WO2012107988A1

WO2012107988A1 - Memory management program, memory management method and information processing device

Info

Publication number: WO2012107988A1
Application number: PCT/JP2011/052488
Authority: WO
Inventors: 丈治加藤
Original assignee: 富士通株式会社
Priority date: 2011-02-07
Filing date: 2011-02-07
Publication date: 2012-08-16

Abstract

This memory management method includes: a step for receiving a request for a page of a first size from an operating system; and an assignment step which, if free pages in a first memory area, which has been partitioned into pages of the first size, are insufficient to accommodate the request, partitions free pages of a second size which is larger than the first size in a second memory area, which has been partitioned into pages of the second size, into pages of the first size so as to assign the pages of the first size in response to the request. Accordingly, memory can be efficiently utilized.

Description

Memory management program, memory management method, and information processing apparatus

The present technology relates to a memory management program, a memory management method, and an information processing apparatus.

A general operating system (OS: Operating System) introduces a mechanism for delaying the allocation of memory to an application program (hereinafter abbreviated as application) until the application actually accesses the memory area. . This mechanism is known as a demand paging mechanism, which can reduce the overhead due to the loading process of the program into the memory and improve the response performance of the system.

This is because, in general-purpose operating systems such as UNIX (registered trademark), interactive jobs such as shells are mainly executed, and thus, responsiveness from the user's perspective is important.

For this reason, in such a general-purpose operating system, the process of confirming whether or not memory allocation is possible, that is, the process of confirming that free memory to which the requested memory can be allocated exists in the system is performed by actual memory access. Will be implemented at the time.

On the other hand, the following memory management is performed in an HPC (High Performance Computing) environment where scientific and technical calculations are performed.
(1) Memory is allocated before execution of an application (calculation job) is started, and then execution of the application is started (pre-allocation of memory resources).
(2) Reserve a memory area that can be allocated to an application (pre-reservation of memory resources).
(3) In order to reduce the overhead caused by the operating system when handling large-scale data, the granularity for performing address translation between the virtual address and the physical address (that is, the page size) is made larger than that of a normal memory. There are various names depending on the operating system and CPU (Central Processing Unit). Here, the minimum size page supported by the address conversion mechanism of the operating system and CPU is called a normal page, and a page with a size other than the minimum size is called here. It will be called a large page.

The reason why the page size for address conversion is larger than the normal page size is mainly due to the following reason.
(1) Realization of efficient access to large-scale data In recent systems that handle virtual memory, a memory management hardware mechanism such as an MMU (Memory Management Unit) in the CPU described above and a virtual address in the operating system. A mechanism for managing the correspondence relationship with the physical address is provided. When handling large-scale data, if the granularity of address translation is fine, the number of address translation processes generated by the operating system increases, which can be an overhead when executing an application. For this reason, overhead is reduced by address conversion by collectively converting a plurality of pages.
(2) Improving the execution efficiency of a calculation job By suppressing or reducing the preemption of various resources such as the CPU time and cache of the calculation job by the operating system during the execution of the calculation job, the execution efficiency of the calculation job is improved and the memory Is assigned in advance.
(3) Suppression of job re-submission due to memory allocation failure Calculation jobs submitted by a batch system or the like often continue long-term execution, so abnormal termination occurs when memory shortage occurs due to long-term execution And the time and money costs for redoing the calculation process are high. For this reason, the memory used by the job is reserved in advance when the system is started.

For this reason, for example, in an HPC environment, a memory area used by a calculation job is reserved at system startup, and managed independently from the memory area used by an interactive process or the like. There is a case where a mechanism for guaranteeing that the memory resource is not consumed by an interactive process other than for a calculation job may be employed. As shown in FIG. 1A, immediately after the system is started, the entire memory area is allocated to the interactive process and used as the area a for the interactive process. On the other hand, as shown in FIG. 1B, when a calculation job is executed, a memory area b is also allocated to the calculation job, and the area a for the interactive process is reduced. The area b is an area used as a large page, and the area a is an area used as a normal page.

Also in Linux, a memory management mechanism called HugeTLBFS (Huge Translation Lookaside Buffer File System) is used.

This HugeTLBFS is a large page memory management mechanism for allocating a data storage area to a calculation job that operates in an HPC environment and handles large-scale data, for example. The memory (large page) managed by HugeTLBFS can be accessed with an overhead smaller than that of a normal memory because the granularity of address conversion is set larger than the granularity of address conversion in a normal memory (normal page). For this reason, programs that handle large-scale data, such as numerical computation jobs, allocate memory using HugeTLBFS.

In order to realize HugeTLBFS, Linux is roughly divided into two types of memory pools.
(1) Normal page pool Memory area used mainly by processes that perform interactive operations (2) HugeTLBFS page pool Memory area used for numerical calculation jobs, etc.

When using HugeTLBFS, the amount of memory that can be used as a data area for calculation jobs (ie, a memory area managed under the control of HugeTLBFS) is set by the system administrator at system startup. At this time, Linux shifts the amount of memory set by the administrator from the memory installed in the system from the normal page pool to the HugeTLBFS page pool, and both page pools are managed independently in the subsequent processing. That is, the memory in the HugeTLBFS page pool is reserved so that an interactive process such as a shell does not use the memory in the HugeTLBFS page pool. This is to ensure that the memory can be used reliably by reserving a memory area that is used by the calculation job to perform calculation processing using large-scale data.

In recent years, an OS used for a network server application such as FreeBSD has been equipped with a function for using a large page. Such a system is equipped with a large page mechanism designed to allow large pages to be used while reducing the administrator's effort by allocating pages without making the user aware of the presence of large pages. .

For example, in FreeBSD, based on the memory acquisition request amount from the application, a large page of the minimum size that can store the memory acquisition request amount is automatically calculated and assigned.

However, some programs operate by acquiring multiple small arrays and exchanging data between the arrays. At this time, memory fragmentation occurs by acquiring a small memory a plurality of times, and there is a problem that a large page having a larger size cannot be used later. As described above, in the system using the existing technology, there is a case where a large-scale calculation program mainly using large pages and an interactive process using a lot of small memory (that is, using normal pages) are run. Since the setting must be changed, there is a problem that operation management takes a lot of work when used in a large-scale cluster.

Japanese Patent Laid-Open No. 1-17137

Conventionally, no special consideration has been given to the distribution of large pages and normal pages, and even if large pages and normal pages can be mixed, efficient memory use may not be possible. is there. For example, if there are too many normal page memory areas, there will be insufficient large page memory areas for numerical jobs, and if there are too many large page memory areas, there will be insufficient normal page memory areas for interactive processes. Things can happen. The types and combinations of applications that run on the system can change depending on the purpose of the system and requirements during system operation, so it is difficult to avoid such a situation in advance and determine the distribution uniformly. .

Accordingly, an object of the present technology is to provide a technology for efficiently using a first size page and a second size page larger than the first size.

The memory management method includes (A) a step of receiving a request for a first size page from the operating system, and (B) a free page in a first memory area divided into pages of the first size. If it is insufficient to meet the requirement, the empty page of the second size in the second memory area divided into the second size page larger than the first size is divided into the first size page, Allocating a first size page for the request.

FIG. 1 is a diagram for explaining conventional memory allocation. FIG. 2 is a diagram illustrating an example of memory allocation in the embodiment. FIGS. 3A to 3C are diagrams for explaining the diversion of the memory area. FIG. 4 is a functional block diagram of the computer according to the first embodiment. FIG. 5 is a diagram for explaining non-reserved large page management information. FIG. 6 is a diagram for explaining reservation large page management information. FIG. 7 is a diagram for explaining normal page management information. FIG. 8 is a diagram showing a processing flow of page allocation processing according to the first embodiment. FIG. 9 is a diagram illustrating a processing flow of the page allocation processing according to the first embodiment. FIG. 10 is a diagram illustrating a processing flow of the page allocation processing according to the first embodiment. FIG. 11 is a diagram illustrating a processing flow of the page release processing according to the first embodiment. FIG. 12 is a diagram illustrating a processing flow of the page release processing according to the first embodiment. FIG. 13 is a diagram illustrating a processing flow of the page release processing according to the first embodiment. FIG. 14 is a functional block diagram of a computer according to the second embodiment. FIG. 15 is a diagram for explaining a buddy page pool. FIG. 16 is a diagram illustrating a processing flow of the page release processing according to the second embodiment. FIG. 17 is a diagram for explaining normal page management information of a buddy page pool. FIG. 18 is a functional block diagram of a computer.

FIG. 2 shows an example of memory allocation according to an embodiment of the present technology. In the present embodiment, it is used for a page having the minimum size among the page sizes supported by the operating system or the like (referred to as a normal page here, but the page having the minimum size is not necessarily set as a normal page). An area A that is divided and managed, and areas B and C that are divided and managed into a page having a size larger than that of a normal page (called a large page) are included. The area C divided for the large page is an area reserved for a specific application or the like using the large page, and an empty large page in the area is registered in the reserved large page pool. On the other hand, as a general rule, the area B divided into large pages is an area used for applications that use large pages and applications that use normal pages, and empty large pages in the areas are unreserved large pages. Registered in the page pool. Large pages in area B are allocated as large pages to applications that use large pages, but if there are not enough normal pages in area A, the large pages in area B are divided into normal page sizes. In some cases, it is assigned to an application using a normal page. The area A divided into normal pages is an area used by an application using a normal page such as an interactive process, and free normal pages in the area are registered in the normal page pool.

The use of the region B will be briefly described with reference to FIG. As shown in FIG. 3A, in a normal state in which there is no shortage of pages, the state is the same as in FIG. 2, and page allocation is performed in each of the normal page area A and the large page areas B and C. An empty large page in the reserved large page pool in area C is allocated if requested by a specific application related to the reservation. An empty large page in the non-reserved large page pool in area B is allocated if requested by an application using the large page. In addition, an empty normal page in the normal page pool in the area A is allocated if requested by an application using the normal page. When a page is released for each area, it is returned to the pool of each original area.

On the other hand, for example, when a large amount of normal page area A occurs due to an interactive job, the normal page area A becomes insufficient. In this case, as shown in FIG. 3B, an empty large page in the area B of the non-reserved large page pool is divided as a normal page area and used as an area A for normal pages. A part of the region B is diverted to the region A as schematically shown by the arrow in FIG. As a result, normal pages can be allocated even when the usage of normal pages increases, so it is possible to start many interactive jobs such as shells and improve responsiveness. To do. Large pages can be reserved for applications that use large memory. That is, it is possible to maintain a state in which large-scale data can be accessed at high speed without affecting the numerical calculation job or the like.

If a normal page is released after a large amount of normal pages is consumed by an interactive job or the like, the large page used as the released normal page is restored as shown in FIG. Return to the non-reserved large page pool area. That is, as schematically shown by an arrow in FIG. 3C, the diversion to the normal page area A is stopped, the large page is restored, and the diversion is returned to the area B. As a result, an application using a large page can sufficiently use the free area allocated to the large page in the non-reserved large page pool. Large pages in the non-reserved large page pool can be used when needed by the application. This is effective when processing such as aggregating calculation results of applications that use large pages is performed by an application that uses normal pages.

In this way, the area B is introduced and the free area allocated to the large page in the non-reserved large page pool is dynamically allocated as a large page or a normal page depending on the situation, thereby flexibly responding to the memory request. Will be able to. That is, the memory utilization efficiency is improved.

Also, the following two points can be made compatible without restarting the system or resetting operating parameters. That is, (1) it is not necessary to estimate the memory consumption of an interactive job or the like in advance. (2) High-speed large-scale memory access can be performed by a calculation job using a large page under the conditions of (1).

In particular, in a parallel computing environment, a plurality of nodes perform calculation processing in cooperation, so when resetting system parameters, it is necessary to reset a plurality of nodes. Since the recent HPC environment is composed of several thousand nodes or more, the above method is useful for reducing the cost for reconfiguration.

[Embodiment 1]
Next, a configuration in the first embodiment for performing the operation as described above will be described with reference to FIGS. The computer 100 includes a normal page management mechanism 101, a large page management mechanism 102, a page allocation mechanism 103, a page release mechanism 104, a normal page pool 106, a non-reserved large page pool 107, and a reserved large page pool 108. And have. In FIG. 4, large pages 108 a to 108 c are registered in the reserved large page pool 108,

large pages

107 a and 107 b are registered in the non-reserved large page pool 107, and the normal page pool 106 is registered. , Normal pages 106a to 106h are respectively registered. The number of pages to be registered is arbitrary, and the mode of page registration shown in FIG. 4 is an example.

The page allocation mechanism 103 is also called a page allocation handler, and is included in the kernel of the operating system of the computer 100.

The non-reserved large page pool 107 stores non-reserved large page management information as shown in FIG. 5, for example. The non-reserved large page management information includes the total number of non-reserved large pages, the number of allocated pages, the number of free pages, the number of diverted pages, and the page list. The total number of pages is a total value of the number of allocated pages, the number of empty pages, and the number of diverted pages. The number of allocated pages is the number of pages allocated to an application using a large page. The number of free pages is the number of unallocated pages. The number of diverted pages is the number of pages that have been diverted to normal pages. The page list is a list of free pages managed by the non-reserved large page pool 107, and each free page is sorted in the order of the top physical address, for example.

Furthermore, the reserved large page pool 108 stores reserved large page management information as shown in FIG. 6, for example. The reserved large page management information includes the total number of reserved large pages, the number of allocated pages, the number of empty pages, and a page list. The total number of pages is a total value of the number of allocated pages and the number of empty pages. The number of allocated pages is the number of pages allocated only to a calculation job using a large page. The number of empty pages is the number of unallocated pages. The page list is a list of free pages managed in the reserved large page pool 108, and each free page is sorted in the order of the top physical address, for example. Note that a memory size corresponding to the number of pages may be used instead of the number of pages.

In addition, normal page management information as shown in FIG. 7 is stored in the normal page pool 106, for example. The normal page management information includes the total number of normal pages, the number of allocated pages, the number of empty pages, a page list, and a diversion threshold. The number of allocated pages is the number of pages allocated to an application that uses normal pages. The number of empty pages is the number of unallocated pages. The page list is a list of free pages managed in the normal page pool 106, and each free page is sorted in the order of the top physical address, for example. The diversion threshold value is a threshold value for the number of continuous free normal pages when a continuous free normal page is converted into a large page and returned when there is a large page diverted to a normal page. The diversion threshold value may be (large page size) / (normal page size) or a larger value.

Next, the operation of the computer 100 shown in FIG. 4 will be described with reference to FIGS. First, the page allocation mechanism 103 outputs a page allocation request including the requested number of normal pages to the normal page management mechanism 101 in response to a request from an application or the like or occurrence of another normal page allocation event (FIG. 8: Step S1). Since this step is the same as the conventional one, further explanation is omitted.

The normal page management mechanism 101 receives a page allocation request including the number of requested pages for the normal page from the page allocation mechanism 103 (step S3), and the normal page free page included in the normal page management information of the normal page pool 106. The number is confirmed (step S5). Here, if the number of empty pages is equal to or greater than the number of requested pages (step S7: Yes route), the same processing as normal page allocation is performed. That is, the normal page management mechanism 101 reduces the number of empty pages by the number of requested pages in the normal page management information, increases the number of allocated pages by the number of requested pages, and creates normal pages for the number of requested pages from the normal page list this time. (Step S9). Specifically, the top physical addresses of normal pages for the number of requested pages are read from the normal page list, and the top physical addresses of normal pages for the number of requested pages read are deleted from the normal page list.

Then, the normal page management mechanism 101 outputs a page allocation result including information on the top physical address of the allocated normal page to the page allocation mechanism 103 (step S11). The page allocation mechanism 103 receives the page allocation result from the normal page management mechanism 101 (step S13), and allows the requesting application to use the allocated normal page. Since this process is the same as that of the prior art, further explanation is omitted.

On the other hand, when the number of requested pages is larger than the number of free pages (step S7: No route), since the current free page cannot be allocated to the requested page, a large page is diverted. Specifically, the normal page management mechanism 101 outputs a page allocation request including the requested number of pages to the large page management mechanism 102 (step S15). In response to this, the large page management mechanism 102 receives a page allocation request including the number of requested pages from the normal page management mechanism 101 (step S17). The processing shifts to the processing in FIG.

9, the large page management mechanism 102 checks the number of large pages that are included in the large page management information of the non-reserved large page (step S19). In addition, the large page management mechanism 102 converts the number of empty large page pages into the number of normal page pages (step S21). For example, it is calculated by (number of empty pages of large pages) * (page size of large pages) / (page size of normal pages). (“*” Represents integration.) In this step, the requested number of normal pages may be converted into a large page. Specifically, (the number of requested pages of normal pages) * (page size of normal pages) / (page size of large pages) is calculated, and a value less than an integer is rounded up, for example.

Then, the large page management mechanism 102 determines whether or not the number of converted pages of empty large pages is equal to or greater than the number of requested pages (step S23). When the number of requested pages is converted into the number of large pages, it is checked whether the number of free pages is equal to or greater than the number of requested pages.

If the unreserved large page pool 107 also has a small number of free large pages, there may be cases where normal pages for the number of requested pages cannot be generated. In such a case, the large page management mechanism 102 outputs an allocation failure to the normal page management mechanism 101 (step S25). When receiving the notification of allocation failure from the large page management mechanism 102, the normal page management mechanism 101 further outputs the allocation failure to the page allocation mechanism 103 (step S27). The page allocation mechanism 103 receives a notification of allocation failure from the normal page management mechanism 101 (step S29), and notifies the requesting application or the like of the allocation failure. Since the subsequent processing is the same as the conventional one, further explanation is omitted.

On the other hand, if the converted number of empty large pages is equal to or greater than the requested number of pages, the large page management mechanism 102 calculates the number of diverted pages of the large page to be rotated now (step S31). The number of diverted pages is calculated by (number of requested normal pages) * (page size of normal pages) / (page size of large pages). Thereafter, the large page management mechanism 102 decreases the number of empty pages in the non-reserved large page management information of the non-reserved large page pool 107 by the number of diverted pages of the large page to be rotated now, and the diverted page from the page list of the non-reserved large page. A number of large pages are allocated to the current request (step S33). Specifically, from the page list of the non-reserved large page management information, the start physical address of each large page for the number of diverted pages is read, and the page management information of each read large page is displayed as the page list in the non-reserved large page management information Delete from. The processing shifts to the processing in FIG.

10, the large page management mechanism 102 increases the number of diverted pages of the non-reserved large page management information by the number of diverted pages (step S35). Further, the large page management mechanism 102 performs a process of dividing the large page assigned in step S33 into normal pages (step S37). Specifically, the page management information of each normal page is accumulated by dividing the large pages that have been allocated in order, until the required number of pages is reached, divided by the normal page size. In some cases, normal pages can be generated beyond the number of requested pages. Similarly, in this case, page management information for each normal page is stored as page management information for surplus normal pages, separated by the normal page size. To do. This is because it is registered in the normal page pool later.

Then, the large page management mechanism 102 outputs a page allocation result including the top physical addresses of normal pages for the number of allocated requested pages to the normal page management mechanism 101 (step S39). The normal page management mechanism 101 receives a page allocation result including the top physical addresses of normal pages for the number of requested pages allocated from the large page management mechanism 102 (step S41), and in the normal page management information of the normal page pool 106 The total number of pages and the number of allocated pages are increased by the number of requested pages.

Further, the normal page management mechanism 101 outputs a page allocation result including information on the top physical address of the normal page included in the received page allocation result to the page allocation mechanism 103 (step S43). The page allocation mechanism 103 receives the page allocation result from the normal page management mechanism 101 (step S45), and allows the requesting application or the like to use the allocated normal page. Since this process is the same as that of the prior art, further explanation is omitted.

Further, the large page management mechanism 102 determines whether or not normal pages exceeding the requested number of pages have been generated (step S47). If a normal page having the same number of pages as the requested number of pages is generated in step S37, no further processing is necessary, and the processing is terminated.

On the other hand, when a normal page exceeding the number of requested pages is generated, the large page management mechanism 102 outputs the above-described first physical address of the surplus normal page to the normal page management mechanism 101 (step S49). On the other hand, the normal page management mechanism 101 receives the information of the head physical address of the surplus normal page from the large page management mechanism 102 (step S50). Then, the normal page management mechanism 101 adds page management information corresponding to the address of the received surplus normal page to the page list in the normal page management information (step S51). Further, the normal page management mechanism 101 adds the number of surplus normal pages to the total number of pages and the number of empty pages in the normal page management information (step S53).

In this way, surplus normal pages can be used effectively. That is, as shown in the transition from FIG. 3A to FIG. 3B, the memory area of the unused large page is diverted to the normal page memory area, so that the memory can be effectively used. In addition, since the large page in the memory area managed by the large page and allocated exclusively for the calculation job or the like is not diverted, the influence on the processing of the calculation job or the like can be reduced.

Next, processing when normal page release occurs will be described with reference to FIGS. First, the page release mechanism 104 is a release page list that is a list of the number of released pages of normal pages and the start physical address of the normal page to be released in response to the release by an application or the like or the occurrence of another normal page release event. Is output to the normal page management mechanism 101 (FIG. 11: step S61). Since this step is the same as the conventional one, further explanation is omitted.

In response to this, the normal page management mechanism 101 receives a page release request including the number of released pages of the normal page and the release page list from the page release mechanism 104 (step S63). Then, the normal page management mechanism 101 registers the page management information of the normal pages included in the received release page list in the page list in the normal page management information of the normal page pool 106 (step S65). Further, the normal page management mechanism 101 decreases the number of allocated pages in the normal page management information by the number of released pages, and further increases the number of free pages by the number of released pages (step S67).

Then, the normal page management mechanism 101 acquires the number of diverted pages in the non-reserved large page management information of the non-reserved large page pool 107 from, for example, the large page management mechanism 102, and confirms whether the acquired number of diverted pages is 0 ( Step S69). When the normal page management mechanism 101 can directly access the number of diverted pages in the non-reserved large page management information, the normal page management mechanism 101 may directly access the number of diverted pages in the non-reserved large page management information.

If the number of diverted pages is 0, the released normal page is not returned to the non-reserved large page pool, so the process ends. On the other hand, when the number of diverted pages is not 0, the processing shifts to the processing in FIG.

The normal page management mechanism 101 scans the page list in the normal page management information of the normal page pool 106, and determines whether the number of consecutive free normal pages is equal to or greater than the diversion threshold in the normal page management information of the normal page pool 106. Confirm (step S71). Instead of checking whether the number of continuous free normal pages is equal to or greater than the diversion threshold, it may be checked whether there are continuous free normal pages for the large page size. Also, when checking whether there are continuous free normal pages for the large page size, it is checked whether there are free normal pages that are continuous for one large page, but it is continuous for two pages or more. It may be confirmed whether there is an empty normal page.

If there is no continuous free page or if the continuous free normal page is less than the diversion threshold in the normal page management information (step S73: Yes route), the process ends here. In step S71, instead of checking whether the number of continuous free normal pages is equal to or greater than the diversion threshold, when checking whether there are continuous free normal pages for the large page size, When there are no consecutive free normal pages, the process cannot be returned to the non-reserved large page pool 107, and the process ends here.

On the other hand, when there are continuous free pages and the number of continuous free normal pages is less than the diversion threshold value in the normal page management information (step S73: No route), the normal page management mechanism 101 uses the normal page management information. The number of empty pages and the total number of pages are reduced by the number of normal pages that are returned (step S75). In step S71, instead of checking whether the number of continuous free normal pages is equal to or greater than the diversion threshold, when checking whether there are continuous free normal pages for the large page size, When there are consecutive free normal pages, the normal page management mechanism 101 reduces the number of empty pages and the total number of pages in the normal page management information by the number of normal pages for return. Further, the normal page management mechanism 101 extracts a list of empty normal pages having continuous large page sizes from the page list in the normal page management information (step S76). The list of free normal pages is a list of the top physical addresses of continuous free normal pages having a large page size. Further, the normal page management mechanism 101 outputs a list of empty normal pages having continuous large page sizes to the large page management mechanism 102 (step S77).

On the other hand, the large page management mechanism 102 receives a list of continuous free normal pages having a large page size from the normal page management mechanism 101 (step S79), and converts the free normal page into a large page (step S81). That is, continuous empty normal pages are handled as continuous areas. The processing shifts to the processing in FIG.

The large page management mechanism 102 registers the top physical address of the generated large page in the page list in the non-reserved large page management information of the non-reserved large page pool 107 (step S83). In addition, the large page management mechanism 102 reduces the number of diverted pages in the non-reserved large page management information by the number of returned pages (“1” in the above example, although it may be “2” or more). Step S85). Further, the large page management mechanism 102 increases the number of empty pages in the non-reserved large page management information by the number of returned pages (step S87). Then, the process ends.

By performing the processing as described above, as shown in the transition from FIG. 3 (b) to FIG. 3 (c), when a non-reserved large page is diverted, a margin is generated in the normal page. The normal pages are bundled in the large page size and returned to the non-reserved large page pool 107. This makes it possible to prepare for a large page allocation request. Further, it is possible to cope with a case where the request for normal pages increases again thereafter.

For requests from applications that use large pages, the large page management mechanism 102 allocates free pages from the non-reserved large page pool 107 as usual. If the request is from a calculation job or the like, free pages are allocated from the reserved large page pool 108 as usual. Therefore, further description of the request for large pages is omitted.

[Embodiment 2]
The normal page management method includes the method described below in addition to the method described above.

FIG. 14 shows a computer 200 according to the present embodiment. The computer 200 includes a normal page management mechanism 201, a large page management mechanism 102, a page allocation mechanism 103, a page release mechanism 104, a buddy page pool 206, a non-reserved large page pool 107, and a reserved large page pool 108. And have. The same reference numerals are given to functions similar to those of the computer 100 shown in FIG. That is, the normal page management mechanism 201 and the buddy page pool 206 are different parts from the first embodiment.

In particular, the buddy page pool 206 is different from the normal page pool 106 in the management mode of empty pages. As schematically shown in FIG. 14, in order to avoid fragmentation of physical pages that occurs during system operation, that is, thrashing, empty pages are managed in units of powers of 2 (referred to as buddies). That is, if the power is 2 to the power of “0”, one empty page is managed, and in the buddy page pool 206, the empty pages 2061a to 2061e are managed in the order of physical addresses. If the power is 2 to the power of “1”, two consecutive empty pages are managed. In the buddy page pool 206, two consecutive empty pages 2062a to 2062d are managed in the order of physical addresses. Further, if 2 is the power of “2”, four consecutive pages are managed. In the buddy page pool 206, four consecutive empty pages 2063a to 2063c are managed in the order of the top physical address.

FIG. 15 shows the management state of the buddy page pool 206 in more detail. The buddy page pool 206 is introduced to prevent memory fragmentation (thrashing) in units of pages based on the rule of thumb that many page allocation requests in the system are made in units of powers of 2. . As shown in FIG. 15, the start physical address of the free page of every two ⁰ pages are registered in the list of free_list [0]. For example, if the size of one page is 8 Kbytes, a page for every 8 Kbytes is registered. In addition, 2 ¹ of the start physical address of contiguous free pages, are registered in the list of free_list [1]. For example, data of up to 8 Kbytes × 2 = continuous 16 Kbytes can be stored.

In addition, top physical address of a free page of every 2 ² pages are registered in the list of free_list [2]. For example, 8 Kbytes × 4 = continuous 32 Kbytes of data can be stored. In addition, the top physical address of a free page of every 2 ³ page is registered in the list of free_list [3]. For example, data of up to 8 Kbytes × 8 = continuous 64 Kbytes can be stored. In addition, top physical address of a free page of every 2 ⁴ page is registered in the list of free_list [4]. For example, 8 Kbytes × 16 = continuous 128 Kbytes of data can be stored. Furthermore, the head physical address of the empty page for every ²⁵ pages is registered in a list called free_list [5]. For example, it is possible to store data of up to 8 Kbytes × 32 = continuous 256 Kbytes. In this way, the head physical address of the empty page for every 2 ¹⁰ pages is registered in a list called free_list [10]. For example, 8 Kbytes × 2 ¹⁰ = continuous 8 Mbytes of data can be stored.

The buddy page pool 206 can handle page allocation and release in consecutive page units.

Even when such a buddy page pool 206 is used, a process similar to that of the first embodiment is performed for a normal page allocation request. However, the method of searching for continuous free pages that satisfy the requested number of pages differs depending on the management method of the buddy page pool 206. That is, a search is made for a continuous free page having the smallest power of 2 that is equal to or greater than the number of requested pages, and allocation is performed from the continuous free pages. If there is no continuous free page that is equal to or greater than the requested number of pages, as in the first embodiment, a large page is converted to a normal page and handled.

On the other hand, the page release is changed as described below.

Referring to FIG. 16, the processing content when releasing the page will be described. First, the page release mechanism 104 includes the number of released pages of normal pages and a released page list that is a list of the first physical addresses of normal pages to be released in response to release by an application or the like or occurrence of another normal page release event. A page release request is output to the normal page management mechanism 201 (FIG. 16: step S101). Since the processing in this step is the same as the conventional one, further explanation is omitted.

In response to this, the normal page management mechanism 201 receives a page release request including the number of released pages of the normal page and the release page list from the page release mechanism 104 (step S103). Then, the normal page management mechanism 201 performs a page return process to the buddy page pool 206 (step S105). For example, if the release page is

page

2, 2 ¹ page page list to search for and verify contiguous pages to be linked there. If there is no page to be continuous to be consolidated, to register as it is 2 ¹ page of the page list. On the other hand, it will be consolidated if there is a page to be continuous to be consolidated, to search the page list of 2 page ^2, to check whether the page to be continuous with the coupling exists. Such processing is repeated until a continuous page is not found when connected. Also, a process for increasing the number of empty pages is performed. In the example of FIG. 7, the total number of empty pages is managed in the normal page management information. However, for example, as shown in FIG. 17, the number of empty pages is managed for each buddy. However, the number of vacancies in units of buddies may be managed.

In step S105, the normal page management mechanism 201 confirms whether the final return destination buddy size matches the large page and the number of diverted pages is 1 or more (step S107). The normal page management mechanism 201 acquires, for example, the number of diverted pages in the non-reserved large page management information of the non-reserved large page pool 107 from the large page management mechanism 102 or the non-reserved large page management information of the non-reserved large page pool 107 Get directly accessed the number of diverted pages in. If either the condition that the final return destination buddy size matches the large page or the condition that the number of diverted pages is 1 or more is not satisfied, the process ends.

On the other hand, if the final return destination buddy size matches the large page in step S105 and the number of diverted pages is 1 or more, the normal page management mechanism 201 manages the normal page management of the buddy page pool 206. In the information, the number of empty pages of buddies whose size matches the large page is subtracted by the number of return pages (step S109). Also, the total number of pages in the normal page management information is subtracted by the number of return pages. Then, the process proceeds to step S77 in FIG.

The introduction of the buddy page pool 206 in this way makes part of the page return process different, but the same is true for the other processes.

Even if the buddy page pool 206 is introduced in this way, the non-reserved large page pool 107 can be introduced and the memory utilization efficiency can be increased as in the first embodiment.

The embodiment of the present technology has been described above, but the present technology is not limited to this. For example, the division of roles between the normal page management mechanism and the large page management function shown in the processing flow is merely an example, and it may be possible to change the processing subject of each step. In some cases, the normal page management mechanism and the large page management mechanism are integrated.

The

computers

100 and 200 described above are information processing apparatuses such as a computer as shown in FIG. 18 as a whole, and include a memory 2501, a CPU 2503 (also referred to as a processor), and a hard disk drive (HDD: Hard Disk Drive). 2505, a display control unit 2507 connected to the display device 2509, a drive device 2513 for the removable disk 2511, an input device 2515, and a communication control unit 2517 for connecting to a network are connected by a bus 2519. The operating system and the application program for executing the processing in this embodiment are stored in the HDD 2505, and are read from the HDD 2505 to the memory 2501 when executed by the CPU 2503. The CPU 2503 controls the display control unit 2507, the communication control unit 2517, and the drive device 2513 according to the processing content of the application program, and performs a predetermined operation. Further, data in the middle of processing is mainly stored in the memory 2501, but may be stored in the HDD 2505. In an embodiment of the present technology, an application program for performing the above-described processing is stored in a computer-readable removable disk 2511 and distributed, and installed from the drive device 2513 to the HDD 2505. In some cases, the HDD 2505 may be installed via a network such as the Internet and the communication control unit 2517. Such a computer apparatus realizes various functions as described above by organically cooperating hardware such as the CPU 2503 and memory 2501 described above and programs such as an operating system and application programs. To do.

Typically, the normal page management mechanism and the large page management mechanism in the present embodiment are implemented as a part of the operating system, for example. That is, it is executed by the CPU 2505 to realize the functions described above.

The above-described embodiment can be summarized as follows.

In the memory management method according to the present embodiment, (A) a step of receiving a request for a page of the first size, and (B) an empty page in the first memory area divided into pages of the first size. If there is a shortage to meet the above request, the empty page of the second size in the second memory area that is divided into pages of the second size larger than the first size is divided into pages of the first size. And allocating a first size page for the request.

If there is a shortage of pages of the first size in this way, empty pages of the second size are diverted, so that the memory utilization efficiency can be improved.

The allocation step described above includes (B1) determining whether the number of first requested pages included in the request exceeds the number of free pages in the first memory area; and (B2) first requested page. If the number exceeds the number of free pages in the first memory area, the second requested page number obtained by converting the first requested page number into the second size page number is the second in the second memory area. A step of securing a page having a size of 2, and a step of (B3) dividing the secured second size page into pages of the first size and allocating them to the request by the number of first request pages. May be included. In this way, the second size page can be specifically transferred.

Furthermore, the memory management method according to the present embodiment includes (C) a step of detecting the release of a first size page, and (D) a first size of free space including the released first size page. When the number of continuous pages is equal to or greater than a predetermined threshold and the second size page in the second memory area is diverted as the first size page, the empty continuous page of the first size To generate a second size page and register it as a free page of the second size. If it does in this way, the page of the 2nd size which was diverted can be returned, and it can use now for the utilization of the page of the 2nd size.

Also, there are cases where the first size of free pages is managed for each continuous free page size. In such a case, the memory management method is generated from (E) the step of detecting the release of the first size page, and (F) the released first size page and the free continuous page that already exists. If the size of the second free continuous page is the second size and the second size page in the second memory area is diverted as the first size page, the second free continuous page The method may further include a step of concatenating the pages to generate a second size page and registering it as a free page of the second size. For example, even when a buddy page pool mechanism managed with a power-of-two size is adopted, it is possible to appropriately cope with it.

Note that the third memory area divided into pages of the second size may be reserved as an area dedicated to a specific job using a large page.

Further, the computer according to the present embodiment has (A) a first memory area divided into pages of a first size and a second page divided into pages of a second size larger than the first size. And (B) in response to a request for a page of the first size, if there are not enough free pages in the first memory area to satisfy the request, the second memory area A memory management mechanism that divides an empty page of size 2 into pages of a first size and allocates a page of the first size in response to the request.

It is possible to create a program for causing a computer to carry out the processing described above, such as a flexible disk, an optical disk such as a CD-ROM (Compact Disk-Read Only Memory), a magneto-optical disk, and the like. The data is stored in a computer-readable storage medium or storage device such as a disk, a semiconductor memory (for example, ROM), and a hard disk. Note that data being processed is temporarily stored in a storage device such as a RAM.

Claims

Receiving a request for a first size page from the operating system;
If empty pages in the first memory area divided into pages of the first size are insufficient to meet the request, the second page is divided into pages of a second size larger than the first size. An allocation step of dividing the second size free page in the second memory area into the first size page and allocating the divided first size page in response to the request;
Memory management program for causing a computer to execute
The assigning step comprises:
Determining whether the number of first requested pages included in the request exceeds the number of free pages in the first memory area;
When the first requested page number exceeds the number of free pages in the first memory area, the second requested page obtained by converting the first requested page number into the number of pages of the second size. Securing a number of pages of the second size in the second memory area,
Dividing the secured second size page into first size pages and assigning to the request as many as the first requested pages;
The memory management program according to claim 1, comprising:
Detecting the release of the first size page;
The number of free continuous pages of the first size including the released first size pages is equal to or greater than a predetermined threshold, and the second size pages in the second memory area are the first size. If the page is diverted as a page of size 1, a step of generating the second size page by concatenating the free continuous pages of the first size and registering it as a free page of the second size When,
The memory management program according to claim 1, further comprising: causing the computer to execute.
The empty pages of the first size are managed for each continuous empty page size,
Detecting the release of the first size page;
The size of the second free continuous page generated from the released page of the first size and the free continuous page that already exists is the second size, and the second size in the second memory area. When the page of the size is used as the page of the first size, the second continuous page is concatenated to generate the second size page, and the second size of the free page is generated. Register as a page,
The memory management program according to claim 1, further causing the computer to execute the following.
Receiving a request for a first size page from the operating system;
If empty pages in the first memory area divided into pages of the first size are insufficient to meet the request, the second page is divided into pages of a second size larger than the first size. An allocation step of dividing the second size free page in the second memory area into the first size page and allocating the divided first size page in response to the request;
Is executed by a computer.
The assigning step comprises:
Determining whether the number of first requested pages included in the request exceeds the number of free pages in the first memory area;
When the first requested page number exceeds the number of free pages in the first memory area, the second requested page obtained by converting the first requested page number into the number of pages of the second size. Securing a number of pages of the second size in the second memory area,
Dividing the secured second size page into first size pages and assigning to the request as many as the first requested pages;
The memory management method according to claim 6, further comprising:
Detecting the release of the first size page;
The size of the first continuous free page including the released first size page is equal to or larger than a predetermined threshold value, and the second size page in the second memory area is the first size. The second size page is generated by concatenating the first continuous free pages, and is registered as the second free page. ,
The memory management method according to claim 5, wherein the computer further executes.
The empty pages of the first size are managed for each continuous empty page size,
Detecting the release of the first size page;
The size of the second free continuous page generated from the released page of the first size and the existing free continuous page is the second size, and the second size in the second memory area is the second size. When a page of size is diverted as the page of the first size, the second vacant continuous page is concatenated to generate the second size page, and the second size vacant page As a step to register as
The memory management method according to claim 5, wherein the computer further executes.
A memory including a first memory area divided into pages of a first size and a second memory area divided into pages of a second size larger than the first size;
In response to a request for a first size page, if there are not enough free pages in the first memory area to satisfy the request, the second size free page in the second memory area is A memory management mechanism that divides the page into one size and allocates the first page divided in response to the request;
Having a computer.