CN103577249A - Method and system for virtual machine online migration - Google Patents

Abstract

The present invention provides an online migration method for a virtual machine. The migration priority is set according to the update heat of the virtual machine memory pages and disk blocks. For the virtual machine memory pages and disk blocks to be migrated, the migration priority is from high to high Migrate in the lowest order, copy the data in it to the target end, until the total amount of data to be migrated in the memory and disk of the virtual machine is less than the predefined shutdown threshold; suspend the operation of the virtual machine, and copy the data to be migrated in the memory and disk of the virtual machine One-time copy to the target side. This method can reduce the invalid migration data generated during the virtual machine online migration process, thereby saving network bandwidth, host machine CPU, memory and other resources, and can also timely discover virtual machine online migration operations that cannot be completed or cannot be completed within a reasonable time. So that users or administrators can take corresponding measures, the utilization rate of resources and the quality of cloud services in the cloud computing environment are improved.

Description

Virtual machine online migration method and system

technical field

The invention relates to cloud computing technology, in particular to a virtual machine online migration method.

Background technique

Online migration of virtual machines is an important supporting mechanism for cloud computing resource scheduling. Existing virtual machines widely use the pre-copy technology (PreCopy) to realize online migration, and complete the migration of virtual machine memory and disk data through three rounds of data copying. The first round is called the "full copy" stage, which copies the virtual machine memory and disk data from beginning to end to the migration target; the second round is called the "incremental copy" stage, which copies those The modified and updated memory pages and disk blocks are migrated to the target end until the data volume of the remaining memory pages and disk blocks is less than a predefined threshold (sometimes also called the downtime threshold); the third round is called "downtime copy" stage, that is, after the amount of remaining data to be migrated is less than a certain threshold through "incremental copying", the operation of the virtual machine is suspended, and all remaining data is copied to the migration target at one time. In the existing virtual machine online migration technology, the first round and the second round of data copying will copy the data to be migrated from the beginning to the end indiscriminately, which will cause those memory pages or disk blocks with high update frequency to be deleted. Multiple copies to the migration target, but only the last copy of the same memory page or disk block is the latest valid data, and the previous data copies will be gradually overwritten, so it can be said to be "invalid migration". data".

Existing virtual machine online migration technology results in a large amount of invalid migration data, and the higher the update frequency of virtual machine memory pages and disk blocks, the more invalid migration data. In the cloud computing environment, a large number of invalid data migration operations will waste shared resources such as network bandwidth, CPU, and memory, and affect the quality of cloud services and resource utilization. In addition, if the amount of updated data in the memory and disk of the migrated virtual machine does not converge for a long time, it will cause unpredictable migration completion time on the one hand, and generate a large amount of invalid migration data on the other hand. Technology cannot detect this situation in time and then carry out effective intervention.

Contents of the invention

Therefore, the object of the present invention is to overcome the above-mentioned defects in the prior art, and provide a method for online migration of a virtual machine.

The purpose of the present invention is achieved through the following technical solutions:

On the one hand, the present invention provides a virtual machine online migration method, including:

Step 1. For the virtual machine memory pages and disk blocks to be migrated, migrate them according to the order of their migration priority from high to low, and copy the data in them to the target end until the total amount of data to be migrated in the virtual machine memory and disk Less than the predefined shutdown threshold; among them, the migration priority of virtual machine memory pages and disk blocks is set according to the update heat of virtual machine memory pages and disk blocks, the lower the update heat, the higher the migration priority;

Step 2, suspend the operation of the virtual machine, and copy the data to be migrated in the memory and disk of the virtual machine to the target end at one time.

In the above method, the step 1 may also include dynamically adjusting the migration priority of the virtual machine memory page/disk block according to the change of the update rate of the virtual machine memory page/disk block.

In the above method, the step 1 may also include regularly counting the update frequency of virtual machine memory pages/disk blocks within a recent period of history, and setting according to statistical data of the update frequency of virtual machine memory pages/disk blocks within a recent period of history Update heat of virtual machine memory pages/disk blocks.

In the above method, the update heat can be determined through the following steps:

For a virtual machine memory page/disk block, each data migration cycle records the update of the virtual machine memory page/disk block. If the memory page/disk block has been updated, it is recorded as 1, otherwise it is recorded as 0;

Sum the update status of the virtual machine memory page/disk block recorded in the last n data migration cycles, and use the sum as the current update heat of the memory page/disk block; where the data migration cycle is the virtual machine online Continuous data copy time during migration.

In the above method, the step 1 may further include recalculating the update heat of the virtual machine memory pages and disk blocks before each data migration cycle and adjusting the migration priority of the virtual machine memory pages and disk blocks based on the update heat.

In the above method, the migration priority of the virtual machine disk block can be set to be higher than the migration priority of the virtual machine memory page, and each disk block/memory page is initialized to the same migration priority.

In the above method, the step 1 may further include: detecting in real time whether the virtual machine has an update hotspot data set, the update hotspot data set refers to a collection of virtual machine memory pages or disk blocks whose update heat continues to be high for a period of time ; and detecting whether the sum of the data volume of the updated hotspot data set is higher than a predefined shutdown threshold, and if it exceeds the shutdown threshold, suspending the migration of data blocks in the updated hotspot data set.

In the above method, the step 1 may also include judging whether the online migration of the virtual machine can be completed within a reasonable time based on the prediction of the convergence trend of the total amount of remaining data to be migrated during the virtual machine migration process; If the online migration operation of the virtual machine is completed within the specified time, the user or administrator is prompted to reduce the load on the virtual machine or give up the migration operation.

In the above method, the prediction of the convergence trend of the remaining amount of data to be migrated during the virtual machine migration process may include: recording the remaining amount of data to be migrated after each data migration cycle; checking whether the sequence generally conforms to the monotonically decreasing Trend, if yes, the prediction result is convergent; otherwise, it is non-convergent.

In yet another aspect, the present invention provides a virtual machine online migration system, including:

The iterative copy device is used to migrate the memory pages and disk blocks of the virtual machine to be migrated according to the order of their migration priorities from high to low, and copy the data in them to the target end until the memory pages and disk blocks of the virtual machine are to be migrated The total amount of data is less than the predefined shutdown threshold; among them, the migration priority of virtual machine memory pages and disk blocks is set according to the update heat of virtual machine memory pages and disk blocks. The lower the update heat, the higher the migration priority ;

The shutdown copy device is used to suspend the operation of the virtual machine when the total amount of data to be migrated in the memory and disk of the virtual machine is less than a predefined shutdown threshold, and copy the data to be migrated in the memory and disk of the virtual machine to the target end at one time.

Compared with the prior art, the present invention has the advantages of:

It greatly reduces the invalid migration data generated during the virtual machine online migration process, thereby saving resources such as network bandwidth, host CPU, memory, etc. Users or administrators take corresponding measures to avoid endless waiting for migration operations, and improve resource utilization and cloud service quality in the cloud computing environment.

Description of drawings

Embodiments of the present invention will be further described below with reference to the accompanying drawings, wherein:

1 is a schematic diagram of a sliding historical time window for predicting the update heat of a virtual machine memory page/disk block according to an embodiment of the present invention;

Fig. 2 is a schematic flowchart of the iterative data copy stage in the online migration method of a virtual machine according to an embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below through specific embodiments in conjunction with the accompanying drawings. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

In the traditional virtual machine online migration method described above, memory pages or disk blocks with high update frequency (also called update heat) are copied to the migration target multiple times, while the same memory page or disk block Only the last copy of multiple migration data is the latest valid data, and the previous data copies will be gradually overwritten, resulting in invalid migration data. The higher the update frequency of virtual machine memory pages and disk blocks, the more invalid migration data will be generated.

In one embodiment of the present invention, a virtual machine online migration method is provided. This method sets the migration priority of virtual machine memory pages and disk blocks based on the update heat of virtual machine memory pages and disk blocks. The lower the update heat, the higher the migration priority, and the migration is performed in the order of migration priority from high to low. Virtual machine memory pages and disk blocks. This method lowers the migration priority of memory pages and disk blocks with high update heat, thereby reducing the number of migrations of such pages, thereby reducing invalid migration data. The virtual machine online migration method may include two stages of iterative data copy and shutdown data copy. In the iterative data copy phase, the virtual machine memory pages and disk blocks to be migrated are migrated according to the order of their migration priorities from high to low, and the data in them is copied to the target end until the virtual machine memory and disk blocks are to be migrated When the total amount of data is less than the predefined downtime threshold, it enters the downtime data copy stage. In the downtime data copy stage, the running of the virtual machine is suspended, and the data to be migrated in the memory and disk of the virtual machine is copied to the target end at one time.

Before discussing the specific details of each stage of the method, the following concepts are introduced first: (1) Data migration cycle: refers to the continuous data copying time during the online migration of virtual machines. Since online migration of virtual machines is often limited by specific network bandwidth quotas to share network bandwidth resources reasonably, the amount of data sent to the network interface at one time during data copying cannot exceed the predetermined threshold. When the amount of data sent to the network interface reaches a predetermined threshold, data copying is suspended for a period of time. Because the threshold is fixed, the data migration cycle is also a substantially constant time. (2) Dirty page/dirty disk block: If the virtual machine memory page/disk block has been updated by the guest operating system and has not been migrated to the target, it is called a dirty page/dirty disk block. (3) Update heat: It reflects the update status of virtual machine memory pages/disk blocks. A high update rate of a virtual machine memory page/disk block indicates that it is frequently updated, while a low update rate of a virtual machine memory page/disk block indicates that it is seldom updated. Generally, the update heat of the virtual machine memory is higher than the update heat of the virtual machine disk block. In an example, the update frequency statistics of virtual machine memory pages/disk blocks for a period of time before the virtual machine starts to migrate can be used as the update heat of virtual machine memory pages/disk blocks, so as to set the memory pages and disk blocks to be migrated. The block's migration priority. However, as mentioned above, the virtual machine migration process is not completed once, but may be completed after multiple data migration cycles. During the virtual machine migration process, the update heat of each virtual machine memory page/disk block may also change. Since the update popularity in the future period is closely related to the update situation in the past period of time, in order to set the migration priority more accurately, in another embodiment, during the migration process of the virtual machine, it is possible to regularly collect statistics on the latest historical period The update frequency of virtual machine memory pages/disk blocks, based on the statistical data of the update frequency of virtual machine memory pages/disk blocks in the most recent historical period, predicts the update popularity of virtual machine memory pages/disk blocks in a period of time in the future. The reference basis of the migrated memory pages and disk blocks; and it is also possible to dynamically adjust the migration priority of the virtual machine memory pages/disk blocks as the update heat of the virtual machine memory pages/disk blocks changes during the virtual machine migration process.

In one embodiment of the present invention, a virtual machine memory page/disk block update heat prediction method based on a sliding historical time window is provided. Update popularity. FIG. 1 is a schematic diagram of a sliding historical time window for predicting update heat of memory pages/disk blocks of a virtual machine according to an embodiment of the present invention. As shown in Figure 1, the sliding unit of the sliding historical time window is a data migration cycle. Before the start of each data migration cycle, the update status of each memory page and disk block since the previous cycle will be recorded. For example, if the memory page or disk If the block has been updated in the previous cycle, "1" will be recorded at the latest position of the time series, otherwise "0" will be recorded. Next, slide the window forward one position, discard the oldest update record, absorb the latest update record, recalculate the update heat of the memory page/disk block, and adjust the memory page/disk block according to the change of update heat migration priority. The length W of the sliding historical time window is actually an integer multiple of the data migration period P, that is, W=n*P, where n is a positive integer. The length of the sliding historical time window can be set according to the actual situation of the system. As shown in Figure 1, n=7, assuming that the data migration cycle is 2 seconds, then for a memory page/disk block, the update of the memory page/disk block is recorded every 2 seconds, during which only the memory page/disk block If it has been updated, it is recorded as 1, otherwise it is recorded as 0. The sliding historical time window includes the update record of the memory page/disk block within the last 2*7=14 seconds. The update popularity of the memory page/disk block is calculated based on the update records covered by the sliding historical time window. For example, the update records covered by the sliding historical time window can be summed, and the total value can be used as the update heat of the memory page/disk block. As shown in Figure 1, in the nth data migration cycle, the update heat of the memory page/disk block is 1+0+0+0+1+1+1=4. For another example, the update heat of the memory page/disk block can be set based on the distribution of 0 and 1 in the update records covered by the sliding historical time window. For example, assuming that the update record in the window is 0001111, it means that the memory page/disk block has been accessed more recently, and its update heat can be set to be higher. For another example, assuming that the update record in the window is 1110000, it means that the memory page/disk block has been accessed less recently, and its update heat can be set to be low. It should be pointed out that the statistics of the above updates and the calculation of the update popularity are only for the purpose of illustration, rather than any limitation. Those skilled in the art can also set the update heat of the virtual machine memory page/disk block according to the statistical data of update frequency of the virtual machine memory page/disk block in a recent period of history.

In one embodiment, during the virtual machine migration initialization phase, as described above, the virtual machine memory page/disk block can be calculated based on the update frequency statistics of the virtual machine memory page/disk block for a period of time before the virtual machine starts to migrate The initial value of the update heat to set the initial migration priority of the memory pages and disk blocks to be migrated. Then, in the process of virtual machine migration, use the above-mentioned virtual machine memory page/disk block update heat prediction method based on the sliding historical time window mentioned above to calculate the current virtual machine memory page/disk block update heat and adjust accordingly Migration priority of virtual machine memory pages/disk blocks.

In yet another embodiment, in the virtual machine migration initialization phase, the migration priority of the virtual machine disk block is set to be higher than the migration priority memory page of the virtual machine memory page, and each disk block/memory page is initialized to the same Migration priority.

As mentioned above, the online virtual machine migration method according to the embodiment of the present invention may include two stages of iterative data copy and shutdown data copy. In the iterative data copy phase, the virtual machine memory pages and disk blocks to be migrated are migrated according to the order of their migration priorities from high to low, and the data in them are copied to the target end. Fig. 2 is a schematic flow chart of the iterative data copy stage in the online migration method of a virtual machine according to an embodiment of the present invention. As shown in FIG. 2 , at the start of the migration, judging whether the amount of migrated data reaches the threshold is actually judging whether the data migration cycle is over. The threshold can be flexibly set according to actual requirements and system conditions. If the threshold is reached, it means that the data migration cycle has ended, and it is necessary to wait for a fixed period of time before starting the next data migration cycle. If the threshold is not reached, select the memory page/disk block with the highest migration priority from the memory pages and disk blocks to be migrated, and copy the data in the selected memory page/disk block to the target end. There are many ways to select memory pages/disk blocks to migrate. For example, the memory pages and disk blocks to be migrated can be formed into a queue according to the migration priority from high to low. For another example, the memory pages to be migrated may form a queue according to the migration priority from high to low, and the disk blocks to be migrated may be formed into a queue according to the migration priority from high to bottom. In yet another example, the memory pages and/or disk blocks with the same migration priority can be set in the same priority queue, and the memory pages/disk blocks with the same migration priority at different priorities are at different priorities, and then from High to low to scan each priority queue. After the data in the selected memory page/disk block is copied to the target end, the memory page/disk block may be removed from the set or queue of virtual machine memory pages and disk blocks to be migrated. If the memory page/disk block is updated again during the migration of the virtual machine, the memory page/disk block may be added to the set or queue of the virtual machine memory page and disk block to be migrated again.

Continuing to refer to Figure 2, if the amount of migrated data reaches the threshold, it means that the data migration cycle has ended, and it is necessary to wait for a fixed time before starting a new data migration cycle to continue the above migration process. The above process is repeated until the total amount of data to be migrated in the memory and disk of the virtual machine is less than the predefined shutdown threshold, and the shutdown data copy stage is entered. In yet another embodiment, the virtual machine online migration method can also recalculate the update heat of the virtual machine memory pages and disk blocks before the start of each data migration cycle, and adjust the migration of the virtual machine memory pages and disk blocks based on the update heat priority. For example, using the above-mentioned virtual machine memory page/disk block update heat prediction method based on the sliding historical time window, record the update status of each memory page and disk block since the previous cycle before the start of each data migration cycle , if the memory page or disk block was updated by the guest operating system in the previous cycle, record "1" at the most recent position in the time series, otherwise record "0". Next, slide the window forward by one position, cover the nearest position, discard the record of the longest position, then recalculate the update heat of the memory page/disk block, and adjust the memory page/disk block according to the change of update heat migration priority. Taking the method of multiple priority queues mentioned above as an example, if the memory page/disk block is a "dirty page/dirty disk block" and the page/disk block is not in the priority queue, it will be The migration priority of the page/disk block is added to the corresponding priority queue; if the page/disk block is already in the priority queue, and its latest migration priority is not equal to the migration priority of the previous cycle, Then move it to the corresponding priority queue; if the page is not a "dirty page/dirty disk block", remove it from the priority queue (if it was in the priority queue before).

Through the above iterative data copy process, the amount of data to be migrated in the memory or disk of the virtual machine can be less than the preset shutdown threshold, so that it can be copied to the migration target in a short enough time, and then it can enter the shutdown data copy stage . In the downtime data copy stage, the running of the virtual machine is suspended, and the data to be migrated in the memory and disk of the virtual machine is copied to the target end at one time. The downtime threshold can be set according to actual requirements and system resource conditions.

In yet another embodiment, the method for online migration of a virtual machine can also detect in real time whether there is an updated hotspot data set for the virtual machine. The so-called update hotspot data set means that the update heat is continuously high for a period of time (for example, it can be higher than a predefined threshold; for example, the average update heat of a certain data block within a period of time is greater than the size of the sliding window half) of virtual machine memory or a collection of disk data blocks. When an updated hot data set is found, it will further detect whether the size of the data set is higher than the predefined shutdown threshold. If the total amount of data in the updated hotspot data set exceeds the shutdown threshold, the migration of this part of virtual machine memory pages and disk blocks (that is, the data blocks in the updated hotspot data set) will be suspended until the update heat of some data blocks decreases to update the hotspot data The size of the set is smaller than the amount of data allowed by the downtime copy, so that the virtual machine migration enters the downtime migration stage, and this part of the data is migrated at one time. In this way, compared with the indiscriminate data copy mode in the traditional virtual machine online migration method, resources such as network bandwidth, CPU, and memory can be greatly saved, and at the same time, the migration time of the virtual machine will not be prolonged.

In the cloud computing environment, virtual machine online migration is the basic method of cloud resource scheduling, and resource scheduling often has strict timeliness requirements, so there is a higher requirement for the completion time of virtual machine online migration. If the virtual machine online migration cannot be completed within a predetermined and reasonable time, effective resource scheduling cannot be completed, thereby affecting the quality of cloud services. The traditional virtual machine online migration process is opaque to users or administrators. After submitting an online migration operation, the administrator cannot predict whether the operation will be successful or how long it will take to complete.

Therefore, in another embodiment of the present invention, the virtual machine online migration method can also predict whether the virtual machine can complete the online migration operation, and if it is predicted that the online migration operation cannot be completed within a reasonable time, the user or administrator is prompted to reduce the load on the virtual machine or abort the migration. Therefore, online migration operations of virtual machines that cannot be completed or cannot be completed within a reasonable time will be promptly fed back to the user or administrator, so that corresponding countermeasures can be taken to meet the timeliness requirements of cloud resource scheduling. The prediction of whether the virtual machine can complete online migration within a reasonable time is mainly based on the prediction of the convergence trend of the remaining data to be migrated during the virtual machine migration process. If the predicted trend can be converged within a reasonable time, the judgment result is yes Complete the online migration, otherwise the judgment result is that the online migration cannot be completed. The specific prediction algorithm can be, for example: by recording the total amount of remaining data to be migrated after the end of each data migration cycle, check whether the numerical sequence generally conforms to a monotonically decreasing trend. If so, the prediction result is convergent; otherwise, it is non-convergent .

In yet another embodiment of the present invention, a virtual machine live migration system is also provided, including an iterative copy device and a shutdown copy device. Wherein, the iterative copy device is used to migrate the memory pages and disk blocks of the virtual machine to be migrated according to the order of their migration priorities from high to low, and copy the data therein to the target end until the memory pages and disk blocks of the virtual machine are to be migrated. The total amount of migrated data is less than the predefined downtime threshold. The shutdown copy device is used to suspend the operation of the virtual machine when the total amount of data to be migrated in the memory and disk of the virtual machine is less than a predefined shutdown threshold, and copy the data to be migrated in the memory and disk of the virtual machine to the target end at one time.

Through the above-mentioned specific embodiments of the present invention, it is possible to reduce the invalid migration data generated during the online migration of virtual machines, thereby saving resources such as network bandwidth, host machine CPU, memory, etc. The online migration operation of the virtual machine allows users or administrators to take corresponding measures, avoids endless waiting for migration operations, and improves resource utilization and cloud service quality in the cloud computing environment.

Although the present invention has been described in terms of preferred embodiments, the present invention is not limited to the embodiments described herein, and various changes and changes are included without departing from the scope of the present invention.

Claims (10)

1. the online moving method of virtual machine, comprising:

Step 1, for the virutal machine memory page and disk block to be migrated, according to its migration priority order from high to low, move, copy data wherein to destination end, until data total amount to be migrated is less than predefined shutdown threshold in virutal machine memory and disk; Wherein, the migration priority of the virutal machine memory page and disk block is according to the renewal temperature of the virutal machine memory page and disk block is arranged, and upgrades temperature lower, and migration priority is higher;

Step 2, suspends the operation of virtual machine, by the disposable destination end that copies to of data to be migrated in virutal machine memory and disk.

2. method according to claim 1, described step 1 also comprises the migration priority of dynamically adjusting the virutal machine memory page/disk block along with the change of the renewal temperature of the virutal machine memory page/disk block.

3. method according to claim 1 and 2, described step 1 also comprises the virutal machine memory page/disk block renewal frequency of the nearest phase of history of regular statistics in the time, and the page/disk block of the virutal machine memory in time renewal frequency statistics arranges the renewal temperature of the virutal machine memory page/disk block according to nearest phase of history.

4. method according to claim 3, described renewal temperature is to come through the following steps to determine:

For a virutal machine memory page/disk block, each Data Migration periodic recording is the update status of this virutal machine memory page/disk block once, if this memory pages/disk block was updated, is just designated as 1, otherwise is designated as 0;

The update status of this virutal machine memory page/disk block that nearest n Data Migration cycle recorded is sued for peace, and the current renewal temperature using gained summation as this memory pages/disk block; Wherein the Data Migration cycle is the continuous data copy time in the online transition process of virtual machine.

5. method according to claim 4, described step 1 is also included in each Data Migration cycle and recalculates renewal temperature and the migration priority based on upgrading temperature and adjust the virutal machine memory page and disk block of the virutal machine memory page and disk block before starting.

6. method according to claim 1, wherein the migration priority of virtual machine disk block is set to the migration priority higher than the virutal machine memory page, and each disk block/memory pages is initialized as identical migration priority.

7. method according to claim 1, described step 1 also comprises: detect in real time virtual machine and whether have renewal hot spot data collection, described renewal hot spot data collection refers to that in a period of time, it upgrades the set that temperature continues the higher virutal machine memory page or disk block composition; And whether the data volume summation that detects this renewal hot spot data collection is higher than predefined shutdown threshold, if surpass shutdown threshold, suspends the migration of upgrading the concentrated data block of hot spot data.

8. method according to claim 1, described step 1 also comprises based on to remaining the prediction of the convergent tendency of data total amount to be migrated in virtual machine (vm) migration process, judges whether to move online completing virtual machine within reasonable time; If determined, can not complete the online migration operation of virtual machine within reasonable time, point out user or keeper to reduce the load in virtual machine or abandon this migration operation.

9. method according to claim 8, wherein comprises remaining the prediction of the convergent tendency of data total amount to be migrated in virtual machine (vm) migration process: record the residue data total amount to be migrated after each Data Migration end cycle; Check whether this sequence meets the trend of monotone decreasing substantially, if so, predict the outcome as restraining; Otherwise for not restraining.

10. the online migratory system of virtual machine, comprising:

Iteration copy device, be used for for the virutal machine memory page and disk block to be migrated, according to its migration priority order from high to low, move, copy data wherein to destination end, until data total amount to be migrated is less than predefined shutdown threshold in virutal machine memory and disk; Wherein, the migration priority of the virutal machine memory page and disk block is according to the renewal temperature of the virutal machine memory page and disk block is arranged, and upgrades temperature lower, and migration priority is higher;

Shut down copy device, for suspend the operation of virtual machine when virutal machine memory and disk data total amount to be migrated are less than predefined shutdown threshold, by the disposable destination end that copies to of data to be migrated in virutal machine memory and disk.

Images