JPH0236458A - Loose-coupled multiprocessor system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to loosely coupled multiprocessor systems.

2. Description of the Related Art In a conventional loosely coupled multiprocessor system, a local memory and a secondary storage device are divided into predetermined page units to manage virtual memory, as shown in FIG. 41 is a logical address, 411 is a logical page number, and 412 is an offset from the top of the page. 42 is a page conversion table, 421 is a valid bit, 422 is a field containing information other than the above, and 423 is a field for storing a physical page number. 43 is a local memory.

In the conventional virtual memory management method for the loosely coupled multiprocessor system configured as described above, the page table 42 is retrieved using the logical page number 411. The valid bit 421 of the drawn entry is checked, and if it is valid, the local memory 43 within the subsystem is accessed. The valid bit 421 is checked, and if it is not valid, a page fault occurs, and if there is free space in the local memory, the page is moved there, and if there is no free space, the page in the local memory is released and the page is moved to the newly created free space. The required pages are read from the omitted secondary storage device.

Problems to be Solved by the Invention However, if each subsystem of a loosely coupled multiprocessor system manages virtual memory as described above, a page fault will occur if a physical address corresponding to a logical address does not exist in local memory. The necessary pages are transferred from the secondary storage device. If the logical addresses referenced in the executed program exhibit locality, the frequency of page faults will be low; otherwise, the frequency of page faults will be high. Since the local memory of each subsystem is managed independently from the local memory of other subsystems, page faults occur in subsystems that are running programs with high locality and in subsystems that are running programs with low locality. Different frequencies. When a page fault occurs, the secondary storage device is accessed, which increases input/output waiting time. In general, large-scale programs have low locality, and subsystems running large-scale programs have the problem of increased input/output wait times and reduced execution efficiency.

In view of this, the present invention reduces the occurrence of page faults in subsystems running large-scale programs or programs with low locality as described above, and reduces the number of page faults caused by waiting for input/output of data from a secondary storage device. The purpose of this invention is to provide a loosely coupled multiprocessor system that reduces execution efficiency degradation.

Means for Solving the Problems The present invention provides a loosely coupled multiprocessor system in which virtual memory is managed by dividing each local memory and secondary storage device into pages of a fixed size, in which each process is required for each entry. a page translation table having a field indicating whether the physical page is in local memory or in local memory in another subsystem coupled to the network;
and a page list A that is accessible from outside the subsystem, and a page list B that exists in a local memory outside the subsystem and that is accessible from the subsystem.
It has means for measuring the frequency of page faults within a certain period of time, and when the frequency of page faults in a certain subsystem 1 is lower than a certain lower limit and there is a page release request from another subsystem 2, the subsystem 1 automatically measures the frequency of page faults within a certain period of time. A page of local memory within the system is released, the page is registered in the page list A of the own system, the index of the page list A that stores the page is also registered in the page list B of the subsystem, and a certain subsystem 3 When the frequency of page faults in subsystem 4 is higher than a certain upper limit and the frequency of page faults in another subsystem 4 is lower than a certain lower limit, move the page indicated by page list A in subsystem 3 to the local memory of subsystem 4. This is a loosely coupled multiprocessor system having a storage management method characterized by the following.

According to the above-described configuration, when the frequency of page faults in a certain subsystem 1 is lower than a certain lower limit and there is a page release request from another subsystem 2, the subsystem 1 frees local memory within its own system. The page is released, the page is registered in the page list A of the own system, and the index of the page list A, in which the previous 5 own pages are distributed, is also registered in the page list B of the subsystem. In addition, if the frequency of page faults in a certain subsystem 3 is higher than a certain upper limit, and there is a page in the local memo J in the subsystem 3 that can be accessed from another subsystem 4, and there is a page in another subsystem 5, When the frequency of absent faults is lower than a certain lower limit, a page in the local memory of subsystem 5 is released to create a free area, a page that can be accessed from subsystem 4 in subsystem 3 is moved to the free area, and the subsystem Page list A within 5
The page is registered in , the information of the page list B in the subsystem 4 that can access the page is registered, and the entry of the page list A in the subsystem 3 that indicated the page is invalidated. These features allow subsystems that frequently experience page faults to have accessible pages in other subsystems in addition to their own local memory, so programs with less locality have more valid pages. This makes it possible to reduce the amount of time waiting for input/output from the secondary storage device.

Embodiment FIG. 1 shows a schematic configuration diagram of a loosely coupled multiprocessor system in an embodiment of the present invention. 1.2 are subsystems of a loosely coupled multiprocessor system,
3 is a network connecting each subsystem.

11.21 is a processor, 12.22 is a storage management unit, 13.23 is a local memory, and 14. ．． 24 is an input/output control device. Although only two subsystems are shown in the figure, there may be many subsystems.

FIG. 2 shows a schematic diagram of a data structure in an embodiment of the present invention. 21.22 are subsystems of a loosely coupled multiprocessor system, 211.221 are logical addresses, 2111°2211 are logical page numbers, 2
112.2212 is the offset value from the top of the page, 2
12,222 is a page table, 2121.2122.
2221.2222 is page table entry 21
211.21221 is a field indicating whether the physical page number indicated in the entry is in local memory within the respective subsystem or in another subsystem; A page list indicating physical pages of local memory that can be accessed, 2141 and 2241 are 214 and 2, respectively.
24 entries, 213.223 is a page list for indicating the page list of another subsystem from each subsystem, and 2131.2331 are entries of 213 and 223, respectively.

Although not shown in the figure, each subsystem has means for measuring the frequency of page faults within a certain period of time and storing the current state of the frequency.

Using the data structure shown in FIG. 2 as an example, a description will be given of how a logical address points to a physical memory using the above configuration.

The page table 212 is retrieved based on the logical page number 2111 of the logical address 211. Although not shown in the figure, a field indicating whether the physical page is valid is checked, and if the entry is invalid, a page fault occurs. Entry 2121 is valid and field 212
11 is on, the physical page number refers to the page in the local memory 15, although it is omitted in the figure, and the offset value 2
112 as well as the physical address. If the entry 2122 is valid and the field 21211 is off, the page list 213 is retrieved.

The entry 2131 retrieves the page list 224 of the subsystem 22. Although omitted in the diagram, entry 2
The physical page number in 241 and the physical address together with the offset value 2112 are determined.

Next, we will explain the behavior when a subsystem receives a page release request from another subsystem and the behavior when a page fault occurs in a subsystem, respectively.
This will be explained according to the flowcharts [a] and [bl] in the figure.

However, a list of pages that can be accessed from other subsystems is called page list A, and a table showing page list A of other subsystems is called page list B. Also,
If the same name is used frequently, add a symbol or number after the name to distinguish it.

[al: (31a) When a certain subsystem S1 receives a page release request from another subsystem s2, (32
a) Compare the frequency of page faults to some lower bound. (33a) If the frequency is lower than the lower limit and no subsystem has provided the page yet, the requesting subsystem S2 is requested to provide the page.

(34a) Subsystem S2 recognizes the page provision by subsystem S1 and notifies subsystem S1 of this. (35a) If the page to be released by subsystem S1 is a page that can be accessed by a process running in another subsystem S3, (36a) If the page has been modified, it is written to the secondary storage device, and subsystem S
Entries of page list A in 3 and subsystem S3
Disable page table entries for processes running within. (37a) If the page to be released is a page that can be accessed by a process running in subsystem s1, and if the page has been modified, it is written to the secondary storage device and the entry in the page table is invalidated. (38a) Subsystem S indicates that the page has been released.
Let 2 know.

(bl: (31b) When a certain subsystem s1 generates a page fault, (32b) Compare the frequency of page faults with a certain upper limit. (33b) If the frequency is higher than the upper limit, (34b) Issue a page release request to the subsystem connected to the network. (35b) If there is a subsystem s3 that provides the page, notify it to the subsystem s3 and wait until the page is actually released. ( 36b) If there is a page P1 in the local memory that can be accessed from another subsystem s2, (37b) copy the page P1 provided by the subsystem s3, and copy the page P1 provided by the subsystem s3, and register it so that page list B of subsystem s2 points to the entry of page list A of subsystem s3. (38
b) Invalidate the entry pointing to page p1 in page list A of subsystem s1. (39b)
Data necessary for the freed area is transferred from the secondary storage device. <310b) When there are no pages in the local memory of subsystem S1 that can be accessed from other subsystems, the data necessary for the freed page is transferred from the secondary storage device, and (311b) the page list of subsystem s3 is (312b) Register the page so that the entry in the page table of subsystem s1 points to the entry in page list B, (313
b) Register so that the page list B entry of the subsystem s1 points to the page list A of the subsystem s3. (314b) If there is no subsystem on the network that provides the page, subsystem s
Free 1 local memory. If the page to be released is a page that can be accessed by a process running in another subsystem s4 (315b), if the page has been modified, it is written to the secondary storage and is added to the page list A entry in the subsystem S4. Invalidates page table entries of processes running in subsystem S4. (316b) If the page to be released is a page that can be accessed by a process running in subsystem S1, if the page has been modified, write it to the secondary storage device;
Invalidate that page table entry. (31
7b) Transfer the necessary data to the freed area. ,(
318b) Point the entry in the page table of subsystem S1 to the previous page.

Effects of the Invention As explained above, according to the present invention, large-scale programs or programs with low locality that frequently generate page faults can secure as many accessible pages as possible,
Reduce the occurrence of page faults in running subsystems,
The decrease in program execution efficiency caused by waiting for input/output of data from the secondary storage device is also reduced, and the memory of the entire loosely coupled multiprocessor system can be used effectively. Note that this method can also be applied when using a multi-stage page table.

[Brief explanation of the drawing]

FIG. 1 is a schematic diagram of a loosely coupled multiprocessor system according to an embodiment of the present invention, FIG. 2 is a schematic diagram of a data structure according to an embodiment of the present invention, and FIGS. 3 [a] and [b] each show a subsystem. A flowchart showing the operation when the system receives a page release request from another subsystem, flowchart 1 and 4 showing the operation when a page fault occurs, show the conventional virtual memory management method. It is a conceptual diagram. 1.2...-Nub system, 3... Network, 11.21... Processor, 12.
22... Memory management department, 13.23...
Local memory, 14°24...Population control device. Name of agent: Patent attorney Shigetaka Kurino, and one other person: ``Gu-411 ・ 4;2 Notes on 5 logical address table scraps, Logical page number S unit, Facility, N page publication number sex! Show field knee'

Claims (1)

[Claims] In a loosely coupled multiprocessor system in which virtual memory is managed by dividing each local memory and secondary storage device into pages of a fixed size, each process can check if the physical pages required for each entry are available in the local memory. A page list A that exists in the local memory within the subsystem and is accessible from outside the subsystem, and has a page conversion table that includes a field indicating whether the page is in the local memory in another subsystem connected to the subsystem. , a page list B that exists in local memory outside the subsystem and is accessible from the subsystem, and means for measuring the frequency of page faults within a certain period of time; When the fault frequency is lower than a certain lower limit and there is a page release request from another subsystem 2, the subsystem 1 releases a page in local memory within its own system, registers the page in its own system's page list A, The index of the page list A that stores the page is registered in the page list B of the subsystem, and if the frequency of page absent faults in a certain subsystem 3 is higher than a certain upper limit and the frequency of page absent faults in another subsystem 4 is higher than a certain upper limit, A loosely coupled multiprocessor system having a memory management method characterized in that a page indicated by a page list A in a subsystem 3 is moved to a local memory of a subsystem 4 when the value is lower than a certain lower limit.