JP3431936B2 - Memory management method and device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a memory management method and device in an information processing device or the like using virtual memory.

[0002]

2. Description of the Related Art Conventionally, in an operating system of an information processing apparatus using memory management by virtual memory, an execution file is loaded by a demand paging method. In that case, if a page fault occurs during paging, paging in one page for the page in which the page fault occurred increases the number of page faults and reduces the processing capability of the information processing device. , The method of putting pages in collectively is proposed. Among them, the demand pre-paging method of performing pre-paging in addition to demand paging is an efficient paging method. A typical one of the demand pre-paging methods is to load pages before and after a page having a page fault called clustering, which is adopted in 4.3BSD and OSF / 1, as a page cluster at a time. There is a method, and there is a simple pre-paging method that simply loads the page beyond the page where the page fault occurred. In the clustering method, as shown in FIG. 18, a cluster including a page in which a page fault has occurred (in the figure, 4 pages are 1 page clusters) is loaded at one time. In the simple pre-paging method, as shown in FIG. 19, several pages (three pages in the figure) after the page faulted are loaded together with the page faulted page.

In addition, when page replacement is performed when there is no empty page at the time of processing a page fault,
A process of writing a page to be paged out and a page out to a secondary storage device, and a process of transferring a page replaced from the main storage from a secondary storage device or an information processing device connected by a network. The page to be replaced is selected from the past usage history information of the page without considering the time required for the page replacement. That is, the page that is the target of page replacement is the LRU (Last Res
It is selected according to an approximation of the method called "cently Used". This LRU leaves the most recently referenced page in main memory and targets the page that has not been referenced for the longest time as a page replacement target, and all pages are treated equally.

[0004]

As described above, when the demand pre-paging method is adopted, compared with the case where the method of simply page-in every page is adopted. , Consumes many pages quickly.
Furthermore, there is no guarantee that pages other than the page that actually caused the page fault among the pages that have been paged in by pre-paging or clustering will be referenced immediately, and as a result, unnecessary page loading was performed. Become. Therefore, when the demand pre-paging method is used when the number of free pages decreases or the load of the entire system becomes heavy, the processing capacity of the information processing device decreases. there were.

Further, in an information processing apparatus having a page replacement method in which all pages are equally selected for page replacement selection, if pages that must be written in the secondary storage device are replaced, they are rewritten. It takes more processing to write back to the secondary storage device than to replace a page that is not stored. Also, if a page that has data is replaced in a secondary storage device that takes time to reload or an information processing device connected by a slow network, the data is stored in the secondary storage device that is relatively fast. It takes more time than reloading the existing page if you have to use the page again. For this reason, there is a disadvantage that the page-out process for page replacement and the page-in process for the entire information processing apparatus take time.

[0006]

A memory management method of the present invention is a memory management method for an information processing apparatus using virtual memory, wherein a setting step of setting a priority corresponding to a processing cost at the time of page replacement for each page. And a replacement target number setting step of determining the number of page replacements required, and a priority target number setting step of determining the number of page replacements for each priority according to the number determined in the replacement target number setting step. And a selection step of selecting a page to be a page replacement target based on the number obtained in the priority-based target number setting step.

Further, the memory management device of the present invention is a memory management device of an information processing device using virtual memory, and a setting means for setting priority to each page according to a processing cost at the time of page replacement, and page replacement. A target number setting means for determining the required number, a target number setting means for each priority for determining the number of page replacements for each priority according to the number calculated by the number setting target for replacement, and the priority And a selection unit for selecting a page to be replaced with a page based on the number obtained by the target number setting unit for each time.

[0008]

[0009]

【Example】

(Embodiment 1) An embodiment of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram of an information processing apparatus according to an embodiment of the present invention. In the figure, reference numeral 1 is a CPU, which controls the apparatus. 2 is a main memory, RO
It consists of M201 and RAM202. ROM201 is C
It stores various programs executed by the PU 1 and processes for implementing the present invention. The RAM 202 stores various information and programs loaded from the secondary storage device 3, the CPU 1
Used as a work area. A secondary storage device 3 stores various data, programs, and the like.
A keyboard 4 as an input device is used to input various data and the like. Reference numeral 5 denotes a display device on which a result processed by the CPU 1 and the like are displayed. 6 is a network, I
It is connected to the CPU 1 via an / O interface (not shown).

FIG. 2 is a flowchart showing the processing operation of the first embodiment of the present invention. In the figure, first, the load of the entire information processing apparatus is detected (S21). In order to detect the load of the entire information processing apparatus, for example, a load average, the number of free pages, the number of active processes, the frequency of reading the disk, and the like are obtained. Further, a plurality of such numerical values may be combined and obtained.

The number of pages to be pre-paged is set according to the load obtained in S21 (S22). The page loading is executed according to the number of pages set in S22 (S23).

FIG. 3 is a flowchart of a process for detecting the load on the entire information processing apparatus and setting the number of prepages. In the figure, the load of the entire system is detected by the load average and the number of free pages (S31, S32), and the load average and the number of free pages are respectively compared with the threshold (S33). Threshold with load average (L
When the number of pages is equal to or more than thresh) or the number of free pages is equal to or less than a certain threshold value (Pthresh), pre-paging is stopped (S34, S36), and if not, N pages are pre-paged (S35, S36).

Here, even if the number N of prepages is not a fixed numerical value, but the number of prepages is changed according to a function (here, it is based on the number of active processes) as shown in FIG. Good. That is, in FIG.
When the number of active processes is 1, the number of pre-pages is 6 pages, when the number of active processes is more than 1 and less than 3, the number of pre-pages is 4, the number of active processes is more than 3 and less than 6 Sometimes the number of pre-pages is 2,
When the number of active processes is more than 6 and less than 10, the number of pre-pages is 1 page and the number of active processes is 10.
If the number is larger than the number, it is assumed that the demand prepage is not performed. As a result, when a large number of active processes exist, it is possible to prevent the processing capacity of the entire system from being lowered by loading a large amount of data at one time.

Further, not only the load in the system is detected, but also the load of the network which is connected and is affected by the load may be detected so that the unnecessary transfer to the network is not required. . For example, in a system in which most files are provided by an information processing device connected by a network, such network load information is important. For obtaining the load information of the network, for example, the response time of the past data request is stored in the RAM 20.
It can be realized by storing in 2 and placing it by hand.
Further, the load information in the system as described above and the additional information such as the network may be used in combination.

(Second Embodiment) Next, a second embodiment of the present invention will be described.

FIG. 5 is a flow chart showing the operation of the second embodiment of the present invention. In the figure, first, the priority of page-out is set (S51). The page is used according to the setting (S52). When setting a priority, whether a page must be written to secondary storage when it becomes a replacement target, and how long it takes to reuse the contents held by that page Is reflected in the priority.

FIG. 6 is a diagram showing an example of priority setting. In the figure, the page replacement with the highest priority is the page that holds the file on the local disk without writing, and the second highest priority is the page that holds the file on the local disk. a write also Nodea is, then is of no write on the page that holds a file on a server connected on the network, the lowest-priority write a page that holds the file on the network There is something.

This priority setting need not be performed at the time of page allocation, but may be dynamically set at the time of setting the page replacement target or at the time of executing the program.

In FIG. 5, the page replacement target is selected according to the priority set for the page (S53).
FIG. 7 is a flowchart showing a process (S53 in FIG. 5) of selecting a page replacement target according to the priority set for a page. In the figure, if there is one or more high priority pages, the page is targeted for page replacement (S71 to S74). In S71 to S74, the page replacement target is selected from the pages with high priority. Next, when there are a plurality of pages with the same set priority, the target is selected from those pages according to the conventional page replacement selection method (such as approximation of LUR).

As the priority setting, as shown in FIG. 8, the page priority may be set according to the type of the secondary storage device. In the figure, when a ram disk, a hard disk, a magneto-optical disk, and a flexible disk are present as secondary storage devices, the access time of them is taken into consideration and the priority of the page holding the file on the ram disk is the highest. Then, the page holding the file on the hard disk is given a higher priority, and the page holding the file on the magneto-optical disk is given a higher priority next. The page holding the file on the flexible disk has the lowest priority.

Further, the number of pages to be replaced may be changed according to the priority. FIG. 9 is a flowchart showing a process of changing the number of pages to be replaced according to the priority. In the figure, first,
The number of page replacements is detected (S91). Accordingly, the number of pages to be replaced from the page of priority 1, the number of pages to be replaced from the page of priority 2, the number to be replaced from the page of priority 3, and the number to be replaced from the page of priority 4 are determined. Set (S92). The number of pages set in this way is selected from pages of the same priority by a method such as LRU approximation. Therefore, pages with higher priority are more likely to be replaced with pages, and pages with low priority that are rarely used are subject to page replacement, and page replacement advantageous in terms of processing cost and frequency of use is performed ( S93).

Further, the following can be applied to a system (4.3 BSD Unix, etc.) in which page data once loaded is always written to a local disk. In such a system, the loading time when the page is replaced and then reused is the same for any page.
However, if a page once written to the local disk is not rewritten in the main memory, it is not necessary to write it back to the local disk if the page is targeted for page replacement. That is, in the case of such a system, by increasing the priority of pages that have been once written to the local disk and have not been rewritten, it is possible to obtain an effect of preventing a decrease in the processing capacity of the system. .

As described above, the priority may be set in consideration of either the processing cost of page-out or the processing cost of loading at the time of reuse, or the priority may be set in consideration of both of them. Good.

Further, in the case of a system in which files on the secondary storage device are loaded on the main storage and no rewriting is allowed in them, for example, in the case of Mach or the like (Carnebigy Mellon University), Set the priority as follows. For files on secondary storage, they do not need to be paged out because no rewriting occurs. In other words, for the page that holds the contents of the file, it is only necessary to consider the processing cost when reloading the contents. Also, pages etc. for data areas (including stack areas) that are dynamically allocated in the execution program must be paged out to the local secondary storage device. Considering the above two types of processing costs, FIG.
As shown in 0, the priority of the page is considered according to the processing cost at the time of reloading, and for pages (data, stack area) that need to be paged out, the degree of request to the local secondary storage device is set. Upon detection, the priority may be dynamically changed to the position 1 or the position 2 in FIG.

(Third Embodiment) Next, a third embodiment of the present invention will be described.

FIG. 11 is a flow chart showing the priority setting process of the third embodiment. In the figure, first, if a page is to be replaced, whether it must be written to the secondary storage device, and how long it takes to reuse the content held by the page The priority is set in consideration of (S111). Then, the page is put into the active queue according to the priority (S11).
2).

FIG. 13 shows an example of setting the priority. In the figure, the highest priority for page replacement is the page that holds the file on the local disk and is not rewritten, and the second highest priority is for the remote disk connected to the network. Pages that hold files on the local disk that are not rewritten, then pages that hold files on the local disk that can be rewritten, and the lowest priority page that holds files on remote disks connected to the network There is rewriting on the page. Various patterns can be considered for these depending on the network used and the secondary storage device connected locally. Also, it is not necessary to set this priority when allocating pages,
It may be set dynamically when the page replacement target is set or when the program is executed.

FIG. 12 is a flow chart showing the process when selecting a page replacement target according to the priority set for a page. In the figure, it is determined whether the number of pages connected to the free queue has decreased (S12).
1). When there are few pages connected to the free queue, page replacement is performed to set the number of pages to be free (N) (S122), and pages are taken out from the inactive queue (S123). Next, it is determined whether or not the reference bit of the fetched page is turned on (S124). If the reference bit is not turned on, the page is replaced and page replacement is performed (S126). If the reference bit is on,
It is returned to the active queue (S125). It is determined whether N pages have been replaced (S127), and if not, S123 and subsequent steps are executed. When N pages have been replaced, it is determined whether the inactive queue has become shorter than a certain constant value (S128). inact
If the IVE queue becomes shorter than a certain constant value,
The number of pages (M) to be put from the active queue to the inactive queue is set (S129), and a plurality of act
A page is taken out of the Ive queue, the reference bit of the page is cleared, and the page is put in the inactive queue (S
130).

FIG. 14 shows the FIFO with seco.
3 queues used in the end change (act
It is a conceptual diagram which shows the relationship of (active, inactive, free). A page used by the program currently being executed is connected to the active queue, and fr
Pages that are immediately available to the operating system are connected to the ee queue. The inactive queue is used by the program currently running,
A candidate page selected for page replacement is connected.

When the operating system allocates a page for a program, it retrieves the page from the free queue. Inactive when there are few pages connected to the free queue after fetching
Takes a page out of the queue and replaces it if the reference bit is not on. Also,
As a result of page replacement, if the inactive queue becomes shorter than a certain constant value (set from a few% to a few tens% of the total number of pages in main memory), a plurality of active
A page is taken from the ve queue, the reference bit of the page is cleared, and the page is put into the inactive queue. In this way, the LRU algorithm is approximated by page-replacing only the pages that are not referenced while connected to the inactive queue.

FIG. 15 is a conceptual diagram showing a page number setting process when a page is moved from a plurality of active queues to a single inactive queue. In the figure, pages used in the program currently being executed are separately connected to the active queue for each priority. When the number of pages connected to the free queue becomes small, the page is taken out from the inactive queue and the reference bit of the page is checked to set whether or not the page is to be replaced. The setting is a normal FIFO with second cha
, which is the same as that of L.sub.nce, which allows the approximation of LRU. Further, when the inactive queue becomes shorter than a certain constant value, the page is taken out from a plurality of active queues, the reference bit of the page is cleared, and the page is put in the inactive queue. At this time, a plurality of a
The setting of the number of pages when moving a page to the ctive queue is obtained as shown in FIG. In this way, when the length of the active queue is different for each priority queue, the queue ratio of pages to be taken out from the active queue is changed according to the ratio determined by the priority. As a result, pages with high priority are more likely to be the target of page replacement, and F
Similar to the IFO with second change, the LRU can be approximated. The initial length of the inactive queue may be set based on the number of pages of the main memory as described above.

The active queue and inacti
A set of ve queues may be prepared for each priority. FIG. 17
FIG. 9 shows a conceptual diagram of processing when a set of active queue and inactive queue is prepared for each priority. In the figure, since there are multiple inactive queues,
The target of page replacement is selected from the inactive queue having a high priority. In this case, each inact
The length of the ive queue cannot be set based on the number of pages in the entire main memory because the number of pages with the priority is not constant. Therefore, based on the number of pages in that priority at present, it is set that some percentage thereof exists in the inactive queue. These ratios do not have to be constant in all priority queues, and may be set for each priority.

As described above, according to the present embodiment, when the information processing apparatus performs paging by the demand pre-paging method, the load of the entire information processing apparatus is detected and the pre-page is detected according to the load. By changing the number or stopping pre-paging, it is possible to prevent the processing capacity of the entire system from being lowered when the load is heavy or the number of free pages is small.
Further, regarding the loading of page data from the information processing apparatus connected by the network, the load of the network is also taken into consideration, so that the transfer capacity of the network can be similarly prevented from lowering.

Further, by reducing the time required for page-out processing when replacing a page and reading processing when the page is reused, when a program on the information processing apparatus is executed. It is possible to reduce the total time required for page replacement and page-in processing, which has the effect of increasing the processing speed of the information processing apparatus.

Further, by setting the priority of page-out at the time of page replacement according to the processing speed of the secondary storage device, there is an effect that the total time required for page replacement and page-in processing is reduced.

Further, by setting a low priority for page data recorded on the information processing apparatus connected by the network, the effect of reducing the load on the network and the information processing apparatus connected by the network can be obtained. It has the effect of reducing the load.

Considering the processing cost of page saving when replacing a page and the processing cost of reading the page again, pages with lower processing costs are preferentially replaced. By targeting, when a certain program is executed on the information processing device, it is possible to reduce the time required for the page replacement process and the page reloading process of the entire information processing device. Only a page with a high priority can be prevented from being selected as a target for page replacement, which has the effect of improving the processing efficiency of the information processing device.

Considering the processing cost of page saving when replacing a page and the processing of reading when reusing the page, pages with lower processing cost are preferentially replaced. By targeting, when a certain program is executed on the information processing apparatus, it is possible to reduce the time required for the page replacement processing and the page reloading processing of the entire information processing apparatus. In selecting a page to be replaced from among pages having the same number, the LRU approximation is performed, the time locality of the application program can be utilized, the page fault rate is reduced, and as a result, the processing efficiency of the entire information processing apparatus is reduced. Can be raised.

[0040]

As described above, according to the present invention, as the number of page replacements is calculated for each priority according to the number of page replacements required, the higher the priority, the easier the page replacement becomes. Pages that are rarely used even if they have a low priority are subject to page replacement, and page replacement that is advantageous in terms of processing cost and frequency of use can be performed.

[0041]

[Brief description of drawings]

FIG. 1 is a block diagram of an information processing apparatus according to an embodiment of the present invention.

FIG. 2 is a flowchart showing the operation of the first embodiment of the present invention.

FIG. 3 is a flowchart of processing for detecting the load on the entire information processing apparatus and setting the number of prepages.

FIG. 4 is a diagram showing a relationship between the number of active processes and the number of preppages.

FIG. 5 is a flowchart showing the operation of the second embodiment of the present invention.

FIG. 6 is a diagram showing an example of priority setting.

FIG. 7 is a flowchart showing a process of selecting a page replacement target.

FIG. 8 is a diagram showing priorities set according to the type of secondary storage device.

FIG. 9 is a flowchart showing a process of changing the number of pages to be replaced according to the priority.

FIG. 10 is a conceptual diagram showing a change in priority according to a processing cost.

FIG. 11 is a flowchart showing priority setting processing according to the third embodiment of the present invention.

FIG. 12 is a flowchart showing processing when selecting a target of page replacement according to priority according to the third embodiment of the present invention.

FIG. 13 is a diagram showing a priority setting example according to the third embodiment of the semi-invention.

FIG. 14: FIFO with second cha
3 queues (active, in)
The conceptual diagram which shows the relationship of (active, free).

FIG. 15: Single ina from multiple active queues
The conceptual diagram which shows the setting process of the page number at the time of moving a page to a ctive queue.

FIG. 16 is a conceptual diagram of setting the number of pages when moving a page to a plurality of active queues.

FIG. 17 is a conceptual diagram of processing when a set of an active queue and an inactive queue is prepared for each priority.

FIG. 18 is a diagram for explaining pre-paging.

FIG. 19 is a diagram for explaining pre-paging.

[Explanation of symbols]

1 CPU 2 main memory 3 Secondary storage 4 keyboard 5 Display device 6 network 201 ROM 202 RAM

─────────────────────────────────────────────────── ─── Continuation of front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 12/08-12/12

Claims (2)

(57) [Claims] 1. A memory management method for an information processing apparatus using virtual memory, wherein a setting step of setting a priority according to a processing cost at the time of page replacement for each page, and a replacement target for obtaining the number required for page replacement Number setting step, according to the number obtained in the replacement target number setting step, the priority target number setting step for determining the number of page replacement for each priority, and the priority target number setting step And a selection step of selecting a page to be a page replacement target based on the number of pages. 2. In a memory management device of an information processing device using virtual memory, a setting means for setting a priority for each page according to a processing cost at the time of page replacement, and a replacement target for obtaining a number required for page replacement. Number setting means, the target number setting means for each priority for obtaining the number of page replacement for each priority according to the number obtained by the number setting means for replacement, and the target number setting means for each priority And a selection unit that selects a page to be replaced based on the number of pages.