DE10310858A1 - Character string compression method for compressing computer data, whereby data on a data bus to a memory component is compared with data stored in the memory to find the longest possible match - Google Patents

Abstract

Method for compressing a character string, especially according to the LZ77 standard, whereby on a data bus connecting to a maskable content addressable memory component, a number of compressible characters are configured as at least a data word and at least a matching of at least a partial quantity of these data characters with the data content of the memory by masking of the data bus is searched for. The invention also relates to a corresponding device for implementation of the inventive method.

Description

The invention relates to a method and a device for compressing a character string from successive ones Characters, especially according to the LZ77 standard or the GZIP standard, with the longest match Character to be encoded with the data content of a content addressable Memory module is searched for and in the case of a match the matching ones Characters at least depending their address in the memory chip and the length of the match are encoded.

For compressing data e.g. Different methods often come from a character string used both individually and in combination with each other can be used. For example, the Huffman coding and a dictionary based Compression, which in its basic idea follows the LZ77 compression.

The combined use of such procedures is due to the widespread compression according to the GZIP standard known, which provides for implementation in software. Here will first a dictionary based Compression according to the LZ77 standard carried out, after which the encoded data continue to be subjected to Huffman encoding.

In this case, a solution falls on conventional computer systems the very high computing requirements and the associated large amount of time as a disadvantage.

In addition to the compression mentioned above The LZ77 standard also includes other processes in accordance with the LZ78 standard known as well as further developments, such as LZW compression, these methods are based on ideas from Lempel and Ziv or Welch. With these types of encoding, dictionaries are used that come from the compressing data itself are formed. In connection with of the present invention are under words or data words sequential combinations of any characters, e.g. Bytes, understood. The data words can therefore any length exhibit.

Basic idea of the compression standards mentioned is a data word to be compressed within the dictionary to search for and, if a match is found, the one to be compressed Word by its address, i.e. essentially its position in Dictionary too replace, which can achieve compression if the address of a word to be compressed in the dictionary with fewer characters, than would be necessary to display the data word itself.

With compression according to the LZ78 standard this becomes the dictionary from a data stream to be compressed, in particular sequential self-educated. The starting point is a dictionary in which initially only all sorts Single characters as words are listed. Subsequently then combinations appearing sequentially in the data stream from already in the dictionary listed words and single characters are added to the dictionary as newly found longer words. In this way, two are first Characters encompassing words added. The length the as words character strings to be recorded is increasing so that gradually longer ones words and word strings in the dictionary be included. Thus, in particular complex long word strings after finding them in the dictionary by a much shorter one Addressing replaced and thus compressed.

In contrast, with the LZ77 standard dictionary from already processed data. Joined a combination of characters a length of three or more characters in the past, data that has already been processed, is indicated by a Save the length and the position of the string relative to the one currently being processed Characters achieved a data reduction. The dictionary has the LZ77 standard a constant size and usually exists from an over the search windows to be encoded or data to be moved, which is a constant number based on the current position of past Character includes. Accordingly, such a search window can, for example 32,768 characters in the past include.

The LZ78 and the LZ77 standard differ is therefore significant in that the LZ78 standard is new words to be recorded just added continuously with all existing entries including their addresses persist while the LZ77 standard is based on a dynamically changing dictionary of constant size, where the addresses of the words go through permanently change the moving window.

The various compression standards mentioned and also the Huffman coding are the relevant Adequate specialist known, so a closer discussion is unnecessary here.

Data compression, in particular according to the LZ77 standard, has so far been carried out predominantly in software on microprocessors, the greatest time being required in the search for the longest possible matches of the next characters to be processed with the character strings contained in the search window. This search in particular can be implemented using a special form of a memory chip, namely a content-addressable memory. Such a memory module is also called CAM (content addressable memory). Such a memory is often also referred to as associative memory. This type of memory can be used to add a given string to the data bus of that memory is created to maintain its position in the memory, ie a CAM memory module has a causality sequence reversed to a conventional RAM (random access memory). While an address is being applied to the address bus of a RAM memory in order to read it out, a content is being applied to the CAM memory in order to obtain the associated address, if it exists. However, the writing of a CAM memory with data content is principally done in the same way as with the conventional RAM memory modules.

A hardware-based compression method according to the LZW standard is, for example, from the patent US 5,838,264 known. In the method known there, individual data, in particular characters, the size of a byte are applied to the data bus of an above-mentioned content-addressable CAM memory module in order to then obtain a possible address within the memory by means of the hardware check within the CAM module. In order to obtain a longest match of character strings with the content of the dictionary stored in the CAM module, even in the LZW method, it is provided in the known method to continue to add more and more individual characters to the data bus of the CAM module until that no more correspondence with the content of the memory chip can be determined. For example, the addresses received after each character created and the number of characters created until no more matches were found can result in compression according to the LZW standard.

In the known method it as distracting and disadvantageously felt that successively to the CAM memory chip Individual characters must be created in order to successively match to consider, until no further match more occurs. Despite the hardware being implemented, this requires Address search for a single character requires a considerable amount of software, since one character at a time from an input data stream must be created and checked in the data register of the CAM.

The object of the invention is To provide methods and apparatus by means of which simple Way several characters to be compressed (i.e. data words) on a possible complete or partial agreement checked with the content of a dictionary stored in a CAM module can, in particular also the determination of matches a length larger than 1 hardware-based.

This object is achieved according to the invention solved, that to a data bus of a maskable content addressable Memory block of several characters to be compressed at the same time are created as at least one data word and matches at least a subset or all of these characters with the data content this memory module is sought by masking the data bus become.

A device with a content addressable Memory module and control electronics for the corresponding Control and implementation of the Accordingly, the method according to the invention has a content-addressable Memory block of a known type that can be masked according to the invention is, i.e. allows masking of the data bus, which is within the scope of this invention means that by masking the locations of the data bus created data word can be selected, their correspondence to be checked with the data within the memory chip. In the case of a data word made up of several characters, it can be significant the one to be considered in the search Characters can be determined.

The data bus can be masked here for example about a separately provided masking bus, masking register or other masking devices can be provided on the CAM memory module according to the invention. A masking device accordingly indicates which of the for one Search for characters to be considered on the data bus should own and can have hardware through control logic or software can be programmed.

Compared to the above status the technology in which only the agreement of one individual Character or when creating a longer one Data word just a complete match the data word with the memory content can be determined, offers the solution of the invention particular advantage that due to a reduced significance, i.e. a reduction of all characters to be found within one Data word on only a part of characters within the data word only the occurrence of a substring in the memory block can be determined.

With such a device according to the invention and the method according to the invention there is therefore a particularly simple possibility of a longest match of the characters applied to the data bus with those within the memory chip find.

In addition to the address, which is a content addressable Provides CAM memory module after creating a data word, can also others, especially binary ones Signals are output which are e.g. a so-called “match Flag "and a" multiple Match Flag ". While that Match Flag indicates the existence of any match the multiple match flag above whether matches occur at multiple memory addresses.

With the method and the device, all compression standards can be implemented in terms of hardware, in which a longest over must be found.

To with the method and the device according to the invention for example to implement compression according to the LZ77 standard, it can be provided that for compression in a specialist known way a search window of predetermined data word width successively over the character string to be processed is moved, but now according to the invention in return from a pure software-based solution, the characters are hardware-based be stored in the memory module within the search window and an at least partial match of one Number of characters arranged in front of the search window with those in the memory block stored characters is searched.

The execution of a compression after the LZ77 standard can accordingly with the inventive method and the device can be carried out hardware-based, since all data or characters within the search window that act as a variable dictionary serve, inscribed in the content-addressable memory chip be and there for a hardware comparison with characters in front of the search window are available stand. It is preferred for a comparison step the number of characters from in front of the search window arranged characters selected such that the data word width of the content-addressable memory chip equivalent.

If such a data word is sent to the Data bus of the memory chip cannot be created, a complete and all matches in between according to the masking chosen being found.

In this case, it is particularly advantageous if each character in the search window along with N-1 trailing Characters in one memory location, especially in successive ones Storage locations of the memory chip is stored, the Memory chip has a data word width of N. So far when selecting, for example, N characters to be compressed a partial match with appropriate masking until a complete match can be found at a single address of the memory chip.

In order to hardware compression of a or several characters can be reached accordingly or several characters as a data word from one of the data word lengths of the memory chip corresponding number of characters from the string in front of the search window to the data bus of the memory module for checking at least one partial match by means of an appropriate masking.

For example, with one Data word consisting of a total of N characters initially a partial match in only one, namely e.g. the first character, checked by using a mask the masking bus or masking register is selected in such a way that only this first character is chosen as significant and the N-1 characters remaining are chosen to be insignificant.

Already with this test the least significance, no match for this individual Character with a character in the first position of any address of the memory module, this first character must be as Characters are encoded themselves and compression is not possible. In In this case, especially after the test, the symbol is displayed together with N-1 subsequent characters in a memory location of the memory chip filed and the search window moved one position. Thus, the entire content of the search window that has now been moved is displayed again available in the memory block.

Will there be a match in the significant If jobs are found, further longer matches are sought by that the significance is increased by a selected masking, for example, that the second character on the data bus is switched to significance, whereas N-2 others follow Characters remain chosen as not significant. In this manner and way can by increasing the significance after a longest match at least a substring from the characters created on the data bus be determined.

With the match found at least one partial character string of those applied to the data bus N characters can be hardware by the CAM memory chip the address of this substring within the memory chip be determined, the length of agreement is automatically implicitly known by the masking selected.

According to the method according to the invention from the determined address the position of the substring relative to the search window to determine compliance with the LZ77 standard to reach. The substring found in the memory chip can therefore according to the rules of the LZ77 standard, i.e. at least by the position within the search window and the length of the match.

After finding the longest match the search window according to the invention the length of agreement shifted further on the character string to be encoded, whereby in a write operation of each of the individual characters from the matching one Substring together with N-1 subsequent characters in turn in successive addresses of the Memory block is stored, the previous character contents overwritten at these addresses become.

Accordingly, the content of the search window is again completely mapped within the CAM memory module, but this Search window begins with a shift according to the LZ77 standard to different addresses within the CAM module. Since the start of the search window within the CAM module is controlled by electronics or software, it is easily possible to infer the position within the search window from the determined address of a character string within the CAM module.

Because the memory chip is usually has only a limited number of storage locations, this number preferably at least the length in particular exactly the length of the search window, it is provided that when registering a new data word in the memory block an old data word overwritten from the memory block becomes. For this purpose, the memory module in the sense of a Ring memory described.

If there is a match with the method according to the invention all characters determined from the data word applied to the data bus, this way, the determined address becomes the characters in the memory block again the position of the characters in the search window and the Characters at least by this position and the data word width (= Length full agreement) encodes, after which the search window is further shifted by the data word width and and each of the individual characters is composed of the data word with N-1 following characters in successive addresses of the memory chip is filed. Accordingly, there are N new data words in the Memory chip written, where N old data words in the sense of the ring memory structure from the memory module are eliminated.

If all are found to be the same Sign, there is still a chance that the following data word within the character string also a partial or full match exists, so in this case if a match is found at least one substring following a full match the length the previous full match around the length the currently determined match is incremented and the determined current address e.g. is discarded.

Furthermore, the search window turns around the length of agreement moved further and put together each of the matching characters with N-1 following characters in an address of the memory chip stored.

If several times in succession full matches of the entire data word applied to the data bus with the content of the Memory module is determined, the process step the incrementing of the determined length is repeated several times in succession, where for coding the position relative to the search window at the first complete agreement preserved.

Are with the inventive method and the device has multiple longest matches found at different addresses, for example by the multiple match flag of the CAM memory module can be displayed, it is preferably provided that there is a match with the lowest Position within the search window is selected for coding. This justifies that in a later, possibly further subsequent Huffman coding a better complementary compression can be achieved.

According to the method and the device it can also be provided that the characters within of the search window only at a single address within of the CAM memory module is stored, this one data word width that corresponds to the search window width. In this case a coding not by an address and a length in the actual, above sense can be determined, however, persists the possibility by a selected one special masking of the data bus to carry out a check for at least partial matches. in this connection can have a mask beginning with every single digit within of a data word can be programmed. This also results a special speed advantage because the CAM module only must be described once and then several tests can be carried out.

For example, the the first position within the data word should be masked as significant, whereupon if a match is found successively the other digits were switched to a high level of significance be until no further match is detected. Because at the first position of the data word with the Programming a high level of significance can start at a partial match found the substring found can be encoded in front of the search window at address 1 and length the programmed significance masking. Furthermore, then according to the inventive method whether, for example, matches begin from the second and subsequent positions within the Search window or the address of the memory chip are available and whether the match is bigger than that previously determined.

For this purpose, if, for example, the search is to start from the second position within the search window, the first position is first masked out, ie switched to no significance, where the second position is masked in or switched to high significance. If a match is found, the process of successively increasing the significance continues from the second position, until the longest match was found from the second digit. Accordingly, a partial character string found is then encoded by address 2 and the length of the significance programming. Accordingly, this procedure can be continued for all other positions within the search window. From all the matches found for each position, the one with the greatest length is selected for coding.

To optimize the hardware compression according to the invention it can be provided that the masking is dependent on statistical Properties of the string. For example, it can be provided be that more common occurring character lengths be checked when programming the masking rather than, for example not that often occurring character lengths.

A speed advantage can can also be achieved in that a write operation in which at least one character along with N-1 subsequent characters in an address of the memory chip is stored before completion a search for matches is started in such a way that the written in the memory block Signs for the next search as soon as the search is completed to disposal stand.

This optimization is based on that usually only after a completed check for maximum matches the matching one String is written into the memory chip, but what required several clock cycles of the controlling electronics, so that before the start of one new search normally waits for this number of clock cycles must be done, because only then the newly written data word or the shifted search window in memory is available for the next search according to matches. In order to optimize this, it is according to the method according to the invention furthermore, this write operation already provided a corresponding one Number of clock cycles to start before completing the search so that the new word written into the memory block at the end of the conducted Search immediately available stands without waiting for a certain number of clock cycles must become.

This can cause an error that occur before the actual completion of the search for match a data word falls out of memory, which is a match would have shown however, in spite of this eventual result Loss of compression considerable time savings in the search.

The method according to the invention can be advantageous performed with a device be a content-addressable memory chip already described and control electronics for controlling the memory module provides. Such a device can, for example, be an interface for data communication with a data processing system, which provides the data to be compressed to the device and according to the procedure then as compressed data back receives.

In this way, e.g. a plug-in card for common computer systems can be realized so that an LZ77 compression within a Computer system can be implemented on the hardware side. GZIP compression can accordingly can be realized by first hardware performed the LZ77 compression and then Huffman coding is either software-based or also hardware-based he follows. Accordingly, they result from the hardware-based Realization of the compression considerable speed advantages especially when sending data over telecommunication networks, such as the Internet.

For example, in an application a host computer, e.g. a web server, data to be compressed provide. The data stream can go through the PCI bus of this computer to a device according to the invention transported and compressed in this. You can do this from any software the data to be compressed via the Interface transferred into the device according to the invention after which the software gets the compressed data back. The Sequence of writing and reading the data to and from the device according to the invention can in the usual ways and ways that are realized by the specialist in electrical engineering and computer technology are known.

As mentioned, it is according to the method according to the invention and according to the device according to the invention provided after a done LZ77 compression apply any other compression methods to the output data stream. By one afterwards Huffman coding can accordingly achieve compression according to the GZIP standard become.

Embodiments of the invention are shown in the following illustrations. Show it:

1 : a schematic representation of a CAM memory module with different input data for checking for agreement;

2 a schematic representation of a CAM memory module with the possibility of masking with various input data for checking for partial correspondence by selecting the significance point in the data word created;

3 : an example of coding by searching for a longest match;

4 : an example of a continued encoding after a previous full match Mood;

5 : An example of a method for saving memory addresses in the CAM memory module and for increasing the speed The data compression according to the LZ77 standard has so far been carried out predominantly in software on microprocessors, the greatest time in the search for the longest possible matches of the next characters to be processed with the strings contained in the search window is required.

For this search for a data word, according to the invention, a data word can be applied to the data bus on a content-addressable memory chip. In addition to the address found, at least two further binary signals can be output, the “match flag” and the “multiple match flag”. Such a CAM module is simplified in the 1 outlined.

In the 1 The memory addresses are marked with their contents in the top left. Below, the reactions of the CAM memory module to, for example, three different input data are given line by line. The binary signal states "True" and "False" are marked accordingly by "+" or "-".

The usual CAM memory modules only have the option issue a single match address. Voices multiple memory contents matches the data word created, consequently, only one address can be output directly. The Addresses of the remaining matching ones Memory cells need then be determined by separate procedures. In the simplest Case, this can be done, for example, by successively deleting memory locations, whose addresses have already been determined. To the initial state later would have to restore finally the previously overwritten Storage locations are restored. Sometimes such functionality implemented for finding multiple addresses in the CAM module itself his.

A CAM memory module according to the invention has, for example, an additional mask bus which allows the data entered on the data bus to be "masked". The mask bus specifies which of the characters applied to the data bus for a search are significant with regard to a search according to the agreement. The procedure is in the 2 clarified. Again, the reactions of the CAM to the input signals given below on the left are shown line by line.

The mask bus is partly used of a CAM memory module also made available as a register, i.e. There is a programmable set within the CAM memory module Registers that are used for masking and that are based on different Species selected can be. So can programmed the registers once and then the desired one Masking selected accordingly become. You can also other implementation options be provided for masking.

Due to the possibility of masking, according to the invention the occurrence of partial character strings, as in the example “abcd” in the second line of the data created, can also occur 2 can be detected. Here the last character "c" of the total data word "abcdc" is masked out as not significant. However, the match can only be found if this character string occurs at the first position of the data word. The substring "acde" of the last search example is only searched for starting at the first position, but the occurrence at memory location / address 3 starting with the second position is not found.

In order to counter this problem, in the method according to the invention, each character located in the search window is stored together with the subsequent characters in a memory location of the CAM memory module, ie the entire search window itself is contained in the first column S1 of the CAM memory locations. The subsequent columns contain the subsequent characters in order to be able to detect longer character string matches. The process is described below using the example in the 3 explains:
For example, the text "ABCDEFGHIJKGHIABCDE", in which the character string GHI appears at the 7th and 12th place, is to be coded. The search window here comprises 10 characters, for example, each location in the CAM comprises 4 characters. The text is preferably from left to right to the right of the search window.

In this example, the CAM memory module has an addressing width of 10, ie the address width corresponds to the data width of the search window. The 10 storage locations of the CAM were as in the 3 shown in the top left. The following character to be encoded into the "K". The procedure for detecting at least partial character matches is the same 3 sketched below.

Left (columns write address and Content) is the change the CAM data, next to it (columns MatchData and MaskReg) the control the comparison functionality of the CAM specified. The columns Match, MultMatch and MatchAddr give the CAM reactions (Editions) again. The Length column shows the implicitly from the selected Masking existing information about the length of the match found.

In the step labeled 1., a number of characters corresponding to the data word length (here 4) of the CAM is read from the text, ie the characters “KGHI” at the designated position. These are made available to the CAM as comparison data by using these characters as a data word the da tenbus of the CAM. In this example, the mask register is programmed to recognize two matching characters, here the "KG" character, by setting the first two digits (++ ––) to significance.

Because the string in CAM so far is not included starting at the first place, so there will be no match found. Consequently, the "K" must be at the current one Position encoded as a character and not as a combination of address and length become.

In addition, in step 2 Character "K" followed by N-1 Character "GHI", ie the character string "KGHI" included in the CAM. Here, N = 4 corresponds to the data word width of the CAM. In the present In this case, storage location 0 is used, which on the one hand removed the character A, which is no longer relevant, from the search window and on the other hand the required number of storage locations in the CAM minimal, namely just the size of the search window equivalent. The CAM memory module is thus a kind of ring memory used which is rewritten from the front when the address end is reached.

The search window is around one Position of the uncoded "K" shifted and for coding the next The character "G" is now properly moved Search window starting with "B" and ending with the last written "K" after a match searched. The data in the search window is still completely in the CAM, but now starting with memory position 1 and ending with Storage position 0, i.e. if there is a match, it must always the determined address of the CAM is converted into a position in the search window which, however, e.g. with a simple address counter leaves.

For the step labeled 3. the CAM is again the next uncoded Sign "GHIA" for comparison to the disposal and it is put in the first place by masking after the string "GH" looking for beginning.

The match with storage location 6 is recognized by the CAM and output as such (Match and MatchAddr). The length of agreement is again chosen by the Masking known.

The following is through change the masking (MaskReg) by increasing the significance points the exact maximum length of agreement searched. In the present example, this is done successively from the left mask registers filled to the right with significance "+" selected or used, whereby gradually longer and longer matches begin each time be found in the first place. Other orders when selecting the Are mask registers also conceivable, for example, the search time or the Search effort for the longest match to minimize.

The location found (here 6) must be dependent the positioning of the search window in the CAM in an address or position of the search window, here 5, because the search window in the considered point in time at storage location 1 of the CAM and on Storage location 0 ends. The string "GHI" would in the present case it is coded by the number combination [5,3] ([position, length]).

In addition, in steps 6-8 you have to the characters now encoded Strings are included in the CAM, what an exact Find the longest match may only be carried out after this search has been completed. For the purpose However, it can optimize the achievable data throughput it makes sense to refrain from the exact search and these strings (Steps 6-8) in the CAM before the search is completed (Steps 3-5) such a small fluctuation in the data rate for the in to reach the data to be recorded.

The search window is as described (Steps 6-8) moved 3 positions. It starts at now CAM address 4 and ends at 3.

Four more will then be used for comparison Character "ABCD", starting with the first uncoded characters are used and the process begins from the front with steps 9 and 10.

In the example presented here for better clarity the data word width of the CAM is accepted with 4 characters and with the Find matching Strings of length 2 started. Both are variable parameters. Preferably one Hardware for realizing data compression with data word widths of 18 bytes (characters) are used, according to the GZIP standard with the search for matching strings the length three is started. This makes sense because there is only a match of 3 characters storing address and length less space in Claimed as storing the characters themselves.

Strings, the length of which exceed the data word width of the CAM e.g. detected in the manner described below.

The text used for coding in the above example is shown in the example of 4 modified to "ABCDEFGHIJKGHIJKGHABCD". The character string to be detected is now, for example, "GHIJKGH". After the detection of the correspondence of the entire data word "GHIJ" applied to the CAM for comparison in the step labeled 5, the recognized CAM address "6" is first transferred to the search window address "5" as described above and together with the length information cached for later output, for example.

The corresponding character strings are then included in the CAM as shown above, which also virtually limits of the search window within the CAM are moved properly. The processing begins in the known manner from the beginning, and this time also matches that only comprise a single character must be recognized, which is why in the present example the further processing is started with the masking of a single character in step 10. Matches that begin after the search window has been shifted to the same address as the previous unshifted search window (5 in the present example) can be taken into account by adding the length information found to the previously stored one.

In the present example the String "GHIJKGH" therefore by the Number combination [5,7] encodes, with the position information to the search window that has not yet been moved (as in step 3) refers. In this way, matching strings can be created Detect any length, by merely monitoring the location of the storage location and the length information is added to the existing one. Monitoring the shift the search window within the CAM will be of additional Processing units guaranteed. This process step can change if a complete match is found again to repeat.

The examples shown so far have yourself with simple matches employed. When matches occur at several storage locations (MultMatch = +) in the CAM can be done by one additional Control logic is the cheapest Alternative selected become. Because e.g. the basic form of the CAM modules mentioned here is only about possibility has To output exactly one address (MatchAddress), all of them must be in order first potential Addresses are read out of the CAM and then the Decision about carry out the selection of the suitable alternative. This can happen that after finding a match, the data this address deleted and buffered with the address information to then another match check make. After all multiple matches e.g. so determined could have been the data back to its original Written back addresses become. In some cases, such functionality can be used to find multiple Addresses can also be implemented in the CAM module itself. After finding of all multiple matches is the cheapest choice from all matches according to further criteria met.

When looking for the longest match within a CAM data word the examination of multiple matches does not initially play Role. Only if there are several longest matches consistently, the one with the lowest position (address) is selected within the search window, e.g. a more advantageous possibility for later coding by means of Huffmann coding.

If the matches over the Length of a CAM data word must go out In addition, according to the above described method for the detection of longer matches, multiple positions within the search window regarding an extension the match already found is monitored become. The entire processing functionality required for this can be from additional Control logic in the hardware can be guaranteed.

In addition to the exact implementation of the previously described The procedure does not result in 100% optimization Compression different ways to simplify the process. For example, like already mentioned above, to obtain one if possible constant write speed in CAM memory locations the shown Order not followed exactly. Those labeled 3rd, 4th and 5th Steps are nested with those labeled 6th, 7th and 8th, so possibly still needed Storage locations at the end of the search window are overwritten too early.

That can result in these being unfavorable Constellations not the greatest possible match is found, i.e. it gets a shorter string through the number combination to be generated [position, length] described as in the optimal case possible would have been. Since this is the case, especially with a large search window (with GZIP 32768 Characters) but extremely rare neglecting such errors can be a significant problem Increase in processing speed can be achieved.

something similar applies to an inadequate Treatment when multiple matches occur. Will be here only the only address output directly from the CAM is used, it may happen that not the longest match in the search window is found, or that the address within the search window too is chosen large so cheaper Position Length pairs exist.

The influence of this inadequacy depends on it on the data word width of the selected CAM block from. The bigger the Data word width, the lower the probability of Occurrence of multiple matches. With a sufficiently large one Data word width, the influence of these shortcomings can be neglected. Especially considering the hardware resources to be saved become so cheap Realizations possible.

The above-mentioned data word widths of, for example, 4 for CAM and 10 in the search window are only given as examples. The exemplary described The above statements also apply to any other data word widths of the CAM or the search window.

In an advantageous further development can CAM addresses through a more targeted special mask programming can be saved, which also means a speed advantage is achieved.

The previous description of the invention goes by providing a memory address for each character in the search window , making the latter in the first column of CAM locations occurs and the number of required Locations of the size of the search window equivalent.

If the first two columns are used in this way to save the search window, this number of required addresses is reduced by half. The first column contains the even and the odd character positions within the search window. To detect matches, however, more complex masking must be carried out. The 5 shows the procedure using the example presented first.

This time only every other combination that occurs of four-element strings stored in the CAM. This can be done generalize to every m-th combination of k-element strings. Consequently, the search window is now in the first in this example contain two columns of CAM locations. For detection the mask register must start once at the first and once at the second (mth) position be programmed. The order of the review, i.e. the programming of the mask register is done so that taking into account of the matches already checked exclusively at all possible Positions are checked.

In the example given, with step 3 designated, the masking "- +++" can be selected directly, without having to test "+++ -" because that Result of the latter after checking "++ ––" in the 1st designated step is already known.

This approach to reduction of the CAM memory requirements continue arbitrarily until only a single memory cell is used in the CAM, which the entire search window as a long Record data word, using the mask to check matches with everyone Must be programmed starting in the long data word.

In a further preferred training can the complex programming of the masking should be provided to detect the longest match to be omitted if a CAM is used that has the functionality of a Longest match detection on the hardware side, i.e. if the CAM can be instructed directly the location with the longest match issue.

Furthermore, the search order based on statistical properties of the text to be compressed be optimized.

By taking into account the statistical Properties of the text, whether it is predominantly longer or longer shorter matching strings further optimization can be achieved. The one in the first Example presented successive filling of the mask from the left could right For example, be modified so that frequently occurring Lengths can be checked faster than less common. The order of the check is adjusted to the given or determined statistical relationships.

In another training course it provided the data transfer rate maximize in the CAM.

Due to the special structure of a CAM usually pass from entering data to storing (creating the information to the data input) up to the possibility to take this data into account when searching several clock periods, being entered in successive clocks Data are also available for searching in successive measures. Search operations can performed simultaneously become.

For the best possible Recognition of character matches, as it was done in the first example, a Data storage (steps 6-8) should be done so that the ongoing search (steps 3-5) will properly complete before relevant CAM memory locations are overwritten by the write operation become.

For an optimized processing speed must be achieved by the Write operation caused delay if possible by anticipated Initiation of the write operation can be compensated, i.e. the write operation is initiated so many bars in advance that the effectiveness the written data is ensured immediately after the search is completed. The latter is only possible if the time for the Completing the search is known a priori, which is the goal of shortening the time to search contradicts.

Is from the requirement of correct Search has gone off, the chronological order can be changed by early Initiation of the write process can be optimized, which then, however some CAM locations can be overwritten too early, so that there is a smaller effective search window. This faulty However, treatment only appears if one is in the search window far behind, character string at the end of the search window in the current one Text appears. Because this case occurs very rarely in reality can corresponding errors, especially with regard to the increased Processing speed can be tolerated.

The application of the GZIP algorithm happens very much in software frequently based on the so-called zlib library, i.e. to use GZIP-based The compression algorithm itself does not have to include data compression a software can be implemented, but it can be based on an existing implementation the needed Algorithms can be used in this library.

Many well-known software products make use of according to the zlib library. By modifying this library, so that data compression in hardware is done in software if GZIP Accelerator hardware is available, everyone benefits on the zlib-based software without modification of the use of GZIP hardware acceleration. Because in multitasking environments Hardware is not duplicated like software a control body has been introduced the order of different compression jobs defines the zlib library.

Claims (16)

Method for compressing a character string from successive characters, in particular according to the LZ77 standard, in which a longest match of characters to be encoded is sought with the data content of a content-addressable memory chip, and in the case of a match, the matching characters at least depending on their address in the memory chip and the The length of the match is coded, characterized in that a number of characters to be compressed are applied to a data bus of a maskable content-addressable memory chip as at least one data word and at least one match of at least a subset of these characters with the data content of the memory chip is sought by masking the data bus. A method according to claim 1, characterized in that for the compression a search window of predetermined data word width successively over the character string to be processed is shifted and the characters in the search window in the memory block and a match a number of characters in front of the search window with in characters stored in the memory chip are searched for. A method according to claim 2, characterized in that every character in the search window together with (N-1) following characters in a memory location of the memory chip is stored, the memory chip having a data word width of N indicates. Method according to one of the preceding claims, characterized characterized that to compress one or more characters a data word from one corresponding to the data word length of the memory chip Number of characters from the string in front of the search window to the Data bus of the memory module for checking at least one partial match is created. Method according to one of the preceding claims, characterized characterized in that especially after checking a match a write operation is initiated when a (partial) match is found each of the individual characters from a matching (partial) character string together with N-1 subsequent characters in an address of the memory chip is filed. Method according to one of the preceding claims, characterized characterized that in the event of an undetermined match encodes the first character in front of the search window as a character itself together with (N-1) subsequent characters in a memory location of the memory chip and the search window by one Position is moved further. Method according to one of the preceding claims, characterized characterized that if there is a match at least one substring of the N characters from the determined Address of the substring in the memory block the position of the Substring is determined in the search window and the substring is coded at least by the position and the length of the match, after which the search window by the length postponed the match and where each of the individual characters from the matching Substring together with N-1 subsequent characters in one Address of the memory module is stored. Method according to one of the preceding claims, characterized characterized that if there is a match all characters from the determined address of the characters in the memory block the position of the characters in the search window is determined and the characters are encoded at least by the position and the data word width, after which the search window is moved further by the data word width and wherein each of the individual characters is composed of the data word with N-1 following characters in an address of the memory chip is filed. A method according to claim 8, characterized in that when a match is found at least one substring following a full match, the length of the previous full match is incremented by the length of the match found and the search window by the length of the Match is shifted further and each individual of the matching characters is stored together with N-1 subsequent characters in an address of the memory chip. Method according to one of the preceding claims, characterized marked that if multiple longest matches were found match at multiple locations with the lowest position within the search window for coding selected becomes. Method according to one of the preceding claims, characterized marked that the characters within the search window under a single address in a memory block with a data word width is stored that corresponds to the search window width. A method according to claim 11, characterized in that the data bus is masked in such a way that the check for (Partial) matches is programmed starting with every position in the data word. Method according to one of the preceding claims, characterized characterized that masking depending on statistical properties of the character string, especially that more frequently occurring character lengths tend to be checked become. Method according to one of the preceding claims, characterized characterized in that a write operation in which at least one Characters together with N-1 subsequent characters in an address of the Memory chip is filed before completing a search for matches is started in such a way that the ones written into the memory chip Signs for the next search as soon as the search is completed to disposal stand. Device comprising a content addressable Memory module and control electronics for performing a Method according to one of the preceding claims. Apparatus according to claim 15, characterized in that it has an interface for data communication with a data processing system, which provides data to be compressed to the device and according to the procedure according to one of the preceding claims gets compressed data back.