JPS63131247A - Garbage collection system - Google Patents

Abstract

PURPOSE:To control paging in GC and to improve the collection efficiency of garbage by considering whether a GC objective page exists on a main storage or not, and whether it includes much garbage or not so as to execute GC. CONSTITUTION:A reference relation between pages on a real storage 1 is held in an interreference table 3, and the reference relation from inside the page on a virtual storage 2 to inside the page on the real storage 1 is held in a real storage indirect pointer table 4 in an indirect pointer shape. The reference relation of interpage on the virtual storage 2 is held in a virtual storage indirect pointer table 5. With referring to respective tables, garbage collection is executed in respective pages on plural pages which are recently used and which are on the real storage 1 among GC waiting pages showed in a GC waiting table 6.

Description

[Detailed Description of the Invention] [Summary] The present invention is capable of performing garbage collection (GC) on recently used pages existing in real memory when a symbolic processing language is executed in virtual memory. , which attempts to suppress paging during GC, keep the processing stop time due to GC activation to a finite value, and increase garbage collection efficiency.

[Industrial application field]

The present invention is a LISP that is executed in a virtual memory environment.

This invention relates to a real-time garbage collection method for symbolic processing languages such as Prolog.

(Prior art) A large amount of garbage (
Garbage) is generated, and the useful parts are collected by removing the garbage page by page to create space and make it reusable. This processing is performed by a garbage collector, and the stop of symbol processing by the garbage collector is important for immediate response functions such as control (
This is a problem in the field of demand.

Many methods have been proposed in the past for the purpose of keeping the processing stoppage time due to garbage collector startup to a finite limit (from the viewpoint of real-time performance and real-time performance), but all of them are limited to real-time performance on real memory. However, it does not take into account the interruption of processing due to the effect of paging during garbage collection in a virtual memory environment.This causes the following problems.

[Problem that the invention seeks to solve]

In conventional real-time garbage collection, reference to data for GC processing is performed regardless of whether the data exists in the actual records 1 and 2.

Therefore, if much of the referenced data is not in real memory, page faults occur frequently. When a page fault occurs, the page is transferred from external storage such as a disk (main culprit)
, α, by ejecting unnecessary pages to create space, but this paging takes several milliseconds or more each time.
Since this is a task that requires a processing time of approximately several tens of milliseconds, if this occurs frequently, symbol processing will be interrupted for a long time. Therefore, the conventional GC method does not guarantee real-time performance in a virtual storage environment.

The present invention takes into consideration whether or not the page to be GC exists on the main memory and whether or not it contains a large amount of garbage.
By performing C, it is intended to suppress paging during GC and increase garbage collection efficiency.

[Means for solving problems]

The present invention uses a symbol processing language in real memory (1) and virtual memory (2).
), a cross-reference table (3) that maintains the reference relationship between pages in the real memory (1), and a cross-reference table (3) that maintains the reference relationship between pages in the real memory (1), and a A real memory indirect pointer table (4) that maintains reference relationships within pages on ill in the form of indirect pointers, and a virtual memory indirect pointer table (5) that maintains reference relationships between pages on virtual memory (2). ), a GC waiting page table (6) showing pages waiting for garbage collection in order of use, and a waiting page table (7) showing pages whose area has been collected in excess of a predetermined threshold value, and a GC waiting page table ( 6
) among the pages waiting for GC, which have been used recently and are on the real memory (1), and are characterized in that garbage collection is performed for each page by referring to each of the tables. .

[Effect]

If the objects of GC processing are limited to pages on the real memory, paging to and from the virtual memory can be suppressed, so that the symbol processing stop time due to activation of the garbage collector can be kept short. Also, in symbolic processing languages, much of the recently used data has a high probability of becoming garbage, so
If the GC target is limited to recently used pages, garbage collection efficiency will be increased. Also, if you specify the reference relationships between cells in various tables, it will result in garbage/
It is possible to easily determine whether or not the page exists, and it is possible to prevent valid data referenced from other pages from being erroneously treated as garbage, thereby making it possible to easily and accurately perform GC processing for each page.

〔Example〕

FIG. 1 is a conceptual diagram of the system of the present invention. First, the method of the present invention will be explained with reference to this.

■If a new cell (storage area of a fixed size, there are multiple cells in one page, data is stored in this cell and referenced on a cell-by-cell basis) is required as the program progresses, the actual memory The storage area is used from the page located in . The pages I used are 1, 2, 3 in order. Give a number such as...

This is called the generation number. When all cells of a page are used up, the page is added to the GC queue and the next new page is used. When a page in real memory is used up, a page in virtual memory is used next.

■Start the garbage collector when the processing progresses and the storage area (logical space) is used up to a certain extent. At that time, G
Among the pages in the C-wait collection, n pages are selected from those in real memory and have larger generation numbers, and classic batch-type GC is performed on each page. 1st
In the example shown in the figure, the GC order is generation numbers 15, 14, 12. ...
The order is... Those with larger generation numbers are those that have been used recently, and there is a high probability that they are in real memory.

(2) As a result of GC, a page (including valid cells and a predetermined number or more of empty cells) for which an area larger than a certain threshold value has been recovered is added to a waiting set for use. For pages that are not (for which a predetermined number or more of free cells could not be obtained), the generation number is decreased by k, and the pages are kept in the GC waiting group.

- When a new area is required for C0N5 (a basic LISP function), etc., the page that is in the collection waiting to be used, is on the real memory, and has the 1° largest generation number is used.

At that time, the generation number of the page is increased by 1. In the example shown, the page with generation number 16 is used, and the generation number is newly changed to 17.
shall be.

■If a page exists in the collection waiting for use, the processes from ■ to ■ are repeated. If there are no more pages during the usage waiting set (if all the pages are filled with a certain amount or more of active cells and added to the GC waiting set), the process (2) is performed for the pages in the GC waiting set. That is, when a new area is required, the GC waiting page with the largest generation number is used (using empty cells within the page), and the generation number of that page is increased by 1.

In order for page-by-page GC to be possible as described above, the existence of data pointed to by data on other pages within that page must be clear at the time the garbage collector is started. Therefore, as shown in FIG. 2, the following process is performed.

(When writing data to a page in an al program, if it is a reference pointer to data in another page, for example, if the data written to page B is a pointer to page A, then If it is a pointer to a page in memory, the page numbers A and B are stored as a pair in the mutual reference table 3 on the real memory 1.

If the data written to page B is a pointer to page D on virtual storage (disk device) 2, nothing is done. Further, since the cross-reference table 3 is simply written with pairs of pages, it only indicates the reference relationships between pages, and the reference relationships between cells are known by scanning the pages.

(If it is necessary to page out the page that used BL, for example, if it is necessary to page out page B, the page B is scanned. At this time, other pages on the real memory l, such as page A If a pointer to data in page B is found, register that pointer in the real memory indirect pointer table (RIPT) 4, and modify the original pointer (in page B) to point to this entry in RIr'T4. do.

If a pointer to another page on the virtual memory 2, for example page D, is found, it is similarly registered in the virtual memory indirect pointer table (VIPT) 5.

(C) When a page on virtual memory 2 is paged in, it is scanned, and if it has an entry in VIPT5, such as page D, it is removed (.Also, when page D is paged in, it is removed). If there is a reference and an entry in RIPT4, remove it and write A to cross reference table 3.
Register the page numbers of C as a set.

Through the above processing, the rules in classical processing systems such as stacks and interned symbols) (G
In addition to the starting point of the active cell when performing C), the RI
By using the pointer obtained by scanning the corresponding page via the PT4 and the cross-reference table 3 as the root, GC for each page becomes possible.

The GC waiting page table 6 is a table showing the GC waiting set in FIG. 1, and each GC waiting page is given a page name, a storage location (r indicates real memory, ■ indicates virtual memory), and a generation number. ing. You can tell whether it is in your real memory or not by looking at these r and v. The waiting page table 7 has a similar configuration.

〔Effect of the invention〕

The garbage collection method of the present invention described above has the following advantages.

(1) Basing during GC can be suppressed. 1 GC
Since the processing upon startup is limited to reading and writing to at most n pages on the real memory, it is guaranteed that the processing stop time due to GC is a finite time.

(2) Due to page management based on the generation number, pages that are subject to GC are n pages for which space has recently been taken. It has been confirmed that in symbolic processing languages, recently used data has a high probability of becoming garbage. Therefore, in this method, GC is performed by selecting an area that is likely to contain a large amount of dust, and it can be expected that the area recovery efficiency (the amount of area recovery relative to the number of times GC is activated) will be high.

[Brief explanation of the drawing]

FIG. 1 is a diagram explaining the principle of the present invention, FIG. 2 is a block diagram showing an embodiment of the present invention.

Claims (1)

[Claims] In a garbage collection method when a symbolic processing language is executed in a processing system having a real memory (1) and a virtual memory (2), a cross reference table (3) that maintains reference relationships between pages in the real memory (1) is used. ), a real memory indirect pointer table (4) that maintains reference relationships from pages on virtual memory (2) to pages on real memory (1) in the form of indirect pointers, and Virtual memory indirect pointer table (5) that maintains reference relationships between pages of
and G which shows the pages waiting for garbage collection in the order of usage.
A C waiting page table (6) and a waiting page table (7) indicating pages whose area has been collected in excess of a predetermined threshold are provided, and among the GC waiting pages shown in the GC waiting page table (6), Recently used and real memory (
1) A real-time garbage collection method that targets a plurality of pages above and performs garbage collection for each page by referring to each of the tables.