 ,e 3 ,…,e N The candidate migrated objects are all task objects on the overloaded host, wherein e j For the energy consumption of the task j, abstracting the selection problem of the migration task set into the following optimization problem:
namely, the migration cost of the container and the virtual machine is minimized under the condition of meeting the requirement of energy consumption reduction.
Wherein f (E) d ) For a set of candidate migrant tasks, p j The migration cost of the task j is a positive correlation function of the migration data volume.
Step B3, giving an optimal migration destination host set: for migrated virtual machines and containers, destination hosts need to be searched for placement, and in the process, the service quality of the cloud data center needs to be considered, so that the migration and placement cannot be obviously reduced. In the aspect of determining the target physical host, the problems are similar to the problems of initial placement of virtual machines and containers in nature and are the problems of multi-target combination optimization.
As a preferred embodiment, in step B3, first, a host with a computing capability, a memory space, and a network performance that all meet requirements is selected as a migration candidate host set, and then, a heuristic multi-objective optimization method is adopted to select an optimal migration host set, taking the host energy consumption, load balancing, resource utilization rate, user service quality, and the like into consideration. In addition, according to the energy consumption level of the cloud data center, different migration destinations are selected. When the energy consumption level of the cloud data center is in a blue state and a green state, carrying out internal migration of the data center, namely the migrated host and the migrated host are in the same data center; and when the migration host is in a red state, performing cross-region migration, namely, the migration host and the migration host are in different data centers, firstly selecting one data center in a blue state and a green state, and then selecting the optimal migration target host in the data center by using the target host selection mode.
Step three, migration synchronization: the migration synchronization aims at achieving data synchronization and state consistency before and after migration of the virtual machine and the container, shielding differences among hardware resources of the cloud computing center and ensuring that tasks can dynamically adapt to resource environments of the migration target host.
As a preferred embodiment, as shown in fig. 4, in step three, an incremental synchronization method is used for migrating a virtual machine and a container, after the migration starts, the virtual machine and the container of a source host are not immediately terminated, but continue to run, memory data of a source node is sent to a target node in a transmission iteration, in a first iteration, all memory page data are sent, in a subsequent iteration, only the modified memory page data are sent, until a memory page modification rate is reduced below a threshold or reaches an upper limit of iteration times, an iteration process is ended, the virtual machine and the container of the source node are closed and the last round of dirty page data is transmitted, and then a virtual machine or a container of a starting target node is recovered, so that the entire incremental synchronization process is completed. The increment synchronization method reduces the data volume to be transmitted from all memory pages to the dirty page of the last iteration through the repeated iteration process, greatly shortens the transmission time and reduces the downtime.
Meanwhile, the invention makes full use of the hierarchical characteristics of the container to preferentially migrate the read-only public mirror image to the destination node, and then only needs to synchronize the read-write layer at the uppermost layer of the container, thereby greatly reducing the data volume and further reducing the data synchronization time. The container technology is not similar to a Hypervisor layer of a virtual machine technology, so that the container is essentially a process with specific parameters, the migration is mainly realized by a process real-time migration tool CRIU, the CRIU is an abbreviation of a Checkpoint/Restore In User Space, and is a tool capable of realizing a Checkpoint/recovery mechanism In a User Space.
The state migration transmits the states of the virtual machine and the container, namely the memory page, the CPU state, the register and the disk content, to the target server, so that the virtual machine and the container at the target are the same as the state before the migration. Stateful migration behavior requires maintaining the running data and the related configuration environment in the memory during the migration process, and ensuring that the service quality after the migration of the computing task is not reduced, wherein the most important point is the network environment. Virtual machines, containers, and other endpoints typically communicate continuously, such as external clients and end-user applications, and it is often desirable to retain such communication after migration, i.e., the external client application continues to access the virtual machine or container in a seamless manner after migration, a problem known as connection maintenance. In addition, the endpoints may communicate using a connection-oriented protocol (e.g., TCP), with the connection representing a state shared between the endpoints. In addition to preserving connections between endpoints, connection migration also requires preserving active connection state, and therefore, endpoints need not establish new connections after migration. The invention adopts a mode of binding the virtual IP address by the task, ensures that the task can be dynamically adapted to the resource environment and the network environment of the migration target host through network address conversion, thereby shielding the heterogeneity and the difference of network resources of different data centers and enabling the migrated virtual machine/container to be seamlessly adapted to the network in the new host environment.
The method is realized based on a computer program and a cloud data system, and the invention further provides a cloud data center system which comprises a plurality of data centers, wherein the cloud data center system is configured with the computer program, and when the computer program is executed, the cross-region migration method of the computing task is realized.
In conclusion, the invention encapsulates computing and data processing services in virtual machines and containers to perform cross-region migration, dynamically reallocates computing resources, fully utilizes huge hardware resources of a data center, reduces energy consumption of a cloud data center system through steps of migration environment perception, migration decision, migration synchronization and the like, and simultaneously meets requirements of system load balance and application performance.
The invention is a technical invention which is suitable for the current development situation of the current cloud computing technology, the research result of the invention can be directly applied to a cloud data center, the domestic technical blank can be filled, the development of the energy consumption management technology of the cloud data center can be promoted, and a reliable dynamic trans-regional migration technology of the computing task can be formed in the aspects of energy consumption optimization and operation migration of the cloud data center. The economic benefit which the invention may produce will depend on the scope of popularizing and applying the degree, if can utilize well, will have very important effects to popularizing and utilizing of the cloud computing technology, play a good promoting effect to the development of the national economy, will produce better economic benefit.
Various modifications, additions and substitutions for the embodiments described may occur to those skilled in the art without departing from the spirit of the invention or exceeding the scope of the invention as defined in the accompanying claims.
Claims (9)
1. A computing task cross-region migration method facing a cloud data center is characterized in that the computing task cross-region migration is carried out by considering data management requirements and dynamic load balancing requirements of the cloud data center and taking system energy consumption reduction as a target, and the method comprises the following steps:
step one, migration environment perception: the method comprises the steps of dynamically monitoring the state of a cloud data center, and acquiring the load and power consumption of the cloud data center and the characteristic attribute and operation information of a computing task in real time;
step two, migration decision: the method comprises the following steps of obtaining node information of a cloud data center through a migration environment sensing step, determining migration triggering time based on an energy consumption level self-adaptive dynamic threshold method, comprehensively considering optimization targets of low energy consumption, load balancing and QoS, determining a migration decision, and giving an optimal migration task set and a migration target host set, wherein the migration decision comprises the following steps:
step B1, determining the migration time: setting low-load and overload migration trigger thresholds, and migrating when the load of the physical host reaches the migration trigger threshold, wherein the overload migration trigger threshold is adaptively adjusted according to the energy consumption level of the physical host;
and B2, giving an optimal emigration task set: when the low-load migration is triggered, all the computing tasks on the low-load host computer are migrated; when overload migration is triggered, arranging migrated computing tasks according to load correlation between the virtual machines and the host, load correlation between the containers and the host and comprehensive load of the computing tasks, and if migration has a requirement on specific reduction of energy consumption of the host, solving an optimal migration task set for all computing tasks on the overload host by taking the minimum migration cost of the containers and the virtual machines as a target under the requirement on the reduction of the energy consumption;
step B3, giving an optimal migration destination host set: for the host with the task immigration condition, comprehensively considering the energy consumption of the host, load balance, resource utilization rate and user service quality, and selecting to obtain an optimal immigration host set by adopting a heuristic multi-target optimization method;
step three, migration synchronization: and data synchronization before and after the migration of the virtual machine and the container is realized by adopting an increment synchronization mode, and the network environment of the migration target host is dynamically adapted by adopting a mode of binding a task with a virtual IP address.
2. The cloud data center-oriented computing task cross-region migration method according to claim 1, wherein the migration environment perception specifically comprises:
a1, acquiring physical node loads at set time intervals, and calculating the node resource utilization rate U = { U = (U) CPU ,U MEM ,U IO ,U BW }，U CPU For CPU utilization, U MEM For memory utilization, U IO For IO utilization, U BW Is the network bandwidth utilization;
step A2: establishing a model of power consumption measurement of the virtual machine, the container and the physical node according to the cloud data center, obtaining the power consumption of the virtual machine, the container and the physical node through the model and the load data, and calculating to obtain the energy consumption of the cloud data center; the energy consumption quantitative analysis of the physical host, the virtual machine and the container is carried out by adopting an intelligent learning method of a decision tree, a random forest, a Support Vector Machine (SVM), an association rule or a deep neural network and combining a multivariate polynomial regression method.
3. The cloud data center-oriented computing task cross-region migration method as claimed in claim 2, wherein in step A1, CPU load information is obtained through a top command, and a CPU utilization rate is calculated by using a formula as follows:
wherein the content of the first and second substances,
percentage of idle CPU directly output for top command;
obtaining memory load information through a free command, and calculating the memory utilization rate by adopting the following formula:
wherein M is total Is the total amount of host memory, M free Is the free memory size;
obtaining the I/O times U of the disk per second through an iostat command iops Calculating the IO utilization rate by adopting the following formula:
wherein, M iops Is the maximum achievable IO number per unit time of the disk;
obtaining network load information through an ifconfig command, and calculating the network bandwidth utilization rate by adopting the following formula:
wherein F in For local data volume flowing in per unit time, F out Is the amount of local outgoing data per unit time, B net The bandwidth of the network card for the host.
4. The cloud data center-oriented computing task cross-region migration method according to claim 2, wherein in the step A2, the energy consumption calculation specifically comprises:
1) Physical host computer: the power consumption of the physical host is in positive correlation with the CPU utilization rate, the memory utilization rate and the bandwidth utilization rate of the physical host, and the physical host is taken as a modeling unit to quantify power consumption models generated under different load states under different hardware specification configuration conditions;
2) Virtual machine: establishing energy consumption quantification models of virtual machines with different configurations under various load states by taking the virtual host as a modeling unit;
3) A container: establishing energy consumption quantification models of containers with different configurations under different working load states by taking the container as a modeling unit;
the energy consumption of the cloud data center is equal to the sum of the energy consumptions of all physical nodes in the center, so the energy consumption E of the center in the time interval from t1 to t2 is calculated as follows:
h is the set of all physical nodes of the cloud data center, and P h (t) is the power consumption of the physical node h at the time t, given by the power consumption model of the physical host, and t belongs to [ t1, t2 ]]。
5. The cloud data center-oriented computing task cross-region migration method according to claim 1, wherein the migration decision specifically comprises:
step B1, determining the migration time: setting a migration trigger threshold, and defining the low-load threshold of a host as Thr low ＝
{CPU low ,MEM low ,IO low ,BW low }，CPU low The component is the lower threshold of host CPU utilization, MEM low The component is the lower limit threshold of memory utilization, IO low The component is IO utilization lower limit threshold, BW low The component is a lower limit threshold of the network bandwidth utilization rate, and the overload threshold of the host is defined as Thr high ＝{CPU high ,MEM high ,IO high ,BW high }，CPU high The component is the upper threshold of the host CPU utilization, MEM high The component is the upper threshold of memory utilization, IO high The component is the IO utilization upper threshold, BW high The component is the upper threshold of network bandwidth utilization, wherein the CPU high Dynamically adjusting according to energy consumption level, configuring pairs for different energy consumption levelsCorresponding CPU high (ii) a Monitoring loads of all physical hosts through a migration environment sensing step, collecting m groups of load data of one physical host within a time interval T, and when at least n groups of data in the m groups of data are smaller than Thr low N is less than or equal to m, the host triggers low-load migration; when at least n groups of data in the m groups of data are more than Thr high If any component of (a), the host triggers overload migration;
step B2, giving an optimal emigration task set: when the low-load migration is triggered, all the calculation tasks on the low-load host computer are migrated without involving the optimal selection; when overload migration is triggered, the following two optimal task migration methods are selected according to requirements:
the method comprises the following steps: selecting the calculation tasks with high load correlation and small comprehensive load value for emigration,
firstly, calculating load correlation, wherein a calculation task is packaged in a virtual machine or a container, namely calculating the load correlation between the virtual machine and a host and between the container and the host:
recording the historical load of the virtual machine as { a ] by adopting a Pearson correlation coefficient method 1 ,a 2 ,…,a k The historical load of the container is recorded as b 1 ,b 2 ,…,b k The historical load record of the physical host is { c } 1 ,c 2 ,…,c k And then the correlation coefficient calculation formula is as follows:
wherein r is ac Is the correlation coefficient of the virtual machine and the physical host, r bc Is the correlation coefficient of the container with the physical host,
are respectively virtual machinesThe average value of historical load data of the container and the physical host, and the value of a correlation coefficient r is [ -1,1]In between, a larger absolute value of r indicates a higher degree of correlation; when it is 0.8<When the r is less than or equal to 1, the virtual machine, the container and the host are in extremely strong correlation, and when 0.6<The | r | is less than or equal to 0.8, the virtual machines, the containers and the host computer are in strong correlation, add the virtual machines and the containers which accord with the strong correlation and the strong correlation into the candidate emigration set;
then, calculating the comprehensive load value of the elements in the candidate migration set, wherein the calculation formula is as follows:
Load l ＝0.5×CPU l +0.3×MEM l +0.1×IO l +0.1×BW l
wherein, load l For the integrated load value of the ith element in the set, the CPU l CPU load value of element l, MEM l Is the memory load value, IO, of element l l IO load value, BW, for element l l Sorting the virtual machines and containers in the candidate migration set in a descending order according to the comprehensive load value for the network bandwidth load value of the element l, and sequentially migrating until the overloaded host recovers the normal load level;
the second method comprises the following steps: when migration has a requirement on specific reduction of host energy consumption, a dynamic planning method is used for providing a set of minimum migration operations meeting the specific requirement on energy consumption reduction, and the specific reduction of energy consumption is assumed to be E d Adding all task objects on the overloaded host into the candidate migration set, and obtaining an energy consumption metric value { e } of the candidate migration set according to the energy consumption perception of the virtual machines and the containers of the cloud data center 1 ,e 2 ,e 3 ,…,e N In which e is j For the energy consumption of the task element j, abstracting the selection problem of the migrated task set into the following optimization problem:
namely, under the condition of meeting the requirement of energy consumption reduction, the migration cost of the container and the virtual machine is minimized, wherein f (E) d ) For a set of candidate migrant tasks, p j The migration cost of the task j is a positive correlation function of the migration data volume;
step B3, giving an optimal migration destination host set: firstly, selecting a host with the computing capacity, the memory space and the network performance meeting the requirements as a migration candidate host set, then comprehensively considering the host energy consumption, the load balance, the resource utilization rate and the user service quality, and selecting an optimal migration host set by adopting a heuristic multi-objective optimization method; when the cloud data center is at a low energy consumption level, carrying out internal migration of the data center, namely the migrated host and the migrated host are in the same data center; and when the cloud data center is at a high energy consumption level, performing cross-regional migration, namely, migrating the host computer out and migrating the host computer into different data centers.
6. The cloud data center-oriented computing task cross-region migration method according to claim 5, wherein in step B1, based on an energy consumption level adaptive dynamic threshold method, a migration trigger threshold is dynamically and automatically adjusted according to different energy consumption levels, a three-color energy consumption representation method is used for describing the energy consumption level of the cloud data center, and the maximum tolerable energy consumption of the center is set as E max When the central energy consumption is 0.5E max The energy consumption status is blue at 0.5E max And 0.75E max Green state at 0.75E max The above state is red, the overload threshold Thr high Of (2) MEM high 、IO high And BW high The component adopts a static threshold value, and the value of the component is 0.8 and Thr no matter how the energy consumption level of the cloud data center changes high Is a CPU high The components adopt dynamic threshold values, and the specific calculation formula is as follows:
wherein, the CPU max The maximum amount of CPU resources owned by the physical host, step is (0, 0.1)]The adjustable parameter of alpha is (0, 1), and when the energy consumption level of the cloud data center is in the blue state, the CPU of the overload threshold value high Keeping the same; when the energy consumption level is in green state, CPU high Entering a fixed step length reduction process; when the energy consumption level is in red state, CPU high Entering a constant coefficient reduction process; the underload threshold remains constant throughout, thr low ＝{0.2，0.2，0.2，0.2}。
7. The cloud data center-oriented computing task cross-region migration method as claimed in claim 6, wherein when the energy consumption level of the cloud data center is in a blue or green state, internal migration of the data center is performed, that is, the migrated host and the migrated host are in the same data center; and when the migration host is in a red state, performing cross-region migration, namely, the migration host and the migration host are in different data centers, firstly selecting one data center in a blue state and a green state, and then selecting the optimal migration target host in the data center by using the target host selection mode.
8. The cloud data center-oriented computing task cross-region migration method according to claim 1, wherein the migration synchronization specifically comprises:
step C1, increment synchronization: after the migration starts, the virtual machine and the container of the source host do not terminate immediately, but continue to operate, the memory data of the source node is sent to the target node in transmission iteration, all the memory page data are sent in the first iteration, only the modified memory page data are sent in the subsequent iteration, the iteration process is finished until the memory page modification rate is reduced below a threshold value or reaches the upper limit of the iteration times, the virtual machine and the container of the source node are closed, the last round of dirty page data is transmitted, then the virtual machine and the container of the target end node are recovered, and the whole increment synchronization process is finished;
step C2, network adaptation: by adopting a mode of binding a virtual IP address by a calculation task and through network address conversion, the task can be ensured to be dynamically adaptive to the resource environment and the network environment of a migration target host, so that the heterogeneity and the difference of network resources of different data centers are shielded, and the migrated virtual machines and containers are seamlessly adapted to the network in the new host environment.
9. A cloud data center system, wherein the cloud data center system comprises a plurality of data centers, and the cloud data center system is configured with a computer program which, when executed, implements the computing task cross-region migration method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211496048.7A CN115718644A (en) | 2022-11-25 | 2022-11-25 | Computing task cross-region migration method and system for cloud data center |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211496048.7A CN115718644A (en) | 2022-11-25 | 2022-11-25 | Computing task cross-region migration method and system for cloud data center |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115718644A true CN115718644A (en) | 2023-02-28 |
Family
ID=85256688
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211496048.7A Pending CN115718644A (en) | 2022-11-25 | 2022-11-25 | Computing task cross-region migration method and system for cloud data center |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115718644A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116860723A (en) * | 2023-09-04 | 2023-10-10 | 合肥中科类脑智能技术有限公司 | Cross-computing center data migration method |
| CN117170870A (en) * | 2023-09-05 | 2023-12-05 | 国网智能电网研究院有限公司 | New energy consumption-oriented data center calculation force migration method and device |
| CN117170870B (en) * | 2023-09-05 | 2024-04-26 | 国网智能电网研究院有限公司 | New energy consumption-oriented data center calculation force migration method and device |
-
2022
- 2022-11-25 CN CN202211496048.7A patent/CN115718644A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116860723A (en) * | 2023-09-04 | 2023-10-10 | 合肥中科类脑智能技术有限公司 | Cross-computing center data migration method |
| CN116860723B (en) * | 2023-09-04 | 2023-11-21 | 合肥中科类脑智能技术有限公司 | Cross-computing center data migration method |
| CN117170870A (en) * | 2023-09-05 | 2023-12-05 | 国网智能电网研究院有限公司 | New energy consumption-oriented data center calculation force migration method and device |
| CN117170870B (en) * | 2023-09-05 | 2024-04-26 | 国网智能电网研究院有限公司 | New energy consumption-oriented data center calculation force migration method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Berger et al. | {AdaptSize}: Orchestrating the Hot Object Memory Cache in a Content Delivery Network | |
| CN111190688B (en) | Cloud data center-oriented Docker migration method and system | |
| CN109120715A (en) | Dynamic load balancing method under a kind of cloud environment | |
| US8429097B1 (en) | Resource isolation using reinforcement learning and domain-specific constraints | |
| CN108182105B (en) | Local dynamic migration method and control system based on Docker container technology | |
| CN113110933B (en) | System with Nginx load balancing technology | |
| CN107196865A (en) | A kind of adaptive threshold overload moving method of Load-aware | |
| CN102176696B (en) | Multi-computer system | |
| WO2023125493A1 (en) | Resource management method and apparatus, and resource management platform | |
| CN108196935A (en) | A kind of energy saving moving method of virtual machine towards cloud computing | |
| CN112835698A (en) | Heterogeneous cluster-based dynamic load balancing method for request classification processing | |
| Zhang et al. | Zeus: Improving resource efficiency via workload colocation for massive kubernetes clusters | |
| CN110597598B (en) | Control method for virtual machine migration in cloud environment | |
| CN115718644A (en) | Computing task cross-region migration method and system for cloud data center | |
| Zhao et al. | Dynamic replica creation strategy based on file heat and node load in hybrid cloud | |
| CN105393518B (en) | Distributed cache control method and device | |
| Selvi et al. | Popularity (hit rate) based replica creation for enhancing the availability in cloud storage | |
| Garg et al. | Optimal virtual machine scheduling in virtualized cloud environment using VIKOR method | |
| CN106775942B (en) | Cloud application-oriented solid-state disk cache management system and method | |
| You et al. | Performance evaluation of dynamic load balancing algorithms | |
| CN110888713A (en) | Trusted virtual machine migration algorithm for heterogeneous cloud data center | |
| CN115604269A (en) | Load balancing method and device of server, electronic equipment and storage medium | |
| CN110308991B (en) | Data center energy-saving optimization method and system based on random tasks | |
| Zhang | A QoS-enhanced data replication service in virtualised cloud environments | |
| CN113377544A (en) | Web cluster load balancing method based on load data dynamic update rate |
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 |