CN104102460A - Cloud computing-based memory management method and device - Google Patents

Abstract

The present invention provides a cloud computing-based memory management method and device, which are applied to a cloud computing platform; the method includes: expanding the local physical address space by using a logically extended address space; memory; performing bidirectional mapping between the global extended address space and the logical extended address space according to predetermined rules. The invention can optimize the utilization of memory resources in cloud computing.

Description

A memory management method and device based on cloud computing

technical field

The present invention relates to the field of computers, in particular to a memory management method and device based on cloud computing.

Background technique

The rise of cloud computing is gradually changing the entire computer industry and academia. This way of using computers as a utility tool and software as a service has brought new challenges to both computer software and hardware design. huge challenge. In terms of software construction and distribution methods, SOA (Service-Oriented Software Architecture) and SaaS (Software as a Service) are increasingly becoming the core methods of software construction and distribution under the cloud computing model. As for the underlying computing infrastructure that supports the operation of services, the challenges mainly focus on two aspects: system software and hardware. Among them, adaptive resource allocation is the biggest challenge faced by system software, and how to allocate resources is the core issue of this challenge. Static allocation of resources based on peak loads will result in serious waste of resources. According to statistics, in current data centers, the utilization rate of CPU resources is between 5% and 20% on average, and the utilization rate of memory resources does not exceed 60%. Obviously, the current resource allocation mode has a high degree of resource waste, which increases the user's total cost of ownership (TCO). Statically allocating resources based on the average load will degrade the service quality under peak load conditions, and ultimately affect the potential profit of cloud users.

Contents of the invention

The technical problem to be solved by the present invention is how to optimize the utilization of memory resources in cloud computing.

In order to solve the above problems, the present invention provides a memory management method based on cloud computing, which is applied in a cloud computing platform, including:

Expand the local physical address space by using the logical extension address space;

Use the global extended address space to record the remote free memory used locally;

The bidirectional mapping between the global extended address space and the logical extended address space is performed according to a predetermined rule.

Optionally, the global extended address space includes an identifier of a remote server where the remote free memory is located and an address segment of the remote free memory in the remote server.

Optionally, when the bidirectional mapping between the global extended address space and the logical extended address space is performed according to predetermined rules, the remote free memory of multiple remote servers is mapped to one logical extended address space.

Optionally, the method also includes:

When the remote free memory exits, copy the data in the remote free memory to the remote free memory provided by another remote server page by page, add the remote free memory provided by the other remote server into the global extended address space, And mapped to the same segment of logical extended address space originally mapped by the exited remote free memory.

Optionally, the method also includes:

The command that reads and writes to the remote free memory is sent by Xen through the shared buffer between Xen and Domain0, and is processed by the Swap daemon process in Domain0.

A memory management device based on cloud computing, applied to a cloud computing platform, comprising:

The virtual machine monitor is used to expand the local physical address space by using the logically extended address space;

The console is used to record the remote free memory used locally by using the global extended address space;

The address mapping module is configured to perform bidirectional mapping between the global extended address space and the logical extended address space according to predetermined rules.

Optionally, the global extended address space includes an identifier of a remote server where the remote free memory is located and an address segment of the remote free memory in the remote server.

Optionally, when the address mapping module performs bidirectional mapping between the global extended address space and the logical extended address space according to predetermined rules, the remote free memory of multiple remote servers is mapped to one logical extended address space.

Optionally, the address mapping module is also used to copy the data in the remote free memory to the remote free memory provided by another remote server page by page when the remote free memory exits, and the data provided by the other remote server The remote free memory is added to the global extended address space, and mapped to the same segment of logical extended address space originally mapped by the withdrawn remote free memory.

Aiming at the needs of memory optimization under the virtualized cloud computing platform, the present invention constructs a mechanism for seamlessly aggregating idle memory resources by using virtualization technology; The double-layer address space mapping mechanism of space and global extended address space; using this mechanism, the memory hierarchy under the virtualized cloud platform is reconstructed, and a high-reliability extended memory area based on a redundant backup mechanism is constructed to optimize memory utilization; By extending the virtual machine monitor, an idle (and low-utilization) page recovery mechanism transparent to the virtual machine is realized. This mechanism has special significance for the situation where the idle memory recovery is not suitable for modifying the operating system code. The present invention realizes access to remote idle memory resources through the addressing technology across physical platforms, and can allocate memory with low utilization rate to virtual machines with high memory utilization rate, thereby optimizing the utilization of the entire memory resource.

Description of drawings

Fig. 1 is a schematic diagram of a system using the method of Embodiment 1;

Figure 2 is a schematic diagram of mapping;

Fig. 3 is a state transition diagram of a memory server page;

Fig. 4 is a schematic diagram of the extended Xen system;

Fig. 5 is a schematic diagram of the device of the second embodiment.

Detailed ways

The technical solution of the present invention will be described in more detail below with reference to the drawings and embodiments.

It should be noted that, if there is no conflict, the embodiments of the present invention and various features in the embodiments can be combined with each other, and all are within the protection scope of the present invention. In addition, although a logical order is shown in the flowcharts, in some cases the steps shown or described may be performed in an order different from that shown or described herein.

Embodiment 1. A memory management method based on cloud computing, comprising:

Expand the local physical address space by using the logical extension address space;

Use the global extended address space to record the remote free memory used locally;

The bidirectional mapping between the global extended address space and the logical extended address space is performed according to a predetermined rule.

In order to utilize free memory resources distributed on each physical server of the entire platform, this embodiment introduces a distributed memory pool composed of two-layer address space mapping in consideration of the characteristics of the distributed platform. The construction of the double-layer address space takes into account both efficiency and reliability, and provides the greatest degree of abstraction of memory resources. As shown in Figure 1, in each node (node 1, node 2, ..., node n) in the cloud computing platform, memory with low utilization rate in the virtual machine (VM) (such as in node 1 and node 2 in Figure 1 The memory of VM#1) is collected into the free resource pool, and then provided to the virtual machine where the memory with high utilization rate is located (such as the memory of VM#k in node 2 and node n in Figure 1).

In an implementation manner of this embodiment, the global extended address space includes a remote server identifier where the remote free memory is located and an address segment of the remote free memory in the remote server.

In this embodiment, the global extended address space can use <server ID, address segment> to mark the address of the remote free memory; in the global extended address space used to record multiple remote free memories of the same remote server, the remote The server IDs are the same, but the address segments are different.

In this implementation manner, the global extended address space can be dynamically adjusted according to changes in free memory. After the free memory of the remote server is marked with the global extended address space, it is mapped to the logical extended address space.

In an implementation manner of this embodiment, when the two-way mapping between the global extended address space and the logical extended address space is performed according to predetermined rules, the remote free memory of multiple remote servers is mapped to one logical extended address space; Realize redundant storage of data and improve the reliability of remote memory.

In this embodiment, as shown in FIG. 2 , the same segment of address space in the logical extended address space can be mapped to different servers and remote free memories in different memories through the global extended address space.

The design of the remote memory system needs to consider two requirements: (1) shorten the memory access delay as much as possible to improve efficiency; (2) the reliability of the distributed system, and provide reliable and stable memory-based storage services. This embodiment fully considers the requirements of these two aspects, and achieves the purpose of high efficiency and reliability through a many-to-one mapping mechanism from the global extended address space to the logical extended address space.

In an implementation manner of this embodiment, an address mapping component is used to provide bidirectional mapping between the global extended address space and the logical extended address space.

In this embodiment, an efficient and reliable remote memory storage abstraction is provided by maintaining related mapping relationships. The address mapping component supports two basic types of data operations: data access and data synchronization. Among them, data access is a basic remote memory operation, which realizes reading and writing of remote data. Data synchronization operations provide support for remote free memory eviction.

In an implementation manner of this embodiment, the method may further include:

When the remote free memory exits, copy the data in the remote free memory to the remote free memory provided by another remote server page by page, add the remote free memory provided by the other remote server into the global extended address space, And mapped to the same segment of logical extended address space originally mapped by the exited remote free memory.

In this embodiment, when a remote server needs to reclaim shared memory due to an increase in its own load or the remote server cannot provide memory services normally, the system will actively or passively start a data synchronization operation to migrate data back up to other remote servers. Using the data synchronization method to migrate data from the server exiting free memory to other remote servers can achieve high reliability of data storage and ensure data validity.

The state transition of the memory server page is shown in Figure 3. After remote writing or synchronous writing of uninitialized pages, the page status becomes dirty; after synchronous reading of uninitialized pages or remote reading of dirty pages, the page status becomes idle; the state is still idle after synchronous writing to the idle page, and becomes dirty after remote writing; the state is still dirty after synchronous reading of the dirty page.

In an implementation of this embodiment, in addition to unallocated memory or memory actively shared by the virtual machine, the remote free memory may also include the part of the virtual machine that has been allocated and used due to the limitation of the data access mode. , but the memory space with extremely low access frequency.

In order to maximize memory sharing, this embodiment introduces a free page reclamation technology that is transparent to virtual machines. In order to realize the transparent reclamation of pages, this embodiment extends the Xen (an open source virtual machine monitor) virtual machine monitor to implement a dynamic "stealing" runtime system based on memory page access information.

The extended Xen virtual machine monitor, as shown in Figure 4, completes two core functions: (1) introduces logical extended address space to support remote memory access; (2) provides transparent addressing support for pages not in local memory.

In order to support the logically extended address space, this embodiment modifies the mapping (p2m mechanism) from the virtual physical address to the machine address in the Xen virtual machine monitor, and adds a new address type to mark the logically extended address space.

When the virtual machine accesses the logical extended address space, it will either trigger a page fault exception, or trigger the logical extended address space addressing mechanism. When running through page fault exception handling and instruction execution rollback, the virtual machine is suspended to wait for the page to be reloaded into the local memory; when the page is reloaded into the local memory, p2m is reset to point to the newly loaded memory page . In the case of paravirtualization, the process of page fault exception processing will continue. In the case of full virtualization, the exception replay mechanism will play a role to reschedule the execution of the virtual machine and restore the abnormal processing process.

The system intercepts the access to the logical extension address space during the instruction simulation execution process, triggers the access exception of the logical extension address space, and passes it to the upper layer function step by step until the execution logic reaches the entry point where the virtual machine falls into Xen. There, the virtual machine is suspended, and the operation commands on the logical extended address space are passed to Domain0 through the shared memory space. After the data is returned, the instruction simulation operation or exception handling operation is continued, and the virtual machine is rescheduled.

In this embodiment, the command that reads and writes remote free memory is sent by Xen through the shared buffer between Xen and Domain0, and the Swap daemon process in Domain0 is responsible for specific processing. The reason for this is that Xen itself lacks a network protocol stack, and such a design can greatly improve system flexibility.

In Domain0, this embodiment implements a flexible global expansion address space management mechanism. Wherein, each segment of continuous address space (address segment) is represented by an address space descriptor and space range, and each remote server may contain one or more address spaces. Redundant storage of data can be realized by mapping multiple address spaces to the same logical extended address space. In the form of modules, this embodiment supports two forms of external storage: remote free memory and local memory. By overlaying and mapping local memory and remote memory, an additional storage system with high reliability and high performance can be obtained.

In this embodiment, information communication with Xen based on a shared page can be realized through mmap, and Xen transmits a request to Domain0 through a shared message channel. In Domain0, the KXenSwap kernel thread of the daemon process is created to respond to control commands issued from Xen, such as read and write operations of remote memory. When KXenSwap completes the relevant commands through the driver, a new Hypercall is invoked to notify Xen to complete the subsequent p2m table modification, page table update and other processes to resume the scheduling execution of the virtual machine.

Embodiment 2. A memory management device based on cloud computing is applied to a cloud computing platform, as shown in FIG. 5 , including:

The virtual machine monitor is used to expand the local physical address space by using the logically extended address space;

The console is used to record the remote free memory used locally by using the global extended address space;

The address mapping module is configured to perform bidirectional mapping between the global extended address space and the logical extended address space according to predetermined rules.

In this embodiment, the address mapping module can also be integrated in the console as an address mapping component (as shown in Figure 5); the memory space of the virtual machine is not only derived from the local memory, but also derived from the logical extended address space through Remote free memory mapped by the global extended address space; local memory can also provide memory space for the console.

In an implementation manner of this embodiment, the global extended address space may include a remote server identifier where the remote free memory is located and an address segment of the remote free memory in the remote server.

In an implementation of this embodiment, when the address mapping module performs bidirectional mapping between the global extended address space and the logical extended address space according to predetermined rules, it maps remote free memory of multiple remote servers to a logical extended address space.

In an implementation manner of this embodiment, the address mapping module is also used to copy the data in the remote free memory to the remote free memory provided by another remote server page by page when the remote free memory exits, and the The remote free memory provided by another remote server is added to the global extended address space, and mapped to the same section of logical extended address space originally mapped by the withdrawn remote free memory.

Those skilled in the art can understand that all or part of the steps in the above method can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a magnetic disk or an optical disk, and the like. Optionally, all or part of the steps in the foregoing embodiments may also be implemented using one or more integrated circuits. Correspondingly, each module/unit in the foregoing embodiments may be implemented in the form of hardware, or may be implemented in the form of software function modules. The present invention is not limited to any specific combination of hardware and software.

Of course, the present invention can also have other various embodiments, and those skilled in the art can make various corresponding changes and deformations according to the present invention without departing from the spirit and essence of the present invention, but these corresponding Changes and deformations should all belong to the protection scope of the claims of the present invention.

Claims (9)

1. A memory management method based on cloud computing, applied in a cloud computing platform, comprising: Expand the local physical address space by using the logical extension address space; Use the global extended address space to record the remote free memory used locally; The bidirectional mapping between the global extended address space and the logical extended address space is performed according to a predetermined rule. 2. The method of claim 1, wherein: The global extended address space includes the remote server identifier where the remote free memory is located and the address segment of the remote free memory in the remote server. 3. The method of claim 1, wherein: When the two-way mapping between the global extended address space and the logical extended address space is performed according to predetermined rules, the remote free memory of multiple remote servers is mapped to one logical extended address space. 4. The method of claim 1, further comprising: When the remote free memory exits, copy the data in the remote free memory to the remote free memory provided by another remote server page by page, add the remote free memory provided by the other remote server into the global extended address space, And mapped to the same segment of logical extended address space originally mapped by the exited remote free memory. 5. The method of claim 1, further comprising: The command that reads and writes to the remote free memory is sent by Xen through the shared buffer between Xen and Domain0, and is processed by the Swap daemon process in Domain0. 6. A memory management device based on cloud computing, applied in a cloud computing platform, is characterized in that, comprising: The virtual machine monitor is used to expand the local physical address space by using the logically extended address space; The console is used to record the remote free memory used locally by using the global extended address space; The address mapping module is configured to perform bidirectional mapping between the global extended address space and the logical extended address space according to predetermined rules. 7. The device of claim 6, wherein: The global extended address space includes the remote server identifier where the remote free memory is located and the address segment of the remote free memory in the remote server. 8. The device of claim 6, wherein: When the address mapping module performs bidirectional mapping between the global extended address space and the logical extended address space according to predetermined rules, the remote free memory of multiple remote servers is mapped to one logical extended address space. 9. The device of claim 6, wherein: The address mapping module is also used to copy the data in the remote free memory to the remote free memory provided by another remote server page by page when the remote free memory exits, and add the remote free memory provided by the other remote server to The global extended address space is mapped to the same section of logical extended address space originally mapped by the withdrawn remote free memory.