JP2001092627A - Method for compressing data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for compressing data capable of improving a processing time at the time of compressing the memory image of a personal computer or the like, and storing the data in a storage device such as an HDD. SOLUTION: At the time of compressing a data stream stored in a main storage part, and storing a code stream obtained as the result of the compression in an auxiliary storage part, when the word length of an arithmetic processing part is 4 bytes, and the processing unit length of the compression processing is 1 byte, a shortest offset code is assigned to offset separated only by (the word length of the arithmetic processing part)÷(the processing unit length of the compression processing)=4.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention uses compression based on a dictionary-based system represented by LZ77, and particularly compresses the entire contents of a main memory (hereinafter referred to as a memory image) such as a personal computer (personal computer). And a data compression method for storing the data in a storage device such as an HDD (hard disk drive).

[0002]

2. Description of the Related Art At present, a method of compressing data using a dictionary base is described in Abraham Lempel and Jacob Ziv's paper "A Universal Algorithm for Sequential Data Compr.
Seen in ession '. This is commonly referred to as the Lempe1-Ziv encoding slide dictionary method or LZ77 method. For example, Seiji Munakata: Ziv-Lempel's data compression method, information processing,
Vol.26, No1 (1985) is known.

[0003] The LZ77 algorithm is a method of dividing encoded data from an arbitrary position in a past data sequence into a sequence of the maximum length that matches, and encoding the duplicate as a copy of the past sequence. More specifically, as shown in FIG. 4, a moving window for storing encoded input data and a look-ahead buffer for storing data to be encoded from now on are provided, and the data sequence of the look-ahead buffer and the data sequence of the moving window are provided. Is compared with all of the sub-sequences, and the maximum-length sub-sequence that matches in the moving window is obtained. Then, in order to specify this maximum-length sub-sequence in the moving window, "start position of the maximum-length sub-sequence"
And the "matching length" and the "next symbol that resulted in a mismatch". Next, the encoded data sequence in the prefetch buffer is moved to the moving window, and a new data sequence for the encoded data sequence is input into the prefetch buffer. Hereinafter, by repeating the same processing, the data is decomposed into sub-series and the encoding is executed.

[0004] Many improved types of such encoding have been proposed. For example, a method of providing a flag for identifying whether the data is an encoded code or raw data, and encoding the raw data when the encoded code becomes longer than the raw data is adopted. This is based on the LZSS coding method (TC Bell, “Better OPM / L Text Compressi
on ", IEEE Transaction Commun., Vol. COM-34, N
o.12, Dec (1986)). In addition, M.
Nelson-Data Compression Handbook, Second Revised Edition, Toppan (1996). ISBN 4-8101-8605-9.

[0005]

In recent years, OS (Operating System) and applications mounted on personal computers have become increasingly sophisticated year by year, and at the same time, large-capacity memories are required. Such a large capacity and large scale of the OS and the application increase the waiting time at the time of power ON / OFF. A user's need is to be able to instantaneously resume processing in the previous processing state after the processing is interrupted and the power is turned off. In order to realize this, the contents (memory image) of the entire main memory at the time of interruption are stored in an HDD or the like, and the power is turned off. At the time of resumption, the memory image stored in the HDD is loaded into the main memory, and the same state as before the interruption There is a method (hibernation processing) of shortening the waiting time at the time of resuming by reproducing.

However, in recent personal computers, 64M
B, it is not uncommon to use a large capacity memory of 128 MB. If such a large-capacity memory image is stored in the HDD as it is, the continuous transfer performance of the HDD becomes 1
Since it is about 6 MB / s or 32 MB / s, it takes several seconds. Therefore, it is conceivable to shorten the storage time by compressing and storing the memory image. The LZ77-based compression method, which has a feature that the compression process is relatively light and does not depend so much on the characteristics of the input data, is suitable for such an application, but is simply a standard LZ7.
Applying the seven methods has many disadvantages.

For example, in a memory image which is input data, there is a high possibility that repetitive data having a word length of the CPU as a unit frequently occurs. However, the existing LZ77-based data compression system makes use of such features of the memory image. There is no such coding system. In addition, there is a drawback that comparison is performed up to a distant offset that is unlikely to match.

SUMMARY OF THE INVENTION An object of the present invention has been made in view of the above circumstances, and it is an object of the present invention to improve processing time when a memory image of a personal computer or the like is compressed and stored in a storage device such as an HDD. It is to provide a possible data compression method.

[0009]

In order to solve the above problems and achieve the object, a data compression method according to the present invention is configured as follows.

(1) The data compression method according to the present invention is used when an input data stream is compressed to output a code stream, and the contents of the entire main memory (hereinafter referred to as a memory image) of a personal computer or the like are compressed to an HDD or the like. , The shortest offset code is assigned to an offset separated by (word length of CPU / processing unit length of compression processing (= symbol length)).

(2) The data compression method of the present invention is for compressing an input data stream to output a code stream, and compressing the entire contents of a main memory (hereinafter referred to as a memory image) of a personal computer or the like to an HDD or the like. When the matching offset is a multiple of (the word length of the CPU / the processing unit length of the compression process) and a certain matching length code is generated, the first word boundary within the matching length from the matching start position is generated. Into three codes: a match code up to, a match code from the first word boundary in the match length to the last word boundary in the match length, and a match code from the last word boundary in the match length to the match end position. And encode.

(3) The data compression method according to the present invention is used when an input data stream is compressed to output a code stream, and the contents of the entire main memory (hereinafter referred to as a memory image) of a personal computer or the like are compressed to an HDD or the like. When storing in the storage device, the matching length code is assigned only to the matching length that is a multiple of (the word length of the CPU / the processing unit length of the compression process) for the portion assigned to the long matching length in the matching length code.

(4) The data compression method of the present invention is used when an input data stream is compressed to output a code stream, and the contents of the entire main memory (hereinafter referred to as a memory image) of a personal computer or the like are compressed to an HDD or the like. , The portion assigned to the long match length code is assigned a match length code only for the match length of a power of two.

(5) The data compression method of the present invention is used when an input data stream is compressed and a code stream is output, and the content of the entire main memory (hereinafter referred to as a memory image) of a personal computer or the like is compressed to an HDD or the like. The number of reference points at the time of compression can be variably set by an external instruction, and the decompression side can decompress the maximum number of reference points that can be set on the compression side. Time and HDD access performance information are stored in a file, and the number of reference points is determined according to the information.

(6) The data compression method of the present invention
In (5), if a storage bottleneck in the HDD occurred during the previous storage, the compression rate is improved by increasing the number of reference points the next time, while the compression processing of the CPU is not performed because the HDD is empty during the previous storage. If so, reduce the reference points the next time.

(7) The data compression method of the present invention is used when an input data stream is compressed and a code stream is output, and the content of the entire main memory (hereinafter referred to as a memory image) of a personal computer or the like is compressed to an HDD or the like. When stored in the storage device, the number of reference points at the time of compression can be variably set by an instruction from the outside, and the decompression side can decompress the maximum number of reference points that can be set on the compression side. The rate information is stored in a file, and the number of reference points is determined according to the information.

(8) The data compression method of the present invention
In (7), if the previous compression ratio is equal to or smaller than the threshold, the number of reference points is increased in the next round. If the previous compression ratio is equal to or larger than the threshold, the number of reference points is decreased in the next round.

(9) The data compression method according to the present invention is used when an input data stream is compressed to output a code stream, and the contents of the entire main memory (hereinafter referred to as a memory image) of a personal computer or the like are compressed to an HDD or the like. When stored in the storage device, the number of reference points at the time of compression can be variably set by an instruction from the outside, and the decompression side can decompress the maximum number of reference points that can be set on the compression side. Retains the compression ratio information in a file. When the user job is not performed for a certain period of time, estimates the compression ratio in the background, updates the previous compression ratio information, and refers to it according to the previous information or information updated in the background Determine the score.

(10) The data compression method according to the present invention is for compressing an input data stream to output a code stream, and compressing the entire contents of a main memory (hereinafter referred to as a memory image) of a personal computer or the like to an HDD or the like. When the user job is not performed for a certain period of time when the image is stored in the storage device of
The data is stored in the DD, and when the hibernation process is performed as it is, the compression of the memory image is omitted.

According to the present invention, the following effects can be obtained.

(1) LZS (for example, M. Nelson: Data Compression Handbook, 2nd revised edition, Toppan (1996))
When the existing LZ77 data compression method is applied to hibernation of a personal computer as described in
It is highly probable that repetitive data with the word length as a unit frequently appears, but the existing LZ77-based data compression method does not have a coding system that takes advantage of the characteristics of such a memory image, or there is a possibility that it matches. There were drawbacks, such as comparing up to distant offsets as low as 2047. In the present invention, the data area of the memory image,
By focusing on the nature of the stack area and assigning the shortest offset code to offsets separated by (word length / symbol length), the compression ratio can be improved and the compression processing time can be reduced. For example, for a 32-bit CPU, the word length is 4 bytes, and LZ7
If the unit of processing (symbol length) of 7 is 1 byte,
(Word length / symbol length) = 4. In this case, the shortest offset code is assigned to offset 4.

(2) As described in the above (1), the memory image is characterized in that repeated data in units of the word length of the CPU frequently occurs. Therefore,
When the matched offset is a multiple of 4, for a given match length, the match code from the match start position to the first word boundary in the match length, the first word boundary in the match length to the last in the match length The code is decomposed into three codes: a match code up to the word boundary, and a match code from the last word boundary in the match length to the match end position.
By doing so, at the time of decompression, it is possible to copy in units of words for the "match code from the first word boundary in the match length to the last word boundary in the match length", so that the speed can be increased.

(3) In a memory image, there is a high possibility that repeated data frequently appears in units of the word length of the CPU.
Therefore, for the part assigned to the long match length in the match length code, codes are assigned only to match lengths that are multiples of 4, thereby simplifying the table of match length codes without significantly deteriorating the compression ratio. Processing can be speeded up. For example, codes are assigned to all match lengths up to match lengths 1 to 32, but after that, 36, 4
Codes are assigned only where the matching length is a multiple of 4, such as 0, 44,.

(4) In a memory image, there is a high possibility that repetitive data having a word length of the CPU as a unit frequently appears.
Therefore, by assigning a code only to a match length of a power of 2 for a portion assigned to a long match length in the match length code, the table of match length codes can be simplified without significantly deteriorating the compression ratio. Can speed up the process. For example, a code is assigned to all the match lengths up to the match lengths 1 to 32, but after that, 64,
Codes are assigned only where the matching length is a power of 2, such as 128, 256, 512, 1024.

(5) In general, when the number of reference points (the number of offsets to be searched) at the time of compression is increased, the compression ratio is improved, so that the size of data to be stored is reduced. The load increases. Since the difference between the time for storing the pressure line data in the HDD and the CPU processing time is the apparent processing time, it is important to balance the two. Therefore, the number of reference points at the time of compression can be set variably, and can be increased or decreased by an external instruction. The extension side is designed to be able to extend the maximum number of reference points. Then, the processing performance of the CPU, H
The optimum number of reference points is predicted based on the DD access performance. Information on the previous processing time and HDD access performance is stored in a file and is used as reference information. If the storage has become a bottleneck in the HDD at the time of the previous storage, the number of reference points is increased next time to improve the compression ratio. On the other hand, if the HDD is empty at the time of the previous storage and the compression processing of the CPU is a processing bottleneck, the number of reference points is reduced next time. By doing so, the storage time of the HDD and the compression processing time of the CPU can be balanced, and the overall processing time can be reduced.

(6) For the first time, the number of reference points is determined based on the compression ratio of prediction. For the second and subsequent times, the previous compression ratio information is stored in a file, and this value is compared with a preset compression ratio threshold to determine the number of reference points. If the previous compression ratio is less than or equal to the threshold, the next round increases the number of reference points. On the other hand, if the previous compression ratio is equal to or greater than the threshold, the number of reference points is reduced in the next round. By doing so, the storage time of the HDD and the compression processing time of the CPU can be balanced, and the overall processing time can be reduced.

(7) In the methods described in (5) and (6) above, the number of reference points is determined based on the previous information. Therefore, when a time interval has elapsed from the previous hibernation process, the compression ratio There is a drawback that estimation cannot be performed accurately. Therefore, when the user job is not performed for a certain period of time, the compression ratio is estimated in the background, and the previous compression ratio information is updated. In this way, the number of reference points can be determined based on information close to the actual compression ratio.

(8) When no user job is performed for a certain period of time, the memory image is compressed in the background and stored in the HDD. When the hibernation process is performed as it is, the compression of the memory image is omitted. By doing so, the time for the hibernation process can be further reduced.

[0029]

Embodiments of the present invention will be described below with reference to the drawings.

As shown in FIG. 1, the personal computer has a CPU
1. Main memory (memory) composed of DRAM or the like 2, C
Host bridge 3 for connecting PU, path and main memory, HD
A D (IDE) controller 4, an HDD 5, and a PCI path 6 as a main path are provided.

FIG. 1 is a diagram showing a state in which the hibernation process is performed without compressing the contents of the main memory. Here, a case where the main memory is 64 MB is shown. At the time of suspension, the contents of 64 MB are stored in the HDD as they are. At the time of resumption, the contents of 64 MB stored in the HDD are directly loaded into the main memory.

As shown in FIG. 2, the personal computer has a CPU 1
1, a main memory 12 composed of a DRAM or the like, a host bridge 13 for connecting a CPU to a path and a main memory, an HDD controller 14, an HDD 15, and a PC serving as a main path.
An I path 16 is provided.

FIG. 2 is a diagram showing a state in which the contents of the main memory are compressed and the hibernation process is performed. Here, instead of using compression / decompression hardware, CPU
Is performing compression processing. 64MB main memory is 20MB
Shows the case where compression was successful. At the time of interruption, the contents of 64 MB are compressed to 20 MB by software processing by the CPU, and
The code data of the MB is stored in the HDD. When resuming, H
The code data stored in the DD is read, decompressed to the original 64 MB by software processing by the CPU, and loaded into the main memory.

FIG. 3 is a diagram showing the merits of compressing and storing the contents of the main memory. When performing the hibernation process without compression, a storage time of 64 MB is required. When the data is compressed and stored, the time required for the compression process is newly generated. However, the storage time is shortened, so that the processing time may be shorter than the case without compression as shown in FIG. Whether it is actually shortened depends on the compression ratio, compression processing time, HDD transfer time, and the like.

FIG. 4 is a diagram showing the principle of a general compression system based on LZ77. In LZ77-based compression, in order to determine the maximum length of the partial data sequence that matches in the moving window during encoding, the data sequence must be compared between the data sequence to be encoded and all positions in the moving window. Must be done. That is, as shown in FIG. 4, a data sequence to be coded (◇ =====... In the figure) is converted into a data sequence starting from an offset 1 position and a data sequence starting from an offset 2 position in a moving window. ,... Compared with the data sequence starting from the position of offset n (n is the size of the moving window),
The goal is to find the offset that gives the maximum match length. As the size of the moving window, for example, in the LZS method, an offset of 1 to 2047 is searched.

The LZ77-based compression method, which has a feature that the compression processing is relatively light and does not depend so much on the properties of the input data, is suitable for the compression for hibernation, but if the standard LZ77 method is simply applied, There are many disadvantages. For example, in a memory image that is input data,
Although there is a high possibility that repetitive data with the word length of the CPU as a unit frequently occurs, the existing LZ77-based data compression method does not have a coding system that makes use of such features of the memory image. In addition, the comparison is performed up to a distant offset that is unlikely to match.

That is, when the conventional LZ77 system is applied to compression by software processing of a CPU, the following is achieved. Here, an example of the LZS method is shown.

Conventional method (a module for simply finding the longest match for 2047 comparison points) {a match length between a data string from an encoding position and a data string from offset 1 is obtained, and the result is len1. Find the match length between the data string from the position and the data string from offset 2 and set the result to len2. Find the match length between the data string from the coded center and the data string from offset 3 and set the result to len3. The match length between the data sequence from the encoding position and the data sequence from the offset 31 is obtained, and the result is len2046. The match length between the data sequence from the encoding position and the data sequence from the offset 32 is obtained, and the result is len2047. Return the maximum value of len1, len2,... len2047 and the offset that gives the maximum match length. Is a diagram showing an example of the personal computer of the memory image that was equipped with a Tsu door CPU.

In a memory image, there is a high possibility that repeated data with the word length of the CPU as a unit frequently appears in a data area, a stack area, or the like. According to the present invention, the compression rate is improved by focusing on the properties of the data area and the stack area in the memory image, and assigning the shortest offset code to an offset separated by (word length / symbol length), thereby improving the compression rate. The compression processing time can be reduced. For example, if a 32-bit CPU has a word length of 4 bytes, and if the unit of processing (symbol length) of LZ77 is 1 byte, then (word length / symbol length) =
It becomes 4. In this case, the shortest offset code is assigned to offset 4.

When obtaining the match length at each offset, the upper limit is determined by the structure of the match length code, so even if a long match is obtained, it will be cut off there. For example, if the maximum length of the match length code is 256, when the number of matches reaches 256, the subsequent comparison is terminated, and
Let 256 be the match length.

The method for obtaining the maximum matching length as described above has a disadvantage that the processing speed is reduced when a long match is obtained for each offset. For example, if a part where identical data is continuous is encoded, the longest match (for example, 256) can be obtained by comparison with all offsets. Therefore, if one data comparison is counted as one time, 256 times for each offset are counted. 8192 comparisons will be made. However, since 256 is the maximum matching length, when the matching length of 256 is obtained, the comparison of the remaining offsets is cut off, thereby realizing high speed.

Normally, the Huffman code is used for both the offset code and the match length code. Therefore, a short offset code is assigned to an offset at which a long match length is likely to be obtained. For example, the offset code of the LZS scheme is configured as shown in FIG. FIG. 7 shows an example of another offset code.

In the present invention, the offset that is (the word length of the CPU / the symbol length) is regarded as the most effective, and the shortest code is assigned to the offset 4.
Short codes are assigned in the order of 5, 6, ..., 32. FIG. 8 shows a configuration example of the offset code according to the present invention.

When the method of the present invention is applied to compression by CPU software processing, the following is obtained.

A module for obtaining the longest match according to the present invention. The match length between the data sequence from the encoding position and the data sequence from offset 4 is obtained, and the result is len4.
If = 256, the offset is set to 4, and the matching length is set to 256. The module ends. The matching length between the data string from the encoding position and the data string from offset 1 is obtained, and the result is len1.
If 56, the offset = 1 and the coincidence length = 256, and the module ends. The coincidence length between the data sequence from the encoding position and the data sequence from the offset 2 is obtained, and the result is 1en2, len2 = 2
If 56, the module ends with offset = 2 and matching length = 256. The matching length between the data string from the encoding position and the data string from offset 32 is obtained, and the result is len3 with the result being len32.
If 2 = 256, the offset is set to 32 and the matching length is set to 256. The module ends. The maximum value of len4, len1, len2,..., Len32 and the offset giving the maximum matching length are returned. Next, referring to FIG. A description will be given of the generation of a code that can perform the decompression process.

When the 11 bytes a to k shown in FIG. 9 coincide with the data starting from $, a code of (offset 11, coincidence length 11) is normally generated.

At the time of expansion, the offset (in this case, 1
11 bytes from the previous position (1 byte) are copied to obtain decompressed data. When one word of the CPU is 4 bytes, the portions c to j shown in FIG. 9 coincide with word boundaries, so that word-by-word copying is possible and processing can be performed faster than byte-by-byte copying.

Therefore, a division such as (offset 11, matching length 2)... A, b (offset 11, matching length 8)... C to j (offset 11, matching length 1). Then, the portions c to j can be subjected to high-speed decompression processing in word units.

Next, the match length code according to the present invention will be described.

FIG. 10 is a diagram showing an example of the configuration of a normal match length code. As shown in FIG. 10, a short Huffman code is assigned to a short match length. In this example, a Huffman code consisting of 91 words is assigned to match lengths 1 to 91 in order. In this case, efficient compression cannot be performed on long matching data. For example, in an unused area of a memory or the like, there may be the same data exceeding 1 kbyte, and a long match occurs there. However, if the upper limit of the match length is small as shown in FIG. 10, the match length code is repeatedly generated, which is not efficient.

According to the present invention, as shown in FIG. 11, the matching length code is assigned only to the matching length that is a multiple of 4 for the portion assigned to the long matching length in the matching length code. Can also generate codes efficiently. Here, since (the word length of the CPU / the processing unit length of the compression process) is 4, a multiple of 4 is assigned with priority.

In another example, as shown in FIG. 12, the part assigned to the long match length in the match length code is 2 bits.
Since a match length code is assigned only to a match length of a power of, a code can be generated efficiently even for a long match.

Next, control for varying the number of offsets used for comparison during compression will be described. In general, if the number of reference points (the number of offsets to be searched) at the time of compression is increased, the compression ratio is improved, so that the size of data to be stored is small.
However, on the other hand, the load of the compression processing performed by the CPU increases. The slower of the time for storing the compressed data in the HDD and the processing time of the CPU is the apparent processing time, so it is important to balance the two. Therefore, in the next embodiment, the number of reference points at the time of compression can be set variably,
It can be increased or decreased by an external instruction. The extension side is designed to be able to extend the maximum number of reference points.

At the first time, the number of reference points is determined based on the compression ratio of the prediction. For the second and subsequent times, the previous compression ratio information is stored in a file, and this value is compared with a preset compression ratio threshold to determine the number of reference points. If the previous compression ratio is less than or equal to the threshold, the next round increases the number of reference points. On the other hand, if the previous compression ratio is equal to or greater than the threshold, the number of reference points is reduced in the next round.

FIG. 13 is a flowchart showing the offset number variable control for performing comparison at the time of compression. First, the HDD storage time and CPU compression processing time saved during the previous hibernation are read (ST1). HDD
The storage time is compared with a threshold value (ST2). If the storage time exceeds the threshold value (ST2, YES), the number N of reference points is increased by one (ST2).
7). Further, the CPU compression processing time is compared with a threshold value (ST
3) If the threshold value is exceeded (ST3, YES), the number of reference points is reduced by one (ST6). After that, the hibernation process (ST4), the HDD transfer time and the CPU compression process time are updated (ST5).

In the soft compression processing, the number of offset points to be compared is controlled by the value of the reference point number N. By doing so, the storage time of the HDD and the compression processing time of the CPU can be balanced, and the overall processing time can be reduced.

The method shown in the flowchart of FIG.
Since the number of reference points is determined based on the previous information, there is a disadvantage that the compression ratio cannot be accurately estimated when a time interval has elapsed from the previous hibernation process. Therefore, when a user job is not performed for a certain period of time,
The compression ratio is estimated in the background, and the previous compression ratio information is updated.

FIG. 14 is a flowchart showing a process for estimating the compression ratio in the background and updating the previous compression information. First, the elapsed time from the previous hibernation process is measured by the timer function (ST
11). After a certain time has passed (ST11, YES),
In the background, only compression ratio estimation without HDD transfer is performed (ST12), and compression ratio information is updated (ST13). In this way, the number of reference points can be determined based on information close to the actual compression ratio.

As another example, when a user job is not performed for a fixed time, a memory image is compressed in the background and stored in the HDD. When the hibernation process is performed as it is, the compression of the memory image is omitted.

FIG. 15 shows a process of compressing a memory image in the background when no user job is performed for a certain period of time and storing it in the HDD, and omitting the compression of the memory image when the hibernation process is performed as it is. It is a flowchart.

First, the timer function measures the elapsed time since the user stops operating the keyboard and mouse (ST21). After a certain time has passed (ST2
1, YES), the memory image is compressed in the background and stored in the HDD (ST22). When the hibernation process is performed as it is (ST23, request immediately after), the compression of the memory image can be omitted (ST25). When the hibernation process is performed after the operation (ST23, a request after the operation), the process proceeds to the normal hibernation process (ST24). By doing so, the time for the hibernation process can be further reduced.

[0062]

According to the present invention, it is possible to provide a data compression method capable of improving the processing time when a memory image such as a personal computer is compressed and stored in a storage device such as an HDD.

[Brief description of the drawings]

FIG. 1 is a diagram showing a schematic configuration of a personal computer and a state in which a hibernation process is performed without compressing the contents of a main memory.

FIG. 2 is a diagram showing a schematic configuration of a personal computer and a state in which the contents of a main memory are compressed to perform a hibernation process.

FIG. 3 is a diagram for explaining the merits of compressing and storing the contents of a main memory.

FIG. 4 is a diagram for explaining the principle of a general compression system based on LZ77.

FIG. 5 is a diagram showing an example of a memory image of a personal computer equipped with a 32-bit CPU of i386 or later.

FIG. 6 is a diagram illustrating an example 1 of an offset code (LZS method).

FIG. 7 is a diagram illustrating an example 2 of an offset code.

FIG. 8 is a diagram showing an example of an offset code according to the present invention.

FIG. 9 is a diagram for explaining generation of a code that can perform a decompression process at high speed.

FIG. 10 is a diagram illustrating an example of the configuration of a match length code.

FIG. 11 is a diagram showing an example of a configuration of a match length code according to the present invention (the latter half is assigned only a multiple of 4).

FIG. 12 is a diagram showing an example of a configuration of a match length code according to the present invention (the latter half is assigned only to the power of 2).

FIG. 13 is a flowchart illustrating offset number variable control for performing comparison at the time of compression.

FIG. 14 is a flowchart showing a process of estimating a compression ratio in the background and updating previous compression information.

FIG. 15 is a flowchart showing a process of compressing a memory image in the background when no user job is performed for a certain period of time and storing it in the HDD, and omitting the compression of the memory image when the hibernation process is performed as it is; .

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... CPU 12 ... Main memory 13 ... Host bridge 14 ... HDD controller 15 ... HDD 16 ... PCI path

Claims (10)

[Claims] When a data stream stored in a main storage unit is compressed and a code stream obtained as a result of the compression is stored in an auxiliary storage unit, (word length of an arithmetic processing unit) ÷
A data compression method, wherein a shortest offset code is assigned to an offset separated by (processing unit length of compression processing). 2. When the data stream stored in the main storage unit is compressed and the code stream obtained as a result of the compression is stored in the auxiliary storage unit, the matched offset is (word length of the arithmetic processing unit) ÷ (compression). When a predetermined match length code is generated in multiples of (processing unit length of processing), the match code from the match start position to the first word boundary within the match length, the first word boundary within the match length to the last within the match length And the matching code from the last word boundary in the matching length to the matching end position.
A data compression method characterized in that the data is decomposed into two codes and encoded. 3. A part for compressing a data stream stored in a main storage unit and assigning a code stream obtained as a result of the compression to a long match length exceeding a predetermined length when storing the code stream in the auxiliary storage unit. , A matching length code is assigned only for a matching length that is a multiple of (word length of the arithmetic processing unit) / (processing unit length of the compression process). 4. A method of compressing a data stream stored in a main storage unit and storing a code stream obtained as a result of the compression in an auxiliary storage unit, wherein a portion assigned to a long match length exceeding a predetermined length among the match length codes. (C) assigning a match length code only for match lengths of powers of two. 5. When the data stream stored in the main storage unit is compressed and the code stream obtained as a result of the compression is stored in the auxiliary storage unit, the number of reference points at the time of compression can be variably set by an external instruction. The decompression side can decompress the maximum number of reference points that can be set on the compression side. The decompression side retains information on the processing time at the time of the previous compression and the access performance of the auxiliary storage unit in a file. Determining a data compression method. 6. When the storage processing of the auxiliary storage unit is a bottleneck at the time of the previous storage, the reference point is increased at the next storage to improve the compression ratio. 6. The data compression method according to claim 5, wherein the reference points are reduced at the time of next storage when the compression processing is a bottleneck. 7. When the data stream stored in the main storage unit is compressed and the code stream obtained as a result of the compression is stored in the auxiliary storage unit, the number of reference points at the time of compression can be variably set by an external instruction. The decompression side can decompress the maximum number of reference points that can be set on the compression side, retains the compression ratio information at the time of the previous compression in a file, and determines the number of reference points according to this information. Data compression method to use. 8. The data according to claim 7, wherein the reference points are increased next time if the previous compression ratio is equal to or less than the threshold value, and the reference points are decreased next time if the previous compression ratio is higher than the threshold value. Compression method. 9. When the data stream stored in the main storage unit is compressed and the code stream obtained as a result of the compression is stored in the auxiliary storage unit, the number of reference points at the time of compression can be variably set by an external instruction. The decompression side can decompress the maximum number of reference points that can be set on the compression side, retains the compression ratio information at the time of the previous compression in a file, and keeps the background A data compression method comprising: estimating a compression ratio, updating previous compression ratio information, and determining the number of reference points in accordance with compression ratio information at the time of previous compression or compression ratio information updated in the background. 10. When a data stream stored in a main storage unit is compressed and a code stream obtained as a result of the compression is stored in an auxiliary storage unit, a memory image is compressed in the background unless a user job is executed for a predetermined time. And storing the data in an auxiliary storage unit, and thereafter, when the hibernation process is subsequently performed, the compression of the memory image is omitted.