CN103729249B - Method and computer system for memory management of virtual machine - Google Patents

Abstract

A memory management method and a computer system for a virtual machine system. The memory management method includes: first, setting a first threshold value by a processor. Then, in a first adjustment phase, the processor sets a balloon target to the allocated virtual memory size and gradually decreases the balloon target by a first decrement value according to a swap in/refault detection result. The processor detects at least one swap in or refault event to generate a swap in/refault detection result. In a cooling phase, the processor stops decreasing the balloon target according to the swap in/refault detection result. In a second adjustment phase after the cooling phase, the processor gradually decreases the balloon target by a second decrement value. The second decrement value is less than the first decrement value, and the balloon target is not less than the first threshold value.

Description

Method and computer system for memory management of virtual machine

technical field

The present disclosure relates to technologies for memory management of virtual machines.

Background technique

Computer virtualization is a technology that involves creating a virtual machine like a physical computer with an operating system, and the architecture of computer virtualization is generally defined by the ability to simultaneously support multiple operating systems on a single physical computer platform. For example, a computer running Microsoft Windows can host a virtual machine with a Linux operating system. When the virtual machine is regarded as a guest machine (guest machine), the host is the above-mentioned physical machine where virtualization occurs. A hypervisor (virtual machine monitor, VMM) is a software layer that virtualizes hardware resources and presents a virtual hardware interface to at least one virtual machine. A hypervisor is similar to the way a traditional operating system manages hardware resources for processing and performs certain management functions with respect to an executing virtual machine. A virtual machine may be referred to as a "guest" and an operating system running within the virtual machine may be referred to as a "guest operating system."

Virtualized environments are currently storage constrained, which means that the physical memory of the host machine is the bottleneck for resource utilization in the data center. Memory virtualization separates physical memory resources from the data center and then aggregates the resources into virtualized memory pools that can be accessed by guest operating systems or applications running on the guest operating systems. As far as memory virtualization is concerned, memory compression is one of the crucial topics of memory resource management and utilization

Similar to traditional operating systems, the last resort to increase the memory used by the hypervisor is through host swapping (i.e., moving a virtual machine's memory page to the entity's swap space, called swapping out) To reclaim memory from the virtual machine, mark the corresponding page table entry (PTE) of the physical address-to-machine address (P2M) table of the virtual machine as absent, and then release the corresponding page to the hypervisor's free memory pool, where a page table is a data structure used by a virtual machine to store the mapping between virtual and physical addresses. Later, if the page is accessed again by the virtual machine, a page fault is triggered and a copy-on access (COA) mechanism is activated to bring the page content from the swap memory space to a newly allocated memory page, called swap-in. However, the cost of long latency due to disk input/output (I/O) can be very unsatisfactory.

As another way to improve memory utilization, memory compression can be performed by compressing swapped-out pages of a virtual machine into smaller-sized data and putting them together in memory to save physical storage disks used to store original content. The idea is that swapping in from compressed memory will be faster than swapping in from disk memory because memory accesses are faster than disk access.

Nevertheless, memory compression is mainly seen as a second choice because it not only initiates copy-on-access (COA) that triggers a hardware trap and stops the current application execution, but also consumes processing cycles of the host's processor. to compress and decompress page content and cause more overhead. Therefore, in an ideal situation, to avoid compacting memory pages (i.e., working set) that are frequently accessed by the guest operating system, find free memory pages (i.e., guest memory pages outside the working set) for memory compaction .

Contents of the invention

The present disclosure provides an embodiment related to a memory management method for a virtual machine system. The memory management method includes the following steps. First, a first threshold is set by a processor. Then the processor sets the balloon target to the allocated virtual memory size and decrements the balloon target by a first decrement value step by step according to the swap-in/re-cache error detection result in the first adjustment stage. A swap-in/re-caching error detection result is generated by the processor detecting at least one swap-in or re-caching error event. The decrementing of the balloon target is stopped by the processor during the cooling phase upon swap-in/re-cache error detection. The balloon target is gradually decremented by a second decrement value by the processor in a second adjustment phase following the cooling phase. The second decreasing value is smaller than the first decreasing value, and the balloon target is not smaller than the first threshold.

Another embodiment of the present disclosure relates to a computer system including a memory and a processor. The processor is coupled to the memory and performs the following operations for memory management of a virtual machine system. The processor sets a first threshold and sets the balloon target to the allocated virtual memory size and decrements the balloon target by a first decrement value in a first adjustment phase according to the swap-in/re-caching error detection result. The processor also generates a swap-in/re-caching error detection result by detecting at least one swap-in or re-caching error event. The processor stops decrementing balloon targets during the cool down phase based on swap-in/re-cache error detections. The processor also steps down the balloon target by a second decrement value in a second adjustment phase following the cooling phase. The second decreasing value is smaller than the first decreasing value, and the balloon target is not smaller than the first threshold.

Several exemplary embodiments accompanied by figures are described in detail below to further describe the present disclosure in detail.

Description of drawings

The accompanying drawings are included to provide a further understanding of the disclosure, and are incorporated in and constitute a part of this specification. The drawings are used to illustrate exemplary embodiments of the disclosure, and together with the description, serve to explain the principles of the disclosure. However, the drawings are not intended to limit the scope of the disclosure, which is defined by the appended claims.

FIG. 1A is a block diagram illustrating a computer system according to an exemplary embodiment of the present disclosure.

FIG. 1B is a block diagram illustrating a virtual machine system according to an exemplary embodiment of the present disclosure.

FIG. 2 is a stage diagram illustrating a method for memory management of a virtual machine according to an exemplary embodiment of the present disclosure.

FIG. 3 is another stage diagram illustrating a method for memory management of a virtual machine according to an exemplary embodiment of the present disclosure.

【Description of main component labels】

100: Computer Systems

100': virtual machine system

110: Processor

120: System memory

150: virtual machine

155: Guest Operating System

160: Super Manager

170: Virtual Hardware

P12: path

P21: Path

P23: Path

P31: Path

P32: Path

P34: Path

S1: The first adjustment stage

S2: cooling stage

S3: Second adjustment stage

S4: another cooling phase

S5: The third adjustment stage

detailed description

Reference will now be made in detail to the exemplary embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers are used in the drawings and description to refer to the same or like parts.

For illustration purposes, one processor and one system memory are used in the following exemplary embodiments, and the disclosure is not limited thereto. In other exemplary embodiments, more than one processor and more than one system memory may be used.

FIG. 1A is a block diagram illustrating a computer system according to an exemplary embodiment of the present disclosure. Referring to FIG. 1 , a computer system 100 includes a processor 110 , a system memory 120 and other standard peripheral components (not shown), wherein the system memory 120 is coupled to the processor 110 .

Processor 110 may be a dedicated or proprietary processor configured to perform specific tasks by executing a machine-readable software code language that defines functions related to operations to perform by communicating with other components of computer system 100 Functional operation.

The system memory 120 stores software such as an operating system and temporarily stores data or application programs that are currently active or frequently used. Thus, system memory 120 (also referred to as physical memory) may be faster memory (eg, random access memory (RAM), static random access memory (SRAM), or dynamic random access memory (DRAM)) for faster short access time.

Virtual memory is a technique used to manage resources of system memory 120 . It provides imaginary mass storage. Both virtual memory and system memory 120 are divided into blocks of contiguous memory addresses, also referred to as memory pages. System memory 120 may, for example, include compressed memory associated with at least one virtual machine running on computer system 100 . Compressed memory temporarily stores less recently accessed memory pages in a compressed format to make more space available in system memory 120 .

The hypervisor is installed on the computer system 100 and provides a virtual machine execution space in which at least one virtual machine can be instantiated and executed in the virtual machine at the same time. FIG. 1B is a block diagram illustrating a virtual machine system according to an exemplary embodiment of the present disclosure. In this embodiment, only one virtual machine will be described, but the present disclosure is not limited thereto. In other embodiments, multiple virtual machines can coexist and operate in a similar manner.

Referring to FIG. 1B and FIG. 1A , the virtual machine system 100 ′ includes a virtual machine 150 having a guest operating system 155 and other application programs (not shown), a hypervisor 160 and virtual hardware 170 . The virtual hardware 170 includes a processor, a memory, and an I/O device, and is extracted and allocated as a virtual processor, a virtual memory, and a virtual I/O device to provide to the virtual machine 150 running on the upper layer. The hypervisor 160 manages the virtual machines 150 and provides emulated hardware and firmware resources. In one embodiment, a Linux distribution can be installed as a guest operating system 155 within the virtual machine 150 to execute any supported applications, and Xen, an open source software that supports most Linux distributions, can be provided as the hypervisor 160 . Guest operating system 155 includes a balloon driver (not shown). In conjunction with hypervisor 160, the balloon driver may allocate or deallocate virtual memory for guest operating system 155 by calling memory management algorithms. For example, by utilizing the existing page reclamation mechanism of the guest operating system 155, a so called true working set based ballooning algorithm (referred to as TWS ballooning algorithm) was developed to detect working set and reclaim idle memory pages as compaction targets. In practical application, we especially focus on the Linux guest operating system, and the true working set balloon algorithm can also be used in other guest operating systems (for example, Microsoft Windows, the window operating system of Microsoft).

In more detail, in order for memory compaction to be effective, it is necessary to identify the working set of the virtual machine 150 and compress memory pages outside the working set by utilizing the page reclamation mechanism. Intuitively, the working set of a virtual machine 150 is defined as the amount of memory that was recently actively used by the virtual machine 150 . For page reclamation, the Linux guest operating system 155 uses the least recently used (LRU) criterion to determine the order in which to evict pages and maintains the two most recent Least used lists: active and inactive lists. The memory pages of the anonymous memory are used by the heap, stack of the user process, and the memory pages of the page cache are backed up by disk data, where the contents are cached in memory after the first access to the disk data to reduce the possibility of subsequent The time spent doing disk I/O. Memory pages in the active list are considered to be accessed more frequently, called hot pages; pages in the inactive list are considered to be accessed less frequently, called cold pages. After allocation, each memory page is put into the active list by default.

In an embodiment of the present disclosure, the page reclamation mechanism may be directly triggered when the kernel of the guest operating system 155 (for example, the domU kernel) cannot allocate memory. For example, when a memory page is requested, the kernel may not be able to find the memory page from the hypervisor 160 free memory pool. The kernel reclaims memory from an inactive list containing memory pages that are considered relatively cool, such that the reclaimed memory will not be accessed anytime soon in the near future. When the number of memory pages in the inactive list is insufficient to fulfill a memory allocation request, the kernel may traverse the active list and move cold pages from the active list to the inactive list.

One way to determine whether a memory page is a hot page or a cold page is to check and clear the hardware reference bit of the page table entry of the memory page. Therefore, if the list reference bit is on during kernel peeking, then the reference bit is cleared, and then the memory page is considered a hot page and is kept in the active list. Otherwise, the memory page is considered a cold page and moved to the inactive list.

If a memory page on the inactive list is anonymous memory, the kernel may swap out the contents to, for example, swap memory space, mark the process's corresponding page table entry as absent, and then free the corresponding memory page. Later, if the memory page is accessed again, a copy-on-access (COA) mechanism is performed by bringing the page content from the swap memory space into a newly allocated memory page (ie, swapping in). Alternatively, if the memory page on the inactive list belongs to the page cache, the kernel can flush the page content to swap memory space if the page content has become dirty, and then free the page. After the next access, the kernel must perform another disk access (called a refault) to bring the contents back to the newly allocated page in the page cache. It should be noted that in one embodiment, the swap memory space may be an area on a hard disk (not shown) in the computer system 100 for unloading LSU pages from the system memory 120 . However, in another exemplary embodiment, if the user experiences slow operation of the computer system 100, a part of the system memory 120 can also be used as a swap memory space.

Whenever a swap-in or re-cache error event occurs, the virtual machine 150 will have degraded performance due to disk I/O latency. In one embodiment, from the perspective of page reclamation, the performance overhead of the virtual machine 150 can be quantified by summing the swap-in count and the re-cache error count, called overhead_count, which can be written as equation (1) :

Consumption count overhead_count=swap-in count+again cache error count equation (1)

In the process of performing memory management of the virtual machine 150, the balloon driver may be instructed to expand or contract. The inflation command will be issued when the computer system 100 is under memory pressure. A shrink command is issued when memory pressure has eased. Each expand or shrink command includes an indication of the number of memory pages of virtual machine 150 (referred to herein as a balloon target). The balloon object associated with the expand or shrink command represents the number of guest physical memory pages reclaimed from or returned to the guest operating system 155 . In the Linux operating system, the reclaim mechanism can be used to identify the total size of anonymous memory consumed by all processes at a moment through the Committed_AS value. The Committed_AS value represents the anonymous memory consumed by the virtual machine 150 , but does not necessarily correspond to the working set size of the virtual machine 150 . In other words, the Committed_AS value is incremented after the first access to each newly allocated anonymous memory page, but decremented when the owner process explicitly frees the memory page. For example, if a program allocates and accesses a memory page only once when the program starts but does not touch the memory page until the program exits, then the Linux kernel may not exclude this cold page from Committed_AS even though it is clearly in the working set The same is true outside. That is, if a page cache page that belongs to the working set is evicted due to memory reclamation, a re-cache fault event can occur and thus can be used as a signal that one more memory page should be added to the working set to accommodate the page cache .

Thus, in an embodiment, a counter of re-caching error events in virtual machine 150 may be maintained and the balloon target may be adjusted according to the re-caching error count such that the performance penalty incurred by evicted page caches may be reduced. to the minimum. Additionally, the true working set of virtual machine 150 may be actively probed. If virtual machine 150 is not performing significant disk I/O, then the true working set of virtual machine 150 is lower than the Committed_AS value, and significant disk I/O requires additional buffer cache pages contained in the working set. That is, for a virtual machine 150 with a physical memory allocation equal to its working set size, the disk access overhead associated with swapping in and recaching errors should be close to zero. Therefore, in order to probe the true working set of virtual machine 150, the balloon driver's balloon target may be incrementally increased, and not stopped until the swap-in and re-cache error counts start to become non-zero.

More specifically, in one of the exemplary embodiments, the estimated working set size EWSS of a virtual machine 150 may be defined as equation (2)

EWSS = allocated memory size + consumption count equation (2)

Where the allocated memory size is the number of memory pages allocated to the virtual machine 150, and the overhead count is the number of pages faulted into the virtual machine 150, and is also defined in equation (1). However, in another exemplary embodiment, the estimated working set size EWSS of the virtual machine 150 may be a linear combination of the allocated memory size and the overhead count as written in equation (3)

EWSS=A1×(allocated memory size)+A2×(overhead count)+C Equation (3)

Where A1≧1, A2≦1, and C is a constant.

When the amount of allocated memory of a virtual machine 150 is higher than the true working set, a custom balloon driver is used to collect swap-in and re-cache error counts every second and the balloon target is adjusted to detect the true working set of the virtual machine 150 . The upper limit of the balloon target is set by the processor 110 to the configured memory size of the virtual machine 150 when the virtual machine 150 is started, and the lower limit is set according to the amount of memory constrained by the guest operating system 155 . In other words, an initial lower bound value is calculated based on a self-ballooning algorithm, adding page requirements reserved for system panics and compressed pages. Without tuning, the guest operating system 155 may be prone to low memory exceptions when the Committed_AS value is low. To better approximate the true working set size of virtual machine 150 , processor 110 uses three runtime phases and adaptively adjusts the balloon target, as shown in FIG. 2 .

FIG. 2 is a stage diagram illustrating a method for memory management of a virtual machine according to an exemplary embodiment of the present disclosure.

Referring to FIG. 2 and the components in FIGS. 1A and 1B , starting from the first adjustment stage S1, the processor 110 initializes the balloon target of the balloon driver of the guest operating system 155 to a first threshold value (for example, Committed_AS value), which Yes so no explicit information exists on the page cache at this time. The Committed_AS value can be retrieved from the exported kernel variable vm_committed_AS, which can be accessed by any kernel component or loadable kernel module in Linux. Next, the processor 110 decreases the balloon target by a first decrement value step by step, wherein the balloon target is not smaller than the first threshold. In this exemplary embodiment, the first decrement value may be a percentage (eg, 5%) of the current Committed_AS value. Processor 110 may generate a swap-in/re-caching error detection result whenever a swap-in or re-caching error event occurs, wherein the swap-in/re-caching error detection result may indicate the occurrence of a swap-in or re-caching error event . Based on the swap-in/re-cache error detection result, the processor 110 may stop lowering the balloon target and switch the balloon driver from the first adjustment phase S1 to the cool-down phase S2 via path P12. Next, the processor 110 raises the balloon target by an amount of memory equal to the total number of swap-in and re-caching error counts in cool-down phase S2 because each of the swap-in or re-caching error events raises the Additional free memory pages are required.

It should be noted that the swap-in count can be retrieved from a variable (e.g. variable pswpin) of a kernel component (e.g. kernel component vmstat) that collects statistics on profile information such as operating system memory in Linux , processes (processes), interrupts (interrupts), paging (paging) and block I/O (block I/O). Information about recaching errors from disk I/O can be obtained by intercepting (for example) I/O operations to swap memory space using facilities in the Linux kernel (for example, the facility blktrace ), which provide an application programming interface (API ) to add hook points to disk I/O and perform statistics. Once the disk I/O is tracked, each block of the disk is initialized with a "0" bitmap and set to a "1" bitmap when the corresponding disk block is accessed. Re-cache error events are counted when the bitmap has been set before changing to one (ie, the corresponding block has been accessed before but requires another access).

This swap-in and re-cache error event indicates that the balloon target is approaching the true working set or a sudden burst of memory demand from the application. Therefore, it is unwise to further reduce the memory allocation of the virtual machine 150 . Even when a swap-in or re-cache error event occurs, the memory allocation to the virtual machine 150 is allowed to exceed the Committed_AS value. This flexible configuration is especially important for virtual machines 150 running I/O-intensive workloads, where the Committed_AS value does not reflect the additional memory requirements caused by page caching.

Also in the cooling phase S2, the processor 110 initializes the cooling counter to a second threshold value (eg, arbitrarily set to 8) and decrements the counter every time period from then on, wherein the time period Can be one second. When the cooling counter reaches zero, the processor 110 may consider the workload burst to have passed and then switch the balloon driver to the second tuning stage S3 via path P23.

The balloon target applies the same logic in the second adjustment stage S3 as in the first adjustment stage S1, except that the processor 110 lowers the balloon target by a second decrement value in a step-by-step manner, where the second decrement value is smaller than the first decrement value value. In this embodiment, the second decrement value may be a percentage (for example, 1%) of the current Committed_AS value. More specifically, the processor 110 reduces the balloon target by 1% of the current Committed_AS in a stepwise manner in the second adjustment stage S3. Whenever a swap-in or re-cache error event occurs, the processor 110 stops lowering the balloon target by switching the balloon driver from the second tuning phase S3 to the cooling phase S2 via path P32. Processor 110 also balloons the target up page count by an amount equal to the total amount of swap-in and re-cache error counts, and re-initializes the cooldown counter to a second threshold.

On the other hand, when the balloon driver is in the cool-down phase S2 or the second adjustment phase S3, when the Committed_AS value of the virtual machine 150 changes, the processor 110 considers that the working set size of the virtual machine 150 will change significantly and enters the first adjustment Phase S1 to reset the balloon target. That is, if before the processor 110 switches the balloon driver from the second tuning stage S3 via path P31 or from the cooling stage S2 via path P21 to the first tuning stage S1, the balloon object bursts with a swap-in or recaching error If the Committed_AS value is exceeded, the processor 110 may reinitialize the balloon object to the Committed_AS value plus the excess amount as a new estimated working set size according to equation (2). That is, the first threshold value changes at this time. In another embodiment, the processor 110 may also reinitialize the balloon target as a linear combination of the Committed_AS value and the overhead count according to equation (3). It should also be noted that when the processor 110 switches the balloon driver from the cooling stage S2 to the first conditioning stage S1 when the cooling counter has not yet reached zero, the balloon driver still enters the first conditioning stage S1, but the processor 110 resumes working Continue counting down until the cooldown counter reaches zero before set detection.

In one embodiment, a swap-in/recaching error event may be detected in more than one situation before the processor 110 switches the balloon driver from the second tuning phase S2 to the first tuning phase S1. That is, the balloon driver may cycle between a cooling phase S2 and a second conditioning phase S2 before entering the first conditioning phase S1 . As an example, FIG. 3 is another stage diagram illustrating a method for memory management of a virtual machine according to an exemplary embodiment of the present disclosure. It should be noted that only the differences from the exemplary embodiment shown in FIG. 2 are explained below. The present disclosure is not limited thereto.

Referring to FIG. 3 and the components in FIGS. 1A and 1B, before the processor 110 switches the balloon driver from the second tuning stage S2 to the first tuning stage S1, it may switch the balloon driver according to the swap-in/re-caching error detection result. The driver switches alternately to at least one further cooling phase S4 and at least one further second adjustment phase (referred to as at least one third adjustment phase S5 ). In this exemplary embodiment, for the sake of clarity and ease of explanation, only one further cooling stage S4 and one third adjustment stage S5 are illustrated. The present disclosure is not limited thereto.

More specifically, when the balloon driver is in the second tuning stage S3, whenever a swap-in or recaching error event occurs, the processor 110 switches the balloon driver from the second tuning stage S3 to another via path P34. A cooling phase S4 stops lowering the balloon target. Processor 110 also balloons the target up page count by an amount equal to the total amount of swap-in and re-cache error counts, and further initializes the cooldown counter to a third threshold (eg, arbitrarily set to 8). In another cooling phase S4, the processor 110 decrements another cooling counter every time period, wherein the time period may be one second. When the other cooling counter reaches zero, the processor 110 may consider the workload burst to have passed and then switch the balloon driver to the third tuning stage S5 via path P45.

The balloon target applies the same logic in the third adjustment stage S5 as in the second adjustment stage S2, except that the processor 110 lowers the balloon target by a third decrement value in a stepwise manner. In this exemplary embodiment, the third decrement value is set equal to the second decrement value, ie, 1% of the Committed_AS value.

In an embodiment, there are a plurality of further cooling phases S4 and third adjustment phases S5 alternating with each other. Whenever a swap-in or re-cache error event occurs, the processor 110 may then stop lowering the balloon target by switching the balloon driver from the third adjustment stage S5 to the next further cooling stage in a similar manner. Certainly, the processor 110 may further switch the balloon driver from the next another cooling stage to the next another third adjustment stage according to another cooling counter. It should also be noted that the third decrement value may be different in each of the further cooling stages. The present disclosure is not limited thereto.

In addition, similar to the second adjustment stage S2 in FIG. 2, when the balloon driver is in another cooling stage S4 or another third adjustment stage S5, when the Committed_AS value of the virtual machine 150 changes, the processor 110 considers the virtual machine The working set size of 150 will change significantly and the balloon target will be reset by entering the first adjustment stage S1 via paths S41 and S51 respectively. For details, refer to related descriptions in the exemplary embodiment in FIG. 3 .

Through the above-mentioned memory management method (i.e., the TWS balloon algorithm), the processor 110 can detect the true working set of the virtual machine 150, and reclaim unnecessary cold pages from the system memory 120 to the memory pool of the hypervisor 160 to save more memory. multiple resources while maintaining application performance without significant degradation.

In the embodiments, the memory management method described above can be implemented by executing prepared programs on computers such as personal computers and workstations. The program is stored on a computer-readable recording medium (eg, hard disk, floppy disk, CD-ROM, MO, and DVD), read from the computer-readable medium, and executed by a computer. The program can be distributed throughout a network (eg, the Internet).

In general, by utilizing the existing page reclamation mechanism of the guest OS, the memory management method in the present disclosure is designed to detect the virtual machine's true working set and idle memory pages are evicted as compaction targets. By using a custom balloon driver, the true working set of a virtual machine can be accurately and dynamically estimated based on the current memory usage of the virtual machine, taking into account swap-in/re-caching error events, and the true working set further provides memory Advantages of resource management.

Claims (22)

1. a kind of storage management method of dummy machine system, including：

First threshold is set by processor；

By the processor in the first metamorphosis stage according to changing to/again cache error detection result and balloon is driven into journey The balloon goal setting of sequence is the virtual memory size of distribution and the balloon target progressively successively decreased the first decrement value, wherein The balloon driver distributes or deallocates the virtual memory of client operating system by using memory management algorithm；

Changed to by processor detection at least one or again cache error event come produce it is described change to/again cache it is wrong Error detection result；

By the processor in cooling stage according to it is described change to/described in cache error detection result and stopping successively decreases again Balloon target；And

The balloon target is progressively successively decreased in the second metamorphosis stage after the cooling stage by the processor Two decrement values,

Wherein described second decrement value is less than first decrement value, and the balloon target is not less than the first threshold.

2. storage management method according to claim 1, also includes：

When the first threshold changes, the balloon driver is switched to from the cooling stage by the processor The balloon driver is switched to first metamorphosis stage from second metamorphosis stage by first metamorphosis stage.

3. storage management method according to claim 1, wherein it is described by the processor by the balloon target Be set as the distribution virtual memory size and by the balloon target progressively successively decrease first decrement value the step of wrap Include：

By the processor by number that the balloon goal setting is the locked memory pages for being assigned to the dummy machine system； And

First percentage ratio of the first threshold of the balloon target progressively being successively decreased by the processor.

4. storage management method according to claim 1, wherein produce it is described change to/again cache error detection tie After fruit, methods described also includes：

The balloon target is incremented by expense by the processor to count, wherein it is at least one according to that the expense is counted It is individual change to and again the total number of cache error event and set.

5. storage management method according to claim 1, wherein it is described by the processor in the cooling stage Change to described in middle basis/again cache error detection result and stop successively decreasing the balloon target the step of also include：

Enumerator is set as and the enumerator that progressively successively decreases in the cooling stage by Second Threshold by the processor.

6. storage management method according to claim 5, also includes：

When the enumerator reaches zero, the balloon driver is switched to from the cooling stage by the processor Second metamorphosis stage.

7. storage management method according to claim 4, wherein when the first threshold changes, by the process Device by the balloon driver from second metamorphosis stage be switched to first metamorphosis stage the step of also include：

The linear combination that virtual memory size and the expense that the first threshold is updated to the distribution are counted；And

The balloon target is reset to into the first threshold of the renewal.

8. storage management method according to claim 1, wherein it is described by the processor in the described second adjustment In stage by the balloon target progressively successively decrease the second decrement value the step of also include：

Second percentage ratio of the first threshold of the balloon target progressively being successively decreased by the processor.

9. storage management method according to claim 1, also includes：

By the processor according to it is described change to/again cache error detection result by the balloon driver from described Two metamorphosis stages were to switch alternately at least one another cooling stage and at least one the 3rd metamorphosis stages；And

When the first threshold changes, will be the balloon driver another cold from described at least one by the processor But each in stage and at least one the 3rd metamorphosis stage is switched to first metamorphosis stage.

10. storage management method according to claim 9, also includes：

By another enumerator is set as in each of the processor in described at least one another cooling stage Three threshold values and another enumerator that progressively successively decreases；

By the balloon target progressively being passed in each of the processor in described at least one the 3rd metamorphosis stages Subtract the 3rd decrement value,

Wherein when another enumerator reaches zero, by the processor by the balloon driver from described at least one Each in individual cooling stage is switched to next 3rd metamorphosis stage in described at least one the 3rd metamorphosis stages, and

Wherein when produce it is described change to/cache error detection result again when, by the processor by balloon driving journey Sequence each from the 3rd metamorphosis stage be switched to the next cooling stage at least one cooling stage.

11. storage management methods according to claim 1, wherein when the dummy machine system starts, by described The upper limit of the balloon target is set as processor the virtual memory size of the distribution, and wherein passes through the processor Lower limit set by the balloon target is the amount of the memorizer constrained by the dummy machine system.

A kind of 12. computer systems, including：

Memorizer；

Processor, is couple to the memorizer, wherein the processor performs operation for the memory management of dummy machine system, The operation includes：

Setting first threshold；

In the first metamorphosis stage according to changing to/again cache error detection result and the balloon target of balloon driver is set It is set to the virtual memory size of distribution and the balloon target is progressively successively decreased the first decrement value, wherein the balloon driver Distribute or deallocate the virtual memory of client operating system by using memory management algorithm；

Changed to by detection at least one or cache error event described changes to/cache error detection knot again to produce again Really；

In cooling stage according to it is described change to/again cache error detection result and stop the balloon target of successively decreasing；And

The balloon target is progressively successively decreased the second decrement value in the second metamorphosis stage after the cooling stage,

Wherein described second decrement value is less than first decrement value, and the balloon target is not less than the first threshold.

13. computer systems according to claim 12, wherein when the first threshold changes, the processor is by institute State balloon driver first metamorphosis stage is switched to or by the balloon driver from described from the cooling stage Second metamorphosis stage was switched to first metamorphosis stage.

14. computer systems according to claim 12, wherein the balloon goal setting is distribution by the processor To the locked memory pages of the dummy machine system number and the balloon target progressively successively decreased into the first of the first threshold Percentage ratio.

15. computer systems according to claim 12, wherein the balloon target is incremented by expense meter by the processor Number, wherein the expense count be change to according to described at least one and again the total number of cache error event and set.

16. computer systems according to claim 12, wherein the processor in the cooling stage by enumerator It is set as Second Threshold and the enumerator that progressively successively decreases.

17. computer systems according to claim 16, wherein when the enumerator reaches zero, the processor enters one The balloon driver is switched to second metamorphosis stage from the cooling stage by the processor by step.

18. computer systems according to claim 15, wherein the processor first threshold is updated to it is described The linear combination that the virtual memory size of distribution and the expense are counted, and the balloon target is reset to into the renewal First threshold.

19. computer systems according to claim 12, wherein the balloon target is progressively successively decreased institute by the processor State the second percentage ratio of first threshold.

20. computer systems according to claim 12, wherein the processor changes to/cache mistake again according to described Testing result and that the balloon driver is to switch alternately at least one from second metamorphosis stage further is another Cooling stage and at least one the 3rd metamorphosis stages, and when the first threshold changes, by the balloon driver from At least one another cooling stage and each at least one the 3rd metamorphosis stage are switched to described first Metamorphosis stage.

21. computer systems according to claim 20, wherein the processor is in described at least one another cooling rank Another enumerator is set as into the 3rd threshold value and another enumerator that progressively successively decreases further in each in section, and in institute State in each at least one the 3rd metamorphosis stages and the balloon target progressively successively decreased the 3rd decrement value,

Wherein when another enumerator reaches 0, the processor will be the balloon driver cold from described at least one But each in the stage is switched to next 3rd metamorphosis stage in described at least one the 3rd metamorphosis stages, and

Wherein when produce described in change to/cache error detection result again when, the processor by the balloon driver from Each in 3rd metamorphosis stage is switched to the next cooling stage at least one cooling stage.

22. computer systems according to claim 12, wherein when the dummy machine system starts, the processor will The upper limit of the balloon target is set as the virtual memory size of the distribution, and wherein described processor according to by the void The amount of the memorizer of plan machine system restriction sets the lower limit of the balloon target.