CN102790793A - Decision method and control module facing to cloud computing virtual machine migration - Google Patents

Abstract

The invention proposes a decision-making method and a control module for cloud computing virtual machine migration. The method consists of three parts: a migration trigger module, a virtual host selection module and a destination node selection module. First, the system load threshold is set through the migration trigger module, and the load forecasting technology is used to effectively avoid the unnecessary virtual machine migration triggered by the instantaneous load peak; then, the virtual host selection module selects the virtual host to be migrated according to the minimum migration cost strategy, Minimize the cost of virtual machine migration and save system resources; finally, the destination node selection module proposes a destination node selection strategy based on a weighted probability selection algorithm, which effectively solves the cluster conflict problem caused by the migration of a large number of virtual machines. Ultimately, this method can achieve load balancing very well, so that resources can be applied fairly and reasonably.

Description

A decision-making method and control module for cloud computing virtual machine migration

technical field

The invention relates to a decision-making method and a control module for cloud computing virtual machine migration. The method is applied to virtual machine migration in a cloud computing network system and belongs to the field of computer science and technology.

Background technique

Cloud computing (cloud computing) is a technology direction that has risen internationally in recent years, and it is also a research hotspot in the field of information technology. Cloud computing is a new computing method based on the Internet. In this way, the hardware resources of the entire system are virtualized into a unified virtual resource pool, and then all resources in the system are allocated to users according to actual needs or other policies with the help of a virtualization management platform. Because cloud computing realizes the unified management of computer resources, users no longer participate in the construction, management and maintenance of underlying hardware resources, system platforms, and even application software, but only need to pay attention to the actual needs of their own products, which greatly reduces the operating costs of enterprises. maintenance costs. Faced with its huge business opportunities, domestic and foreign information industry giants such as IBM, Amazon, Microsoft, Google, Huawei, China Mobile, Baidu, etc. have released their own cloud computing platforms.

Virtual machine migration technology is a key technology in cloud computing, through which virtual machines and their services can be completely migrated from one physical node (source computer) to another physical node (destination computer). With the help of virtual machine migration technology, the cloud computing system realizes online system maintenance and upgrade, resource dynamic management, system fault tolerance and dynamic load balancing, thereby improving system resource and energy efficiency, and improving system security performance.

According to the trigger factors of the virtual machine migration process, virtual machine migration can be divided into four types: CPU trigger, memory trigger, network bandwidth trigger, and external storage trigger; according to whether the response to user requests and service provision are interrupted during the migration process, virtual machine migration can be divided into four types: There are two types of static migration and dynamic migration. Static migration is a kind of downtime migration, that is, after the migration is triggered, the migrated virtual machine immediately stops executing and refuses to respond to new services until the virtual machine migration is completed; while dynamic migration is for the system to provide services to the outside world without interruption or without being interrupted. Complete virtual machine migration under the premise of user awareness. At present, mainstream virtualization software providers have proposed their own virtual machine online migration technologies, such as VMotion of VMware and Live Migration of Xen.

The existing migration mechanism generally manages the entire migration process through the virtual machine controller, and mainly uses the host CPU utilization as the basis for the node to start the migration, and the selection strategy of the migration destination node is usually based on random selection or based on the minimum bandwidth occupation, resulting in Existing migration strategies generally lack comprehensiveness in system resource sensitivity and resource analysis. In addition, the use of existing migration strategies for virtual machine migration may easily lead to unnecessary virtual machine migration triggered by instantaneous load peaks, or cause migration service conflicts caused by clustering effects, and even trigger secondary migration due to unreasonable selection of migration destination nodes . In view of the fact that virtual machine migration in the cloud computing system has the characteristics of large amount of migration data transmission and long time consumption, this patent proposes a decision-making method and control module for cloud computing virtual machine migration, which includes a migration trigger module, a virtual host selection module and Destination node selection module. This method can effectively avoid triggering unnecessary virtual machine migration, reduce the system load caused by virtual machine migration, solve the cluster conflict problem caused by the migration of a large number of virtual machines, and balance the system load well, and finally realize the cloud computing network Decision and control of virtual machine migration in .

Contents of the invention

Based on the current technical background and conditions, and aiming at the characteristics and requirements of cloud computing network and virtual computing resource management and scheduling, the present invention proposes a decision-making method and control module for cloud computing virtual machine migration.

The technical scheme that the present invention adopts is as follows:

The invention is a decision-making method and a control module for cloud computing virtual machine migration, which mainly consists of a migration trigger module, a virtual host selection module and a destination node selection module. Figure 1 shows the above three components and their interrelationships. Among them, the functions of each part are as follows:

The migration trigger module is a component of the virtual machine migration control function module in the cloud computing environment. Its function is to monitor the resource usage of the physical node (server) and its virtual machine in real time, and determine whether the source node is The virtual machine migration process needs to be initiated to avoid triggering unnecessary virtual machine migrations.

The virtual host selection module is a component of the virtual machine migration control function module in the cloud computing environment. The host machine minimizes the cost of virtual machine migration and saves system resources.

The destination node selection module is a component of the virtual machine migration control function module in the cloud computing environment. Its function is to select the destination node for virtual machine migration according to the resource usage of other nodes (except the source node) to avoid clustering The effect causes service conflicts and helps the system to achieve load balancing.

In a cloud computing system environment, the virtual machine migration control module can be deployed in a distributed or centralized manner. In distributed mode, the migration control module is deployed on each network node, and the nodes publish their load status and migration requests in the form of broadcast; in centralized mode, the migration control module is only deployed on the management host (Administrator) of the system. The management host collects statistics on all nodes in the network and the resource allocation and utilization of virtual machines running on them, and manages all virtual machine migration requests in the network in a unified manner. To facilitate management, centralized deployment is generally adopted in cloud computing systems. If there is no detailed description in this patent, the migration control module adopts a centralized deployment mode by default.

As shown in the flow chart of Figure 2, completing the virtual machine migration in the cloud computing environment can be divided into the following 5 steps: a, the source node (server) starts the migration trigger module, b, the migration trigger module starts the virtual machine migration program, c, The virtual host selection module selects the virtual host to be migrated, d. the destination node selection module selects the destination node for virtual machine migration, and e. the system establishes a network connection to complete the virtual machine migration.

(1) The source node (server) starts the migration trigger module. When the node load exceeds the load threshold for a long time, it will cause slow service response or network congestion, and even cause the virtual machine to crash or the source node (server) to crash. In this patent, the migration trigger module sets load thresholds for each node's CPU, memory, external storage, and bandwidth according to the specific resource configuration and services provided by the node. When the node load exceeds the specified threshold, the source node (server) starts the migration trigger module and records the trigger type.

(2) The migration trigger module starts the virtual machine migration program. The migration trigger module continuously monitors the node load with a period T. In the continuous N monitoring periods, if the source node load meets the following three conditions: a. The load average value exceeds the set threshold; b. There are at least n (0<n≤N ) times the monitoring value exceeds the threshold, c, using the load prediction technology to obtain the load of the source node (server) in the next N cycles also meets the first two conditions, then the migration trigger module starts the virtual machine migration program on the source node. If the above three conditions cannot be satisfied at the same time, the migration trigger module continues to listen. Among them, the smaller n and N are, the looser the trigger conditions are, and the greater the possibility of invalid migration is; on the contrary, the larger n and N are, the stricter the trigger conditions are, which may cause system performance degradation or even system crash due to heavy load more sex

(3) The virtual host selection module selects the virtual host to be migrated. The virtual host selection module selects the virtual host to be migrated mainly considering the following two aspects: a. The computing resources occupied by the migrated virtual host must be greater than the part where the node load exceeds the load threshold, that is, the migrated node load can return to within the load threshold; b. The migration time and downtime (down time) should be as short as possible, and the amount of data transfer required for the completion of the migration should be as small as possible. According to the type of migration trigger, the type of service provided by the virtual machine, and the impact of migration on system load and service requests, this policy assigns a migration cost weighted coefficient (h _c , h _m , h _s , h _b ), where h _c +h _m +h _s +h _b =1. The specific steps are as follows:

(1) The virtual host selection module detects the computer resources actually used by all virtual machines {VM|VM1, VM2, ...} on the node, among which the resources (CPU, memory, external storage, network bandwidth) of the virtual machine VMi (VMi∈VM) ) is recorded as: (VMi _c , VMi _m , VMi _s , VMi _b ).

(2) The virtual host selection module screens out all virtual machines that can restore the load of the source node (server) to within the threshold after relocation according to the load size and trigger type of the source node (server) exceeding the threshold.

(3) The virtual host selection module calculates the migration cost weighted value H of all selected virtual machines. The migration cost weight value H _i of VMi =VMi _c *h _c +VMi _m *h _m +VMi _s *h _s +VMi _b *h _b .

(4) Select the virtual machine with the smallest H value as the virtual host for migration.

(4) The destination node selection module selects a destination node for virtual machine migration. After receiving the relevant information about the virtual host provided by the virtual host selection module, the destination node selection module selects the migration destination node according to the demand of the virtual host for computing resources. The selection of the destination node is mainly considered from the following two aspects: a. The idle resources of all the resources of the destination node must meet the resource requirements of the migrated virtual host. Idle resources refer to the remaining resources after subtracting the used resources from the load threshold of the destination node. Wherein, the used resource amount is similar to the calculation method of obtaining the load of the source node (server) in step (2), which is obtained by first counting and predicting the system load values of M periods and then calculating the average value. b. In the set of nodes that meet the conditions, the node with the lower comprehensive load is more likely to receive virtual machine migration, and at the same time, service conflicts caused by the clustering effect of multiple migration services should be avoided. Based on the above considerations, this patent assigns corresponding weight values to node (server) resources (CPU, memory, external storage, network bandwidth) that meet the conditions: (d _c , d _m , d _s , d _b ), where d _c + d _m +d _s +d _b =1.

The specific steps are as follows:

(1) The destination node selection module obtains the free resource size of all nodes {DN|DN1, DN2, ...} through the load monitoring technology similar to the observation first and prediction in step (2), and assigns the node DNi (DNi∈DN) The occupancy of resources (CPU, memory, external storage, network bandwidth) is recorded as: (DNi _c , DNi _m , DNi _s , DNi _b ).

(2) According to the principle that the amount of free resources of the destination node must be greater than the amount of resources required by the migrated virtual host, the destination node selection module screens out all nodes (servers) that can accept the virtual host.

(3) The destination node selection module calculates the selected node (server) as the weighted value D of the virtual machine migration destination node, wherein DNi is used as the weighted value of the destination node D _i = DNi _c * d _c + DNi _m * d _m +DNi _s *d _s +DNi _b *d _b .

(4) The destination node selection module calculates the weighted probability value of each node that may become a destination node: p _i =D _i /ΣD _i , and selects a node from the selected node set as the destination node by means of probability selection. Among them, the node with the larger Di value is more likely to be selected as the destination node.

(5) During the execution of the virtual machine migration program, the destination node selection module will lock the source node and the destination node, and use these two nodes as the migration destination nodes of other virtual machine migration programs. The execution of the secondary virtual machine migration program ends.

(5) The source node (server) establishes a network connection with the destination node and completes the migration of the virtual machine.

Description of drawings

Figure 1 Schematic diagram of the decision-making method and control module for cloud computing virtual machine migration

Figure 2 Flowchart of Cloud Computing-oriented Virtual Machine Migration

Figure 3 The decision-making method and control module instance diagram for cloud computing virtual machine migration

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiments of the present invention. Obviously, the described embodiments are only the embodiments of the present invention. Some examples, but not all examples. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

In order to specifically illustrate the cloud computing-oriented virtual machine migration decision-making method and control module, an embodiment as shown in FIG. 3 is given here. In this embodiment, the system consists of a virtual machine migration control module that adopts centralized deployment and four servers, which are source node, node 1, node 2, and node 3 (IPs are: 10.108.36.100, 10.108.36.101 , 10.108.36.102, 10.108.36.103). The source server system is a dual-core x86 system with a CPU frequency of 2.4GHz and a CPU load threshold of 68%. There are three virtual machines (VM1, VM2, VM3) running on the source server.

In the first step, as shown in Figure 3(a), at time t ₀ , the CPU load of the source server exceeds the set threshold of 68%, and the migration trigger module is triggered to start and records that the trigger type is CPU trigger.

In the second step, as shown in Figure 3(a), the migration trigger module continuously monitors for 20 periods with a period of 0.5s. The migration trigger module monitors that the average CPU load of the source server is 83% (greater than the threshold of 68% and exceeding 15%), and the load value is greater than the load threshold for more than 10 times; in addition, through the load prediction technology, the source server will The node CPU load also meets the above two conditions, and the migration trigger module starts the virtual machine migration program and triggers the virtual host selection module.

The third step, as shown in Figure 3 (b), the virtual host selection module monitors the usage of three virtual machine computing resources (CPU, memory, external storage, network bandwidth) as (14%*4.8GHz, 2.0G, 2.5T, 1.2GB/s), (25%*4.8GHz, 2.5G, 3.5T, 1.5GB/s), (30%*4.8GHz, 3.0G, 3.7T, 1.6GB/s), where VM2 and The average CPU load of VM3 is greater than 15%, which meets the preliminary conditions for being a virtual host. Let migration cost weighting coefficients (h _c , h _m , h _s , h _b ) be (0.1, 0.3, 0.4, 0.2), the migration cost weighted value H ₂ of virtual machine VM2 = 25%*4.8*0.1+2.5 *0.3+3.5*0.4+1.5*0.2＝2.57, similarly, the weighted value H ₃ of the migration cost of the virtual machine VM3＝2.84; finally, the virtual host selection module selects VM2 as the virtual host to be migrated, and simultaneously triggers and converts VM2 The relevant information informs the destination node selection module.

The fourth step, as shown in Figure 3(c), the destination node selection module extracts the resource vacancy information of all other nodes, which are node 1 (1.0GHz, 2.0G, 4T, 2.0GB/s), node 2 (2.0GHz, 3.0G, 4.3T, 2.5GB/s), node 3 (2.5GHz, 3.5G, 5.0T, 2.5G B/s), according to the VM2 related information provided by the virtual host selection module, due to the CPU and memory remaining value of node 1 None of them can meet the needs of VM2, so Node 2 and Node 3 are preliminarily screened out as alternative destination nodes, and (d _c , d _m , d _s , d _b ) are set to (0.3, 0.3, 0.2, 0.2); the destination node selection According to the weight value calculation strategy, the module obtains the weight value D of node 2 as the destination node as D2=2.0*0.3+3.0*0.3+4.3*0.2+2.5*0.2=2.86, similarly, D3=3.3. According to p ₂ =45% and p ₃ =55% obtained from the weighted probability value pi=Di/ΣDi, the destination node selection module finally selects node 3 as the destination server according to the probability selection strategy.

Finally, as shown in Figure 3(d), a network connection is established between the source server (10.108.36.100) and node 3 (10.108.36.102), the corresponding computer resources are migrated and the entire virtual machine migration process is completed.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (4)

1. A decision-making method and control module for cloud computing virtual machine migration, characterized in that: the method is composed of a migration trigger module, a virtual host selection module and a destination node selection module. Wherein, the migration trigger module judges whether the system needs to trigger virtual machine migration; the virtual machine selection module decides which virtual host on the source node needs to be migrated; the destination node selection module decides the destination node (server) for migration. 2. The migration trigger module according to claim 1, characterized in that: the module determines whether the source node (server) needs to trigger the virtual machine migration program by setting the system load threshold and using load prediction technology, so that the system can effectively avoid instantaneous load Spikes trigger unnecessary virtual machine migrations. 3. The migration host selection module according to claim 1, characterized in that: the module first screens out all virtual machines that can make the source node (server) run below the load threshold after all migrations, and then calculates the weighted migration cost of the virtual machines value, and finally select the virtual machine with the smallest weighted value as the migration host; this module minimizes the cost of virtual machine migration and saves system resources. 4. The target node selection module according to claim 1, characterized in that: the module selects the target node by using a weighted probability algorithm, and conducts a comprehensive analysis of the resources of the target node, effectively solving the cluster problem caused by the migration of a large number of virtual machines. Aggregate conflicts, and achieve load balancing well, so that resources can be used fairly and reasonably.

Images