CN102968344A - Method for migration scheduling of multiple virtual machines - Google Patents

Abstract

The invention relates to a scheduling method for multi-virtual machine migration. In the multi-virtual machine migration process, a strategy for selecting a target host machine for virtual machine migration and a strategy for planning the migration sequence of multi-virtual machines include: Get the target clustering information divided by the virtual machine; get the information of the host machine in the system and the information of all the virtual machines running on it, and calculate the appropriate mapping from the clustering of the virtual machine to the target host machine, so as to select the appropriate target host machine for the virtual machine; collect The load of the host machine; according to the load and migration mapping, arrange the migration sequence of multiple virtual machines and control the execution. The invention controls the total times and the total time of multi-virtual machine migration from the perspective of system macro, so that the migration efficiency is higher. Since the algorithm idea of the present invention has certain universality, and because the clustering of the virtual machine is used as the input, which is separated from the calculation process of the algorithm, it has relatively wide applicability.

Description

A method for multi-virtual machine migration scheduling

technical field

The present invention relates to a scheduling method for multi-virtual machine migration, and mainly relates to a strategy for selecting a target host machine for virtual machine migration and a strategy for planning the migration sequence of multi-virtual machines during the multi-virtual machine migration process. It belongs to the technical field of virtualization and cloud computing.

Background technique

At present, with the widespread application of virtualization technology in cloud computing and other fields, a large number of virtual machines run on servers and data centers, and the management requirements for virtual machines are gradually highlighted. The current management needs of virtual machines include many aspects. For example, optimize virtual machine deployment to reduce server power consumption; optimize virtual machine deployment to increase physical CPU utilization; migrate virtual machines with high memory similarity to the same host, and use content-based page sharing technology (CBPS), Allow multiple virtual machines to share the physical memory of the host machine to save the memory resources of the host machine, improve the memory usage efficiency of the host machine, and so on. The above-mentioned virtual machine management requirements all involve multi-virtual machine migration. The migration process of multiple virtual machines has a certain overhead. Therefore, for the migration of multiple virtual machines in different situations, the research of various scheduling algorithms has become a research hotspot in the field of virtualization and cloud computing.

The current virtual machine scheduling algorithms include scheduling for the load balance of the host machine, and scheduling for the resource utilization of the host machine. There are also some scheduling algorithms that migrate virtual machines with similar memory to the same host machine. to share host memory. These scheduling algorithms, no matter they are static algorithms or dynamic algorithms, are mainly focused on solving "how to achieve the scheduling goal", and do not pay attention to how to reduce the overhead during the migration process of multiple virtual machines.

With the large-scale application of cloud computing, the number of virtual machines in cluster servers or data centers is increasing, and the cost of migration involving multiple virtual machines is increasing. If the multi-virtual machine scheduling method can effectively reduce the total number of migration times of multi-virtual machine migration, and in the case of the same number of migration times, thereby reducing the total migration time, it will directly reduce the total number of downtime and downtime of virtual machines in the entire system. Time is of great significance for improving the overall service quality of the system. The present invention aims at multi-virtual machine migration and proposes a scheduling method to reduce the total migration times and migration time of the system.

Contents of the invention

The purpose of the present invention is to propose a scheduling method for multi-virtual machine migration. The input of the method is the clustering of virtual machines. The function of the method is to select a suitable host for each virtual machine cluster, and design a suitable migration sequence for all virtual machines that need to be migrated. The scheduling method minimizes the total number of migrations of multiple virtual machines and minimizes the total migration time under the same migration times, thereby reducing the total overhead of the system during the migration process and improving the service quality of the system.

According to one aspect of the present invention, a scheduling method for multi-virtual machine migration is provided, which is characterized by comprising:

Step 1. The user enters the target clustering of multiple virtual machines on the master control node machine. The division of target clusters is determined by the user according to the target of multi-virtual machine migration. Virtual machines in the same cluster will eventually be migrated to the same target host.

Step 2. The master control node obtains information of all host nodes in the system and information of all virtual machines running on the host through feedback from the host node by sending a request to the host node.

Step 3. Based on the information obtained in Step 1 and Step 2, the master control node invokes the "target host selection algorithm" to select a suitable target host for virtual machine clustering and form a migration map from the virtual machine to the target host.

Step 4. The master control node sends a request to the host node to collect the load status of each host node.

Step 5. The master control node, according to the migration map determined in step 3 and the host load information collected in step 4, arranges the migration sequence for multiple virtual machines according to the "migration order determination algorithm", and issues instructions to control the relevant hosts Nodes perform migrations sequentially.

Description of drawings

FIG. 1 is a system overall framework diagram of the multi-virtual machine migration scheduling method of the present invention.

FIG. 2 is a specific flow chart of the multi-virtual machine migration scheduling method of the present invention.

Fig. 3 is a specific flowchart of an implementation example of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention more clearly, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

The main idea of the present invention is to select a suitable target host machine and arrange a suitable migration order for the migration of multiple virtual machines in the system under the condition of a given migration target. Migration time to improve the overall quality of service of the system. This has a very wide application value in the current cloud computing era, when virtual machines are widely used.

Technical scheme of the present invention is as follows:

A scheduling method for multi-virtual machine migration, the overall framework of the system is shown in FIG. 1 . Among them, the host node is used as the machine carrier for the virtual machine to run, and multiple virtual machines are run through the virtual machine monitor, which is the main place for multi-virtual machine migration. As the running carrier of the scheduling method, the master control node reads the virtual machine target clustering information input by the user, and obtains the information of each host node (including the configuration information of the host machine and the information of the virtual machines running on it, etc.). According to these known conditions, the target host selection algorithm is invoked to output a suitable multi-virtual machine migration strategy, and the control module sends virtual machine migration instructions to each host node according to the strategy to control the multi-virtual machines among the nodes. migrate. The scheduling method for multi-virtual machine migration is mainly embodied on the main control node, and the specific flow of the scheduling method is shown in FIG. 2 . The method includes the following steps:

Step 201. The user inputs the target clustering of multiple virtual machines on the master control node machine. The division of target clusters is determined by the user according to the target of multi-virtual machine migration. Virtual machines in the same cluster will eventually be migrated to the same target host.

Step 202. The master control node obtains information of all host nodes in the system and information of all virtual machines running on the host through feedback from the host node by sending a request to the host node.

Step 203. According to the information obtained in steps 201 and 202, the master control node invokes the "target host selection algorithm" to select a suitable target host for virtual machine clustering and form a migration map from the virtual machine to the target host.

Step 204. The master control node sends a request to the host node to collect the load status of each host node.

Step 205. The master control node, according to the migration map determined in step 203 and the host load information collected in step 204, arranges the migration sequence for multiple virtual machines according to the "migration sequence determination algorithm", and issues instructions to control the relevant hosts Nodes perform migrations sequentially.

Wherein, the target clustering of virtual machines input in step 201 is provided by the user. According to the target of virtual machine migration, the user sets the clustering of virtual machines, and the virtual machines in the same cluster will eventually be migrated to the same host machine. In this way, the wide applicability of the scheduling method of the present invention is guaranteed. Users can divide virtual machine clusters according to different application scenarios, for example, according to the memory similarity between virtual machines, according to the resource usage characteristics of virtual machines, according to the requirements of load balancing, and so on.

Wherein, in step 202, the information collected by the master control node from the host node specifically includes: various hardware configuration information of the host, the maximum number of virtual machines that the host can carry, system versions of all virtual machines currently running on the host, CPU configuration, memory size, and virtual machine number. On the master control node side, fill in the above information for each virtual machine data entry, and record the host number in the entry. The host number collected at this time is the source host number of the virtual machine. In the master node, a record is also maintained for each host.

Among them, in step 203, the "target host selection algorithm" selects a suitable target host for the virtual machine target clustering, and forms a migration map from the virtual machine to the target host. This process is divided into two steps. First, use the traversal method to select the set of mappings that minimize the total number of migrations of the system from the mapping solution space of virtual machine clusters to host machines (because there may be multiple virtual machine clustering to host mapping methods, with total number of migrations of the same). Second, in the set of mappings with the least total migration times, select the mapping with the smallest total migration flow. This is because, even for the same number of migrations, the traffic will be different due to the different virtual machines being migrated. The mapping between virtual machine clusters and host machines formed in this step directly determines the target host machine to which the virtual machines in each cluster should be migrated, that is, determines the migration path of all virtual machines, but does not determine the multi-virtual machine The sequence of machine migration.

Wherein, in step 204, the load condition of the host computer is collected, which is completed by the master control node. The master control node sends a request to each host node, and each host node feeds back its usage data such as cpu, memory, and number of virtual machines, and is recorded in the corresponding table entry by the master control node.

Wherein, in step 205, the migration map calculated in step 203 is further processed according to the load information collected in step 204. According to certain principles, arrange the sequence of multi-virtual machine migration, and issue instructions accordingly to control each host node to perform migration in order. The basis can be varied, such as moving in first based on the least-loaded hosts, or moving out first based on the highest-loaded hosts. At the same time, the requirements of network and parallel processing can also be taken into account.

3. Advantages and effects

The present invention is a method for multi-virtual machine migration scheduling. Compared with the prior art, it has the following main advantages and effects: (1) From the perspective of the macro system, the total number and total time of multi-virtual machine migration are controlled , making migration more efficient and filling the gaps in related fields. (2) Since the algorithm idea of the present invention has a certain degree of universality, and at the same time, the scheduling method uses virtual machine clustering as input, which is separated from the calculation process of the algorithm. For different migration purposes, only the input virtual machine clustering needs to be modified. , so that the scheduling method of this method has wider applicability. For example, it can be used for virtual machine migration based on memory similarity, or virtual machine migration based on load balancing. The idea of the algorithm can even be used in path selection and other fields beyond virtual machine migration.

The hardware equipment required by the present invention is shown in Fig. 1, and the host machine is required to support the virtualization technology, and the host machines are required to be interconnected through the network, so that they can communicate directly without going through an intermediate node. There is no requirement for a virtual machine monitor, and current popular mainstream products are acceptable, the difference is only in modifying individual function interfaces. To simplify the migration process, it is best to require the host and its hypervisor to be isomorphic. Of course, heterogeneous is also possible, the difference is that the migration process will be slightly complicated, which does not involve the core part of the present invention.

A detailed description will be given below with an example.

The scheduling method of the present invention is used in this example, and the specific steps are divided into five steps, as shown in FIG. 3 .

Step 301. The user inputs the target clustering of multiple virtual machines on the master control node machine. The division of target clusters is determined by the user according to the target of multi-virtual machine migration. Virtual machines in the same cluster will eventually be migrated to the same target host. In this example, according to the memory similarity, the virtual machines with large memory similarity are divided into a cluster, so that they can be migrated to the same host machine, so as to facilitate the memory sharing of the host machine.

Step 302. The master control node obtains information of all host nodes in the system and information of all virtual machines running on the host through feedback from the host node by sending a request to the host node, and records them in the master control node.

Step 303. Based on the information obtained in steps 301 and 302, the master control node invokes the "target host selection algorithm" to select a suitable target host for virtual machine clustering and form a migration map from the virtual machine to the target host. "Target host selection algorithm" selects a suitable target host for virtual machine clustering and forms a migration map from a virtual machine to a target host. The idea of the "target host selection algorithm" is: among all the mappings from virtual machines clustered to the target host (called "migration mapping"), use the stack and use traversal to select the mapping with the least number of migrations , forming a set, and then from this set, select a mapping that minimizes the total migration time, that is, as the final target host selection strategy. A mapping of virtual machine clusters to target hosts, the idea is to map virtual machine clusters to the source hosts of the clustered virtual machines first (for a mapping f, select one of the hosts ), if the corresponding host machine has been fully occupied by the previous clustering, select x units from the remaining host machines that have not been mapped to (x can be set), and then try one by one. For the last processed cluster, all the remaining idle hosts can be tried to map once. The value of x can be set, which increases the flexibility of the algorithm. The larger the value of x, the larger the mapping set F, the greater the possibility of the optimal solution, and the more traversal times of the algorithm, the greater the complexity.

Step 304. The master control node sends a request to the host node to collect the current load conditions of each host node.

Step 305. The master control node, according to the migration map determined in step 303 and the host machine load information collected in step 304, arranges the migration order for multiple virtual machines according to the "migration order determination algorithm", and issues instructions to control the relevant host machines Nodes perform migrations sequentially. In step 305 of this example, first generate the migration paths of all virtual machines according to the migration mapping selected in step 303, and each migration path is expressed as a triplet of <virtual machine number, source host machine number, target host machine number>, and then Sorting is performed based on the principle that hosts with the least load are migrated in first. The sorting method of the migration path is to sort the triplets according to the order of the current load from small to large, using "target host" as the key. Finally, according to the sorted result, perform the sorting task represented by the triplet. In order to ensure the real-time performance of the system, it can be set that after each host has been migrated in once, the current load status will be updated once, and then sorted again according to the updated host load, and then continue to perform the rest of the migration.

Finally, it should be noted that the above embodiments are only used to illustrate and not limit the technical solutions of the present invention, although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the present invention can still be modified Or an equivalent replacement, any modification or partial replacement without departing from the spirit and scope of the present invention shall fall within the scope of the claims of the present invention.

Claims (6)

1. the dispatching method of multi-dummy machine migration is characterized in that comprising:

Step 1. the user inputs the target cluster of multi-dummy machine on master control node machine.The division of target cluster is determined according to the target of multi-dummy machine migration by the user.The virtual machine of same cluster is moved on same the target host the most at last.

Step 2. main controlled node by host node feedback, gets access to the information of whole host nodes in the system by sending request to the host node, and the information of the whole virtual machines that move on the host.

Step 3. according to step 1 and the obtained information of step 2, main controlled node call " target host selection algorithm ", be the suitable target host of virtual machine Clustering and selection, form virtual machine to the migration mapping of target host.

Step 4. main controlled node sends request to the host node, collects the loading condition of each host node.

Step 5. main controlled node, the host load information that the migration mapping of determining according to step 3 and step 4 are collected, according to " migration sequentially determine algorithm ", be multi-dummy machine arrangement migration order, and send instruction, control each relevant host node and carry out in order migration.

2. the dispatching method of a kind of multi-dummy machine migration according to claim 1, it is characterized in that: " user is on master control node machine; the target cluster of input multi-dummy machine " described in the step 1 refers to: the target cluster of the virtual machine of input is provided by the user.The user sets the cluster of virtual machine according to the target of virtual machine (vm) migration, and the virtual machine of same cluster is migrated to same host the most at last.This mode has guaranteed the broad applicability of dispatching method of the present invention.The user can come the cluster of partition virtual machines according to different application scenarioss, such as according to the internal memory similarity between virtual machine, and according to the resource operating characteristic of virtual machine, according to the requirement of load balancing, etc.

3. the dispatching method of a kind of multi-dummy machine migration according to claim 1, it is characterized in that: described in the step 2 " main controlled node is by sending request to the host node; feed back by the host node; get access to the information of whole host nodes in the system; and the information of the whole virtual machines that move on the host " refers to: the information that main controlled node is collected to the host node, specifically comprise: the various hardware configuration informations of host, the maximum virtual machine quantity of host's function carrying, the system version of whole virtual machines of the current operation of host, the CPU configuration, memory size, the virtual machine numbering.At the main controlled node end, give each virtual-machine data list item, to fill in above information, and the host numbering also is recorded in the list item, the host numbering of collecting this moment namely is the sourcesink main frame numbering of virtual machine.In main controlled node, can safeguard a record for each host simultaneously.

4. the dispatching method of a kind of multi-dummy machine migration according to claim 1, it is characterized in that: described in the step 3 " according to step 1 and the obtained information of step 2; main controlled node calls ' target host selection algorithm '; be the suitable target host of virtual machine Clustering and selection; form virtual machine to the migration of target host shine upon " refer to: " target host selection algorithm ", be the suitable target host of virtual machine target Clustering and selection, form virtual machine to the migration mapping of target host, this process is divided into 2 and goes on foot.At first, adopt the traversal mode, the mapping solution space from the virtual machine cluster to host, selection makes the set (because may exist a plurality of virtual machine clusters to the mapping mode of host, having identical migration total degree) of the mapping of the total migration least number of times of system.Secondly, in the mapping set of these gross migration least number of times, select the mapping of migration total flow minimum.This is because even identical migration number of times, because the virtual machine of migration is different, flow also can be different.The virtual machine cluster that this step forms has directly determined the target host that the virtual machine in each cluster should be moved to the mapping of host, has namely determined the migration path of whole virtual machines, but does not determine the sequencing of multi-dummy machine migration.

5. the dispatching method of a kind of multi-dummy machine migration according to claim 1, it is characterized in that: described in the step 4 " main controlled node sends request to the host node; the loading condition of collecting each host node " refers to: collect the loading condition of host, finished by main controlled node.Main controlled node sends request to each host node, behind the usage datas such as separately cpu of each host node feedback, internal memory, virtual machine quantity, is recorded in the corresponding list item by main controlled node.

6. the dispatching method of a kind of multi-dummy machine migration according to claim 1, it is characterized in that: " the main controlled node; the host load information that the migration mapping of determining according to step 3 and step 4 are collected; with " migration determines that sequentially algorithm "; be multi-dummy machine arrangement migration order; and send instruction; control each relevant host node and carry out in order migration " refers to: the load information of collecting according to step 4, the further processing of migration mapping that step 3 is calculated described in the step 5.According to certain principle, arrange the order of multi-dummy machine migration, and send accordingly instruction, control each host node and carry out in order migration.The principle of foundation can be various, preferentially moves into such as foundation minimum load host, and perhaps the maximum load host is preferentially moved out etc.Simultaneously, can also take into account the requirement of network and parallel processing.

Images