JPH08161230A - Look-ahead method in on-demand paging - Google Patents

Abstract

PURPOSE: To lower a page fault generation rate at the time of program execution for noncontinuously referring to pages and to accelerate a processing speed. CONSTITUTION: A program 1 during execution accesses a main memory unit 2, and when a page fault is generated, several pages including the page concerned are read. In this case, the page referred to next the page concerned in the past is stored in fault history information 8. Thus, when the page fault is generated, the pages are looked ahead in the order when the fault history information 8 is present and the pages are looked ahead in the order of storage in a secondary storage device in the other case.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an on-demand paging prefetching method for efficiently prefetching a page when a page fault occurs in an operating system using a so-called on-demand paging system.

[0002]

2. Description of the Related Art In an information processing apparatus, when an application program executes various processes, the program and data are stored in a main storage device. However, the main storage device has only an address space limited by the hardware configuration. On the other hand, if the application program is large, it becomes impossible to store all the application programs in the main storage device. Therefore, conventionally, a virtual storage system is used to set a free virtual address space that is not limited by the amount of hardware. In such a method, the program being executed accesses a wider virtual address space regardless of the physical address space of the main storage device, and if the required page does not exist, the corresponding page from the secondary storage device is accessed. To the main storage device. Such a method is called an on-demand paging method and is widely used in operating systems of information processing devices.

[0003]

By the way, when a page fault occurs, not only the corresponding page is transferred from the secondary storage device to the main storage device, but also several pages following the page are collectively sent to the main storage device. To prevent frequent page faults. This is called page prefetching, but in an actual program there are many repetitions and branches of the same processing, and it is assumed that processing proceeds in accordance with the page order (virtual address order) stored in the secondary storage device. Not necessarily. Further, the data area and the like are often accessed ignoring the storage order, and many non-continuous page references occur. Therefore, there has been a problem that the page read-ahead process in consideration of continuous page references as in the past does not always work effectively for high-speed processing of a program.

[0004]

The present invention adopts the following constitution in order to solve the above problems. According to this method, a main memory is accessed, a page fault is generated when a page not stored in the main memory is accessed, and a group of pages including the corresponding page are stored in the main memory from the main memory. When performing prefetch to transfer to the device,
If there is preset fault history information, a group of pages is prefetched according to the transfer page order represented in the fault history information and transferred from the secondary storage device to the main storage device. In other cases, The page in which a page fault has occurred and the pages sequentially stored in the secondary storage device that follow the page are prefetched and transferred to the main storage device.

It should be noted that, with respect to the generated page fault, the page read without reference to the page read in advance is transferred in addition to the corresponding page stored in the secondary storage device and other arbitrary pages have been read in advance. When a fault occurs, the relevant page and the page referenced immediately after that may be associated with each other as pages to be prefetched successively in order, and the fault history information may be generated.

Further, the program to be executed is analyzed in advance, a case where discontinuous page references different from the virtual address order are made is listed and tabulated, and a page to be read next is associated with each relevant page. Fault history information may be generated.

[0007]

[Function] The running program accesses the main memory,
When a page fault occurs, several pages including the page are read. In this case, the page referenced next to the page in the past is stored in the fault history information. So when a page fault occurs,
If the fault history information is present, the pages are prefetched in that order, and in other cases, the pages are prefetched in the order stored in the secondary storage device. As a result, the page fault occurrence rate of a program or the like that refers to pages discontinuously is reduced, and the processing speed is improved.

[0008]

The present invention will be described in detail below with reference to the embodiments shown in the drawings. FIG. 1 is a functional block diagram of an apparatus for implementing a prefetch method in on-demand paging according to the present invention. In the figure, the running program 1 executes a certain process while referring to the programs and data stored in the main storage device 2. A part of the programs and data stored in the secondary storage device 3 is mapped in the main storage device 2. An on-demand paging control unit 4 is provided to control predetermined paging when a page fault occurs due to access by the executing program 1. The on-demand paging control unit 4 includes a page fault management unit 5, a page fault history holding unit 6, and a prefetch management unit 7.
Is provided.

The page fault management unit 5 is a control unit for executing paging and prefetching, which will be described later, when the running program 1 causes a page fault. The page fault history holding unit 6 is newly provided in the present invention. When a page is referred to in accordance with the content of a page fault that occurred in the past, the page referred to next is detected, and the page is kept constant. It is a part to be stored in the format of. When a page fault occurs, the prefetch management unit 7 collectively transfers not only the page in which the page fault has occurred but also the page predicted to be referred to next from the secondary storage device 3 to the main storage device 2 in a batch. This is the part that controls.

FIG. 2 is an explanatory diagram of the virtual storage system and paging. As shown in this figure, when the program or data referenced by the executing program 1 is larger than the physical address space 2A set by the main memory 2, the executing program 1 accesses the virtual address space 3A assuming the virtual address space 3A. However, if there is no page to be referred to in the main storage device 2, a page fault is generated. That is, when a portion of the virtual address space 3A that is not mapped in the main memory 2 is accessed, a page fault occurs.
The programs and data of that portion are stored in the secondary storage device 3 shown in FIG. 1, and this is the main storage device 2.
Transferred to. In this case, in the on-demand paging method, the program and data are stored in the main storage device 2 page by page.
Further, in the case of the present invention, a predetermined number of pages are collectively transferred by adopting the prefetching method.

FIG. 3 shows a reference operation of fault history information. On the right side of this figure, the page numbers in the virtual address space with serial numbers 1 to 11 are displayed. The content of fault history information is shown on the left side. The number on the left side of this fault history information is the page number that was referenced immediately before the page with the number on the right side was referenced. That is, for example, in this example, two pages are referenced after page 1, and four pages are referenced after page 2. Such fault history information is generated by actually executing the program and detecting and recording the reference order. Alternatively, a program to be executed in advance is analyzed, and cases where discontinuous page references different from the virtual address order are listed are created manually.

For example, in this example, it is assumed that a page fault occurs when the fourth page of the virtual address space is referenced. In the normal case, for example, the fifth page and the sixth page are prefetched together with the fourth page and transferred to the main storage device. On the other hand, in the present invention, the fault history information is referred to here. According to the fault history information, the 10th page is referred to after the 4th page. Also, after the 10th page, the 8th page is referred to.
Therefore, in the present invention, prefetching is performed in order of the fourth page, the tenth page, and the eighth page. In this way, the pages are referred to in the optimum order in consideration of the past history and the like, and the occurrence rate of page faults can be suppressed sufficiently low.

FIG. 4 is an explanatory view of a method of updating such fault history information. For example, in the case of generating this fault history information based on the past program execution history, it is initially 4 in virtual address order as before,
Pages are read ahead such as pages 5 and 6 and transferred to the main storage device. However, when a page fault occurs again after that, the progress is recognized and the fault history information described above is obtained.

For example, as shown in this figure, when the fourth page is accessed and a page fault F1 occurs, the fourth page, the fifth page, and the sixth page are prefetched,
Transfer to main memory. At this time, the page table entry is not updated for the fifth page and the sixth page fetched after the fourth page. That is, the state is maintained while referring to the fourth page. Then, although it is preferable that the fifth page or the sixth page is referred to next, a page fault F2 occurs when the tenth page is referred to. Here, page 10 and page 1
Pages 1 and 12 are read ahead again in the normal procedure. At this time, the page table entry held until immediately before is taken out, and it is detected that the 10th page is referred to after the 4th page. The result is stored as fault history information in the page fault history holding unit 6 shown in FIG.

Similarly, when the page fault occurs again for the eighth page after the tenth page, the eleventh page and the twelfth page are transferred, it is detected that the eighth page is referenced after the tenth page. And similarly stored in the page fault history holding unit. Thus
The fault history information is updated one after another, and efficient page prefetch processing is executed.

On the other hand, it is also possible to analyze the program in advance and generate and prepare fault information. In this case, this is processed by the file management unit 9 shown in FIG. That is, the fault history information prepared in advance in the file management unit 9 is read into the page fault history holding unit 6 as an initial value immediately before the program execution. As a result, the efficient page prefetching process according to the present invention becomes possible immediately after the start of program execution. The fault history information is updated as described with reference to FIG. 4 according to the execution of the program, and the result is stored again in the file management unit 9 after the program ends. As a result, the fault history information stored in the file management unit 9 is optimized each time the program is repeatedly executed, and more efficient processing becomes possible. Note that such fault history information can be created by a compiler when compiling a program.

The present invention is not limited to the above embodiments. INDUSTRIAL APPLICABILITY The present invention can be applied to information processing apparatuses having various configurations that employ paging using a virtual memory system, and the configuration and functional blocks of the on-demand paging control unit can be freely selected according to the configuration of the information processing apparatus. It doesn't matter. In addition, the format of the fault history information, for example, is such that the page referred to immediately before is associated with the page referred to next, but when prefetching and transferring several pages in a batch, several pages are linked. An embodiment in which the prefetch processing is efficiently performed by using the content information is also possible.

[0018]

As described above, the prefetch method for on-demand paging according to the present invention stores the reference history as fault history information when the program refers to pages discontinuously, not in virtual address order. ,
If this exists, the group of pages is prefetched according to the transfer page order represented in the fault history information.
Since it is configured to transfer from the next storage device to the main storage device,
The page fault occurrence rate is reduced as compared with the case where pages are prefetched unconditionally in order of virtual addresses. In addition, the program execution speed can be increased and its efficiency can be improved.

[Brief description of drawings]

1 is a block diagram of an apparatus for performing the method of the present invention.

FIG. 2 is an explanatory diagram of a virtual storage system and paging.

FIG. 3 is an explanatory diagram of a reference operation of fault history information.

FIG. 4 is an explanatory diagram of a method of updating fault history information.

[Explanation of symbols]

 1 Program in Execution 2 Main Storage Device 3 Secondary Storage Device 4 On-Demand Paging Control Unit 5 Page Fault Management Unit 6 Page Fault History Holding Unit 7 Prefetch Management Unit 8 Fault History Information 9 File Management Unit

Claims (3)

[Claims] 1. A main memory is accessed, a page fault is generated when a page not stored in the main memory is accessed, and a group of pages including the corresponding page is transferred from the secondary memory to the main memory. In the case of performing pre-reading for transfer to the storage device, if preset fault history information exists, the group of pages is pre-read in accordance with the transfer page order represented in the fault history information, and the pre-reading is performed from the secondary storage device. Transfer to the main storage device, and in other cases, prefetch the page in which the page fault has occurred and the pages stored in the secondary storage device consecutively to that page, and transfer to the main storage device. A prefetch method for on-demand paging. 2. For a generated page fault, 2
When a page fault occurs without referencing the prefetched page even though the other page is prefetched and transferred together with the relevant page stored in the next storage device, the relevant page and immediately after that. The prefetch method for on-demand paging according to claim 1, wherein the page referred to in (1) is associated with each other as a page to be prefetched successively in order to generate fault history information. 3. The program to be executed is analyzed in advance, a case where discontinuous page references different from the virtual address order are made is listed and tabulated, and a page to be read next is associated with each corresponding page. The prefetching method for on-demand paging according to claim 1, wherein fault history information is generated.