WO2009141161A1

WO2009141161A1 - Method for hosting a plurality of versions of memory pages in a storage system and accessing the same

Info

Publication number: WO2009141161A1
Application number: PCT/EP2009/003679
Authority: WO
Inventors: Marc Yves Maria Kramis
Original assignee: Universität Konstanz
Priority date: 2008-05-23
Filing date: 2009-05-25
Publication date: 2009-11-26
Also published as: WO2009141162A1; DE102008024809B3

Abstract

A method for hosting a plurality of versions of memory pages in a storage system, wherein memory pages (Pp) are divided into memory page cells (PCp, c₁,..., PCp, c_n) and unique memory page cell identification information (c) is associated with each memory page cell, wherein in response to a revision (R) of a certain memory page (Pp), a counter (r) of revisions (Rp) and valid for the respective memory page is changed by a value and a memory page version (PVp, r) is logically managed by the storage system, and a memory page part (PPp, r) is physically stored in the storage system. The memory page part (PPp, r) comprises at lease one memory page cell (PCp, c) modified during the revision (Rr, p) and in response to the revision (Rr, p) of a memory page (Pp), a memory page version indicator (PVRr, p) is managed by the storage system, said indicator comprising m memory page part indicators (PPRr_max,p,..., PPRr_max-m+1,p) for the memory page parts (PPp,r_max,..., PPp, r_max-m+1), wherein m is a parameter value of m>0 that can be determined.

Description

A method of maintaining and accessing a plurality of versions of memory pages in a memory system

description

The invention relates to a method for holding a plurality of versions of memory pages in one

A storage system, and a method for accessing correspondingly preserved versions of memory pages in a storage system.

From the prior art are a variety of

Storage systems known, which include file systems as well as database systems. A memory page, which is also referred to as a data page as an alternative, is to be understood, for example, as a file of a file system or a database page of a database system, depending on the respective storage system considered. A database page in this case represents a unit in which a database system handles the data, wherein it comprises a definable maximum amount of data and / or data records that the database system can read from the storage medium with a single access and / or write to the storage medium. As synonymous with the term memory page is considered in the context of this application, the term data page.

There are also known memory systems with which it is possible to store several processing statuses, ie versions, of respective memory pages and at the same time manage. This preserves earlier versions of memory pages - at least for now - and thus provides the ability to restore pages with a history that they had at some point in the past.

In practice, the term revision is often used interchangeably with the term revision. In the context of this application, however, the term version refers to the respective retrievable processing status of a memory page, while the term revision refers to the process of modifying a memory page, which may comprise a plurality of processing steps which, after being stored, ultimately lead to the new processing status of the memory page , That is a version of a

Memory page represents the result of a revision of the memory page. Modifications during a revision may be the insertion, deletion, or modification of data or a memory page. The first creation of a memory page is the first revision.

Saving multiple versions of a memory page may be advantageous in a variety of situations, such as a malfunction of the computer system or user errors such as erasure or modification of data or entire memory pages, as well as analysis of temporal changes of a memory page to a particular state.

As an example of file systems with a directly integrated and therefore transparent for the user versioning, here log-structured file systems (English: log-structured file system, LFS) called. They write data in a continuous stream to disk, so for example a disk, so that Modified memory pages are always written into a free memory space and thus no existing memory pages are overwritten. With this approach, log-structured file systems on the one hand enable high-performance writing of the data to the data carrier, since the write accesses are only a minimum

Require stylus movements. On the other hand, they provide access to earlier versions of memory pages by continuously updating the data.

There are also the so-called

Version management systems Software systems that provide the ability to version memory pages, i. H. of files, regardless of the underlying file system at the application level. Such systems are used primarily for the source code or program code version management in software development in order to log all changes made to the code and to be able to restore old development statuses if necessary. For this saves

Version management system all files belonging to a software project as well as the respective modifications made in a so-called repository. Version management systems are available both as a stand-alone software application and in an integrated form, for example for software development environments but also for word processing and

Spreadsheet software and content management systems. Known examples of version management systems are CVS (Concurrent Versions System) or Mercurial.

In addition to the file systems with versioning and the version management system, storing multiple versions of memory pages also takes place in backup systems. However, unlike the first two systems in which the versions are automatically saved as soon as a memory page, ie a file, has been modified, saving new versions of backup systems is also automatic but usually dependent on fixed, regular times and then also for several files of a directory, a partition or even an entire computer system. In addition, the created so-called backup copies are stored on separate storage media, while file systems with versioning and

Version management systems write the versions of the memory pages to the same storage medium.

Furthermore, there are also database systems which, for example by means of correspondingly implemented database management systems, offer the possibility of versioning data, so that the data is not only available in the current version but can also be restored as it was at a certain earlier point in time.

In its earlier German Patent Application 10 2008 024809.6-53, the disclosure of which is hereby incorporated by reference, Applicant describes a method which, in particular in connection with native XML databases, makes it possible to store several versions of tree-structured data families.

A relatively straightforward way of having multiple versions of a memory page available is to store the entire memory page, including the modifications for each revision. This approach, which is similar to the copy-on-write approach, may be advantageous to have, that the respective desired version of the memory page is immediately accessible. However, it mainly leads to a considerable storage space consumption on the storage medium.

Based on the knowledge that two successive versions of a memory page usually differ only slightly, so that would be redundant in a respective complete storage of the memory page after each revision a large amount of data, in the so-called difference storage after each revision only the modified and thus the different data compared to the previous version of the memory page, the so-called deltas stored. For example, such a delta may consist of a binary difference between the newer and the previous version of a memory page. In this way, differential storage or delta encoding leads to a reduction of storage space usage in the

Storage of sequential data, that is data that exists in several versions, and thus can also be considered as a form of data compression. In contrast to the storage of always complete memory pages per version, the differential storage not only reduces the storage space requirement but also the time required to write the data to the storage medium. This method of delta storage can be applied as often as desired and eventually provides a complete history of the versions of a memory page. Thus, according to the principle of delta storage, each memory page would only have to be completely stored once more, namely at its first creation. A major disadvantage of this type of storage of versions of a memory page, however, is that to provide the respective desired version of the memory page a considerable effort is required if there are already a number of other versions between the first and the desired version. Because starting from the completely stored memory page of the first revision must then be completed in succession for all other versions to the desired the only deposited as a delta modifications in order to obtain the desired memory page version.

However, to enable the reconstruction of each version of a memory page with reasonable effort and speed, a so-called snapshot is now written at regular intervals after a revision of a memory page instead of a delta. As with the first revision, the entire memory page is written to the storage medium with a snapshot. A snapshot covers all the data in a memory page, including the modifications of all previous revisions, as well as the revision to which it will be saved. Thus, instead of the first version of a memory page, the respective closest snapshot can then serve as a new starting point in the reconstruction of respectively subsequent versions of a memory page for which only one delta has been written, so that only a relatively small number of deltas is to be evaluated. The frequency at which a snapshot is saved instead of a delta can be defined automatically, for example, time-dependent, in particular for data backup systems, and also dynamically, for example, depending on the number of revisions or the size or size of the modifications since the last one snapshot. In its above-mentioned earlier German patent application 10 2008 024809.6-53, the applicant describes, inter alia, a further possibility for efficiently storing the modifications made to a data page during a revision. It is exploited there that the data in the data page as a node of an encoded tree, and thus the data content of the data page shows a unique structure. If modifications are made to the data of a data page during a revision, these can be assigned to one or more specific nodes and then each node modified during the revision is completely stored in its modified version in a new data page part for that version of the data page, so that the Processing status of a respective node directly from this data page part is recognizable. In a determinable frequency, the data page part stored after a revision is a snapshot, such a data page part also containing the nodes not modified during this revision, and thus all nodes of the data page. Along with a data page part which is not a snapshot, reference pointers are also stored on the data page parts for previous versions of the data page as well as on the last previously stored data page part, which is a snapshot. For reconstructing a particular data page verse, the respective referenced data page parts are to be considered, wherein the nodes contained in the referenced snapshot corresponding to the other referenced data page parts are either replaced by the modified node versions contained therein or to be deleted or supplemented with additional nodes. Compared to the method of differential or delta storage can in the approach according to the above-mentioned German patent application 10 2008 024809.6-53 a certain version of a data page so to speak composed in a modular manner from the nodes contained in the referenced data page parts, while consuming calculations, such as Example delta calculations would be required.

A problem with all these known methods, however, continues to be that the data volume to be written to the storage medium after a revision varies relatively widely and is therefore distributed very unevenly when viewing several successive memory page versions, since either only a comparatively small delta or data page part or but to write a comparatively large snapshot is.

A disadvantage of the known procedures also that the number of accesses to the storage medium, which are necessary for the reconstruction of a memory page varies relatively widely and the required effort is therefore not predictable, since the number of memory accesses depends initially on whether for the desired version a snapshot was saved, so that the memory page is directly available, or only a delta, so that in addition still the closest previous snapshot and any deltas still intervening there are to read and process. In addition, the variance depends on the number of

Reconstruction of a memory page version of course requires memory accesses directly from the (dynamic) frequency in which snapshots were stored. With regard to the method according to the above-mentioned German patent application 10 2008 024809.6-53, the problem has been noticed that too much memory space is still needed, especially in the case when there are frequently alternating modifications between two nodes of a data page from revision to revision , This case leads to the fact that snapshots of the data page, including the respectively unmodified node, are stored correspondingly often. Above all, this method has hitherto been used for data pages with clearly structured data content, for example in the form of nodes of a tree.

It is therefore an object of the invention to further develop a method for holding a plurality of versions of memory pages in a memory system such that it is suitable for memory pages with essentially any data content and thereby preferably eliminates at least one of the aforementioned disadvantages of previously known methods.

The inventive solution of the problem is given in a surprising way already by a method according to claim 1.

Accordingly, the invention provides a method of maintaining a plurality of versions of memory pages in a memory system in which memory pages are divided into a number of memory page cells and a unique one of the memory page cells

Memory page cell identification information is assigned. In addition, in response to a revision of a particular memory page, a counter for revisions applicable to the respective memory page is changed by one count, in particular always in the same way changed, such as up or down, so for example, incremented, and logically manages a memory page version of the storage system, but physically a memory page part stored in the storage system. In this case, such a memory page part contains at least one memory page cell modified during this revision. Further, a memory page will respond in response to each revision

Memory page version pointer managed by the memory system, said memory page version pointer in turn m memory page partial pointer to the corresponding memory page parts, where m is a definable parameter value of m> 0.

In this case, each memory page is preferably subdivided into a number of memory page cells.

Also preferably, each of the memory page cells is assigned unique memory page cell identification information.

It should be understood that the revision counter provides both the revision revision number and the version number of the versions, by which the individual revisions and versions are clearly distinguishable, with a versinon representing a revision with a fixed number.

In addition, the invention also proposes a method according to claim 10 for accessing according to the method

Claim 1 held versions of memory pages in a storage system before, in which under

Consideration of the desired

Memory page version of the memory system managed memory page version pointer and the m of this Memory page sub-pointers have the desired memory page version reconstructed and provided from the memory page cells contained in the referenced memory page parts by assembling these memory page cells. It is, if one

Memory page cell in several of the m referenced memory page parts is included, used that memory page cell, which was included in the newer of these memory page parts.

In addition, the invention provides a memory system which is prepared for carrying out the method according to claim 1 and / or 10 or the method according to one of the preferred embodiments described below, and a computer-readable medium with a computer program stored thereon, which defines instructions which control a computer or the memory system according to the invention such that it or this is suitable for carrying out a method according to claim 1 and / or 10 or a method according to one of the preferred embodiments described below.

By means of the method according to claims 1 and 10 or one of the embodiments described below, it is now possible to have a plurality of versions of memory pages with essentially any data content in a memory system ready to be accessed. Furthermore, these methods are essentially transferable and applicable for any storage systems, but especially for the storage systems described at the beginning, which hitherto worked according to one of the known methods for storing a plurality of versions of memory pages. An advantageous aspect of the invention was, inter alia, in the method according to the aforementioned German Patent Application 10 2008 024809.6-53 of the applicant from the structured nature of the data of a data page for the storage of multiple versions of this data page resulting advantages thereby for any other Memory pages can be used by the memory pages are divided into cells, these cells then offer a similar structure as the nodes of the aforementioned data page and are handled in a corresponding advantageous manner.

Another aspect of the invention is to no longer store snapshots for entire pages of memory, due to the recognition that a memory page cell, which is completely stored in its modified version in response to a revision, so that the state of processing of the cell immediately it is recognizable, in a way already one

Represents memory page cell snapshot. Thus, it also became clear that a snapshot of an entire memory page whose function it was previously to merge the individual modifications stored as a number of deltas on one memory page and a new one

Starting point for the reconstruction of subsequent memory page versions to create, in principle has become superfluous. In accordance with the invention, therefore, a memory page snapshot is decomposed into a number of memory page cell snapshots stored on the storage medium distributed over a number of m memory page portions.

Thus, an already resulting from the method according to claim 1 advantage is that after a revision in each case on the storage medium to be written data volume evenly distributed, since instead of a relatively large memory page snapshot after every third revision of a memory page now only some relatively small

Memory page cell snapshots are written to the storage medium after each revision of a memory page.

In addition, thanks to the method of claim 1, the number of accesses to the storage medium required during the method of claim 10 for reconstructing and providing a particular memory page version now varies less and the effort required is thus more predictable, since all for a particular memory page version required memory page cells in the m referenced memory page parts are included. It is additionally advantageous that, on the one hand, an inherently parallel read access to the referenced memory page parts and the contained memory page cells is possible, and, on the other hand, no deltas based on a snapshot are to be sequentially evaluated, but rather the respective memory page version can be modulated in a simple manner from the referenced memory page cells is composable, which accelerates the reconstruction again.

Further preferred and / or advantageous embodiments of the method according to claim 1 emerge from the respective subclaims.

Preferably, the memory page part contains each memory page cell modified during the revision. More preferably, the memory page part contains the copy of at least one, expediently, each memory page cell last contained in the memory page part previously stored in revisions.

Thanks to such a copy, there is direct accessibility for the corresponding memory page cell during a later reconstruction of the memory page via the memory page version pointer and its m memory page sub-pointer, without any further access

Memory page sub-pointers, calculations or searches would be required.

By means of this preferred embodiment of the method can be compared to the previously known methods a

Reduction of the required memory space can be achieved by up to one third, especially in the case when there are often alternately between two nodes of a memory page from revision to revision modifications. Because, provided that the parameter m is set to a value greater than or equal to 2 (two), a memory page portion containing only this memory page cell in the fully modified version is written in response to the revision of a single memory page cell an immediately following revision in which the other memory page cell has been modified

Memory page part is stored, which contains only this other memory page cell in the full modified version. So if the

Parameter value m is greater than 1 (one), so even after further revisions, which relate alternately to the modification of one or the other cell, no memory page part would be written to the storage medium, which in the form of a memory page snapshot both Memory page cells contains and thus redundant way again each unmodified memory page cell, as would otherwise be provided in one of the previously known procedures.

It is preferably provided that for each memory page cell of a memory page both their own

Memory page cell identification information and also each memory page cell identification information c of

Memory page cells located in the memory page Pp in their immediate vicinity and each of a ranking information related to a respective one of these memory page cells are managed by the memory system, and that for each memory page cell in the memory page part both their own

Memory page cell identification information as well as each memory page cell identification information of the memory page cells located therein

Memory page version located in their immediate vicinity and each of a ranking information with respect to a respective one of these memory page cells are managed by the storage system.

It is advantageous if, according to an embodiment of the method, for each memory page cell in the memory page part, information about the reason for its storage during this revision is managed by the memory system.

Appropriately, as reasons for the storage of a memory page cell within a memory page part at least their insertion into the memory page, their change and their deletion from the Memory page and their Modifikationsfreiheit distinguishable for a number of m revisions of the memory page. As a reason for storing a memory page cell within a memory page part, however, a read access may already preferably suffice.

Particularly preferably, for each memory page cell contained in a memory page part, another memory page sub-pointer to the memory page part, which contains the memory page cell in the respectively older modified version, is managed by the memory system. Preferably, it is sufficient, in particular in order to keep the storage space requirement low, only to note the difference between the version numbers of the two versions, to which the memory page parts belong, which contain the respectively considered memory page cell or memory page cell in the next older modified version that at any time by a simple calculation determines the correct version number and the ensprechenden

Memory page version pointer and the associated, anyway existing memory page sub-pointer to the respective required memory page part can be accessed.

With regard to the parameter value m, it is preferably provided that this applies as valid for the entire memory system and / or in each case valid for a number of specific memory pages and / or dynamically as a function of the current utilization of the memory system with regard to the read and / or write accesses can be stated on the storage medium.

According to another preferred embodiment of the method, each memory page cell of a memory page may contain a variable number of data bytes. Advantageously, it is additionally or alternatively provided that each memory page can contain a variable or, for better handling of the memory page for the memory system, a maximum of a definable number of memory page cells.

With regard to the method according to claim 10, further preferred and / or advantageous embodiments emerge from the respective subclaims.

Thus, it is preferably provided that the reconstruction and provision of the desired memory page version taking into account the information managed by the memory system to the respective memory page cell about the reason for their storage during the respective revision, and / or taking into account the memory page cell identification information managed by the memory system to the respective memory page cell the memory page cells located in the respective memory page version in their immediate vicinity as well as the respectively managed ranking information with respect to a respective one of these memory page cells.

These and other embodiments of the method according to the invention and further associated advantages are explained below with reference to an exemplary embodiment with reference to FIG. 1, without, however, restricting the invention thereto. It should be understood that the graphic representation is to be regarded merely as a roughly schematic sketch of the data structures used in the method according to the invention. FIG. 1 shows an exemplary, schematic illustration of a plurality of versions of a memory page held in accordance with a preferred embodiment of the method according to the invention.

From FIG. 1, the advantageous aspect of the invention is very clearly shown, namely to subdivide a memory page from which several memory page versions are to be kept into a number of memory page cells.

In addition, it can be seen from FIG. 1 that a total of six memory page versions PVp, r of PVi, i to PVi, 6 are kept in a memory system (not shown) from a memory page P1. Each of these seven

Memory page versions represents a state of processing and thus the result of a corresponding revision Rr, p from Ri, l to R7, i, wherein during each revision a number of specific modifications to the memory page Pi has been made. The individual processes during each of the revisions Ri, i to R7, i will be described in detail below.

For ease of comprehension, but without limiting the invention thereto, the following explanations are based on a database system as a storage system in which the versions PVi, i to PVi, 6 of the memory page Pi are kept. With reference to a database system, a memory page is thus to be understood as a database page in accordance with the explanation given at the outset.

In principle, however, the method according to the invention is also applicable to essentially any other memory systems and in particular to data backup systems, Versioning systems and versioning file systems, such as log-structured file systems, are applicable.

Revision Ri, 1: First, the database page Pi is created with a lot of data contained. This represents the first revision Ri, i of the database page Pi, hereinafter also referred to as page Pi for short. The result of this first revision Ri, i logically forms the first database page version PVi, i of the database page Pi.

Prior to the physical storage of the page Pi in the database system or on a storage medium located in the access of the database system, which is also not shown in detail in FIG

Database side Pi is divided into two database page cells PCp, c, each of the database page cells a unique database page cell identification information c, hereafter referred to as cell identification information c, so that the two memory page cells as PCi, i and PCi, 2, hereinafter also briefly only page cells or Cells PCi, l and PCi, 2 are identifiable.

Physically stored is a so-called

Database page part PPp, r, ie PPi, i, hereinafter also referred to briefly only side part PPi, i, which in the case of the first revision Ri, i contains the complete data content of the page Pi, so both side cells PCi, i and PCi, 2.

Logically, the database system manages the first database page version PVi, i, hereinafter also referred to simply as page version or version PVi, i, for example, in a database-internal, list, table or tree-structured directory Version pointer PVRi, i for the version PVi, i is recorded. Although the parameter m is set to the value 2, the version pointer PVRi, i only includes a page sub pointer PPRi, i referencing the page part PPi, l, since it is the first version of the page Pi and consequently no further, the does not mean an older side part for the side Pi, which could be referenced with a second side part pointer. Namely, a function of the parameter m is to indicate how many side part pointers PPRr, p include a version pointer PVRr, p. A second function of the parameter m, for which a value, in particular an integer value, greater than 0 (zero) can be freely defined, will be discussed later.

For each page cell PCp, r stored after a revision Rr, p, the database system also manages information as to why the cell is contained in the respective page part PPp, r, that is to say in particular as to which type of modification at the respective cell PCp, r during a revision Rr, p. This information can, for example, also be recorded in a database-internal, list, table or tree-structured directory together with the unique cell identification information c of the respective memory page cell PCp, c. When

Modification types and thus also as reasons for the storage of a page cell PCp, c can be distinguished in particular the insertion, modification and deletion of a cell, and preferably also a read access to a cell. Another reason why a side cell PCp, c may be included in a side part PPp, r will be discussed in more detail later.

Particularly preferably, the database system manages together with the cell identification information c and the Information regarding the storage reason also another page sub-pointer PPRr, p, which refers to that side part PPp, r, which contains the side cell PCp, c in the next older version. On an advantage of this, will be discussed in more detail later.

Like the second page sub-pointer of the version pointer PVRi, l, however, after the revision Ri, i also can not give the further page sub-pointer since this is the first version of the page Pi and consequently no further, ie no older, side part for the page Pi, which could be referenced with another page sub-pointer.

The information about the reason for storing the page cells PCi, i and PCi, 2 within the page part PPi, 1 is shown in FIG. 1 as 'I' for engl, 'inserted ¹ ' in the respective one, to simplify the illustration Side cell shown. Analogously, in other illustrated side cells, which will be discussed in more detail later, a 'C for engl,' changed 'so' changed 'or a' D 'for engl,' deleted 'so' deleted '. The digit after I, C, or D symbolizes the further page sub-pointer, which each referenced that side part PPp, r, which contains the side cell PCp, c in the next older version. Preferably, however, especially in order to keep the storage space requirement low, no further page sub-pointers are used, but instead merely notes the difference between the version numbers of the two versions, to which the memory page parts belong, which the memory page cell in the current or in each next older modified version. This version number difference is in By means of it, the correct version number can be determined at any time by a simple calculation and accessed via the ensprechenden memory page version pointer and the associated, anyway existing memory page sub-pointer to the respective required memory page part.

After saving the side part PPi, i, the first revision Ri, i of the database page Pi is completed and a first page version PVi, i is now kept in the database system.

The database system is configured such that each page cell PCp, c may contain a variable number of bytes of data, but is limited by the fact that each database page Pp should only contain a maximum of 128 memory page cells PCp, c to keep it manageable. In principle, however, another maximum value could also be defected or the number of side cells PCp, c could be left variable.

Revision R2, i: At a later time, the first version PVi, i of the database page Pi is called again so that further modifications can be made to it during a second revision R2, i. Accessing the page version PVi, i via the page version pointer PVRi, i managed by the database system and the page sub pointer PPRI, i. Since only one side part PPi, i, which contains both existing side cells PCi, l and PCi, 2 and thus the complete data content of the first version PVi, 1 of the page Pi, is referenced, the page version PVi, 1 can be readily provided , As a further modification on the side Pi, a change takes place on the side cell PCi, 1, while the cell PCi, 2 remains unmodified. This further revision of the page is detected and, in response thereto, the revision counter r, which has hitherto had the value 1, for this memory page Pi increased by the count value 1, ie to 2. The result of this revision R2, i logically represents the page version PVi, 2. Physically, a side part PPi, 2 is stored which, in contrast to the revision Ri, 1 after the revision R2, i does not contain both side cells but only the again modified side cell PCi , 1 contains. Accordingly, physically one page cell is now less stored than logically associated with the version PVi, 2 of the database page Pi. For the logical management of the page version PVi, 2, in the above-mentioned internal directory of the database system, another version pointer PVR2, i with the two

Side sub-pointers PPR2, i and PPRi, 1 on the side parts PPi, 2 and PPi, 1 added. In addition, the database system manages as information about the reason for storing the page cell PCi, 1, that this has been changed and as a replacement for another page sub-pointer on the page part, which contains the next older version of the page cell PCi, 1, only a version number difference to the Side part PPi, 1 leads. In FIG. 1, this is the same as that mentioned above

Explanation with a C and a 1 in the cell PCi, 1, which is included in the side part PPi, 2, symbolizes.

After the storage of the side part PPi, 2, the second revision R2, i of the database page Pi is completed, and in the database system, besides the first page version PVi, 1, also a second version PVi, 2 is kept ready.

Revision R3, i: At a later time, the second version PVi, 2 of the database page Pi is called again so that again further modifications can be made during a third revision R3, i. The page version PVi, 2 is accessed via the page version pointer PVR2, i managed by the database system, and the two page sub-pointers PPR2, i and PPRi, 1. As already mentioned, the number of page sub-pointers is based on the value defined for the parameter m, in this example with m = 2 is defined.

Since in the one referenced side part PPi, 2 only the

Side cell PCi, 1 contained, results from the fact that the cell PCi, 2, which is included in the other referenced page part PPi, 1, also for the second page version PVi, 2 is still valid. However, for the cell PCi, 1 likewise contained in the side part PPi, 1, it is recognized that it represents the older version in relation to the cell PCi, 1 contained in the side part PPi, 2. Thus, the page version PVi, 2 of the database page Pi is reconstructed and provided from the page cell PCi, 1 contained in the page part PPi, 2, and the page cell PCi, 2 contained in the page part PPi, 1.

As a further modification on the side Pi now takes place a change to the side cell PCi, 2, while the cell PCi, 1 this time remains unmodified. This further revision of the page is detected and, in response thereto, the revision counter r, which has hitherto had the value 2, for this memory page Pi is increased by the count value 1, ie to 3. Logically, the result of this revision R3, i represents the page version PVi, 3. Physically, a side part PPi, 3 is stored, which again contains only one cell, namely the modified cell PCi, 2 this time. Accordingly, physically one page cell is less stored again than logically belonging to the version PVi, 3 of the database page Pi. For the logical management of the page version PVi, 3, a further version pointer PVR3, i with the two page sub-pointers PPR3, i and PPR2, i is added to the side parts PPi, 3 and PPi, 2 in the above-mentioned internal directory of the database system. In addition, the database system manages as information about the reason for storing the page cell PCi, 2, that it has been changed and as a replacement for another page sub-pointer on the page part, which contains the next older version of the page cell PC2, i, only one

Version number difference leading to the side part PPi, 2. In the figure 1 this is according to the above-mentioned explanation with a C and a 2 in the cell PCi, 2, in the side part.

After storing the side part PPi, 3, the third revision R3, i of the database page Pi is completed and in the database system, a third page version PVi, 3 of the database page Pi is now already available.

Revision Ra, 1: At a later time, the third version PVi, 3 of the database page Pi is called again, so that a further modification can be made to it during a fourth revision R4, i. The access to the page version PVi, 3 via the managed by the database system page version pointer PVR3, i and the two page sub-pointers PPR3, i and PPR2, i.

Since in the referenced side part PPi, 3 only the

Side cell PCi, 2, results from the fact that the cell PCi, 1, which is included in the other referenced page part PPi, 2, for the third page version PVi, 3 is still valid. The page version PVi, 3 of the database page Pi is thus made in the Side part PPi, 2 contained side cell PCi, 1 and contained in the side part PPi, 3 side cell PCi, 2 reconstructed and provided.

As can be seen from FIG. 1, the additional version number differences managed by the database system for both the side cell PCi, 2 in the side part PPi, 3 and for the side cell PCi, 1 in the side part PPi, 2, which in both cases to the Side part PPi, 1 lead, in a simple manner, access to the next older or second most up-to-date version of the respective cell, although these are not in one of the side parts PPi referenced by the two side part pointers PPR3, i and PPR2, i of the managed side version pointer PVR3, i 3 and PPi, 2 are included.

With regard to the further modification on the side Pi, it is initially noticeable that a new cell PCi, 3 is inserted in the database page Pi, while the cell PCi, 2 remains unmodified. On the side cell PCi, 1 in

Side part PPi, 4 will be discussed below.

For example, a database page Pp includes nodes of a tree, each node comprising specific data from a set of tree-structured, hierarchically linked data, such as XML (extensible markup language) XML, and each Node such as a page cell PCp, c is manageable, then inserting a new node into the tree means inserting a new cell PCp, c into a database page Pp.

For example, one node may include an XML element and another node may include the textual content of the XML element or a first XML element in the tree structure hierarchically parent parent XML element or a hierarchically child child XML element. Furthermore, nodes may also include other XML components such as attributes of an XML element or the value assignment of an attribute.

If, for example, a page contains nodes as page cells PCp, c, where the nodes each comprise an XML component from an XML tree structure of hierarchically structured data, then it may be necessary in addition to the possibility of inserting new cells that already exists a read access to one or more nodes, that is, to the data respectively covered by the latter, to a new revision and to a new page version PVp, r, and thus to storage of a page part PPp, r, which gives the cells read access is carried out, so that at a later time each read access to a cell is traceable.

In addition, it should be noted that although it is described in the context of this embodiment, it is not absolutely necessary that before each modification of a page cell PCp, c, the entire associated page is first reconstructed. Rather, it is also possible to access and modify a particular cell, with the modification or only the read access to that cell meaning a revision for the page containing the cell.

Each node becomes a unique one

Identification information c assigned. Furthermore, for each node, the database system manages, in addition to its own identification information c, also the identification information c of the nodes, which XML Include constituents which were hierarchically directly superordinate, subordinate or directly adjacent to each other in the XML tree structure, in each case together with the corresponding ranking information. This has the advantage that the nodes or page cells PCp, c in the database page Pp can be stored in substantially any order, since the original hierarchical structure of the XML data contained in the nodes can be reconstructed from the managed identification information c and ranking information ,

Thus, assuming that for the example shown in FIG. 1, the page cells PCi, i and PCi correspond to 2 nodes that comprise certain XML components of an XML tree structure, then the modification with respect to the cell PCi, 3 reflects the insertion a new node that includes a particular XML part into the XML tree.

Assuming that the cell PCi, 1 is a node comprising an XML element, while the cell PCi, 2 is a node containing the associated text content, and that the node corresponding to the cell PCi, 3 has another XML element comprises, in the XML structure to the XML element, which is comprised of the cell corresponding to the cell PCi, 1 parent should be parent as parent XML element, then the respective two nodes or cells PCi, 1 and PCi, 3 related ranking information adapted from the database system.

Thanks to the managed identification information c and ranking information for the individual nodes or cells, the newly inserted node or the cell PCi, 3 in the physically stored side part PPi, 4 need not be at the top. This further revision of the page Pi is detected and in response thereto the revision counter r, which has hitherto had the value 3, for this memory page Pi by the count value 1, ie to 4, increased. Logically speaking, the result of this revision R4, i represents the page version PVi, 4. Since, during the revision R ₄ , i, a modification on the side Pi took place merely in the form of an insertion of the cell PCi, 3, the physically stored side part PPi needed , 4 also contain only the cell PCi, 3. If, however, one follows the two page sub-pointers PPR4, i and PPR3, i which would belong to a version pointer PVR ₄ , i, then it is found that neither in the page section PCi, 5 nor in the page section PCi, 4 contain the page cell PCi, i would.

Therefore, the parameter value m and in particular its second function should be discussed again here.

In addition to specifying the number of page sub-pointers PPRr, p, which includes a version pointer PVRr, p, the second function of the parameter value m is to indicate after how many revisions Rr, p of a page Pp during which a page cell PCp, c of that page has not been modified, a copy of this side cell PCp, c must be contained in a side part PPp, r.

In this way it is ensured that in a number of m side parts PPp, r referenced by a number of m side part pointers PPRr, p of a version pointer PVRr, p, each side cell PCp, c belonging to a memory page Pp at least once in one of the referenced page parts PPp, r, and thus for a reconstruction of the memory page Pp in the page version referenced by the version pointer PVRr, p PVp, r is available.

Thus, the copying of a page cell PCp, c into a page part PPp, r after a number of m revisions Rr, p of a page Pp during which the page cell PCp, c of this page has not been modified, constitutes one of the modification types insert, modify and delete Side cell, another distinguishable reason why a side cell PCp, c is stored during a revision Rr, p.

By suitably specifying the parameter value m, the method according to the invention can be optimized in the direction of the data volume to be stored, the time required for storage and the required storage space on the storage medium or in the direction of the number of read accesses to the storage medium and the time required to reconstruct and provide the storage page in the desired version. Due to this importance of the parameter value m the memory system, it is particularly preferred that this is not only fixed once rigidly for the entire memory system, but dynamically depending on the prevailing load or load of the memory system in terms of read and / or read accesses the storage medium is definable and, additionally or alternatively, a value m can also be defined for specific individual or groups of storage pages, for example as a function of their size or their frequency or scope of updating.

With a view back to the revision R3, i it should be noted that there is the case in which by means of the method according to one of claims 1 to 9, a reduction of the required memory space by up to one third compared to the previously known methods is achieved. Because there is an alternating modification between two cells of a memory page from revision R2, i to revision R3, i, wherein even with a parameter value of m = 2, the side parts of the revisions R2, l and R3, l each contain only the modified side cell , By means of the method according to one of claims 1 to 9, even after further revisions, which relate alternately to the modification of one or the other cell, the side parts from the revisions R2, l and R3, l would each contain only the modified side cell. In contrast to this, by means of one of the previously known methods with a parameter value of m = 2, the side part PP1, 3 from the revision R3, i would be a so-called snapshot and the complete memory page, ie both side cells and not just the modified side cell PCi, 2 included.

For the revision R4, i, a parameter of m = 2 means that from the side cell PCi, 1, as it was still available via the second side part pointer PPR2, i of the version pointer PVR3, i in the side part PPi, 2, a copy in the side part PPi, 4 is included.

Physically, therefore, a side part PPi, 4 is stored, which contains the newly inserted cell PCi, 3 and the copied cell PCi, 1. Thus, physically one again

Page cell less stored than logical to the version PVi, 4 belonging to the database page Pi.

For logically managing the page version PVi, 4, a new version pointer PVR4, i with the two page sub-pointers PPR4, i and PPR3, i is added to the page sections PPi, 4 and PPi, 3 in the above-mentioned internal directory of the database system. In addition, the database system, as information about the reason for storing the page cell PCi, 1, manages that it is a copy without being changed, and a version number difference which leads to the side part PPi, 2, which contains the version of the side cell PCi, i, which serves as a copy sheet, so to speak. Since the cell PCi, 3 has been inserted for the first time in this version PVi, 4, there can therefore be no further, that is, no older,

Give side part. In FIG. 1, this is symbolized by an I and a 0 (zero) for the cell PCi, 3 according to the above-mentioned explanation.

The following symbolism applies to the copied cell PCi, i in FIG. 1: The C and the 1 in parentheses were adopted from the version of the side cell PCi, i in the side part PPi, 2 which serves as the copy template. The 2 outside the parenthesis represents the version number difference leading to the page part PPi, 2, which, so to speak, the as

Copy template serving side cell PCi, 1 contains.

After storing the page part PPi, 4, the fourth revision R4, i of the database page Pi is completed and stored in the database system _. em already has a fourth page version PVi, 4 ready.

Revision Rs, 1: At a later time, the fourth version PVi, 4 of the database page Pi is called again so that modifications can be made to it again during a fifth revision Rs, i. The accessing to the page version PVi, 4 via the managed by the database system page version pointer PVR4, i and the two page sub-pointers PPR4, i and PPR3, i.

Since only the page cells PCi, 1 and PCi, 3 are contained in the referenced page part PPi, 4, it follows that the cell PCi, 2 contained in the other referenced page part PP3, i, for the fourth page version PVi, 4 is still valid. Thus, the page version PVi, 4 of the database page Pi is reconstructed and provided from the page cells PCi, 1 and PCi, 3 contained in the page part PPi, 4 and the page cell PCi, 2 contained in the page part PP3, i.

However, assuming that the page cells PCi, 1, PCi, 2, and PCi correspond to 3 nodes that comprise certain XML components of an XML tree, then when reconstructing and providing the page version

PVi, 4 of the database page Pi from the page cells PCi, 1, PCi, 2 and PCi, 3 above all the other from the database system to the respective cell, that is, the jeweilgen knot managed information for brights. For each node manages the database system in addition to its own identification information c and the identification information c of the nodes, which comprise XML components that were in the XML tree hierarchically directly superordinate, subordinate or at the same level directly adjacent, respectively together with the corresponding ranking information.

Thus, taking into account these further identification information c managed by the database system and ranking information during the reconstruction of the page version PVi, 4 of the database page Pi from the page cells PCi, 1, PCi, 2 and PCi, 3, it becomes clear that the cell PCi, 1 is an XML Element, while the cell PCi, 2 is a node containing the corresponding textual content, and that corresponding to the cell PCi, 3

Node comprises a further XML element which, in the XML tree structure, is superior to the XML element which is included in the node corresponding to the cell PCi, 1 as the parent XML element. Only the order of the nodes within the data page would be hierachical Structure was not apparent.

As a modification on the side Pi, there is a change to the newly inserted during the revision FU, i cell PCl, 3 or the corresponding node. Nothing has been changed in the ranking of the nodes, however, so that the ranking information managed by the database system for the version PVi, 5 of the page Pi can be retained.

This further revision of the page is recognized and incorporated in

Reaction thereon, the revision counter r, which has previously been the value 4, for this memory page Pi by the count value 1, ie, increased to 5.

Since during the revision Rs, i a modification on the side Pi took place only in the form of a change to the cell PCi, 3, the physically stored side part PPi, 5 also needed to contain only the modified version of the cell PCi, 3. If, however, one follows the two page sub-pointers PPR5, i and PPR ₄ , i which would belong to a version pointer PVRs, i, then it is noted that neither in the page section PCi, 5 nor in the page section PCi, 4 contain the page cell PCi, 2 would.

However, for the revision Rs, i, a parameter of m = 2 means that one copy is available from the page cell PCi, 2, which was still available via the second page sub-pointer PPR3, i of the version pointer PVR ₄ , 1 in the page part PPi, 3 in the side part PPl, 5 is included.

Logically, the result of this revision Rs, 1 represents the page version PVi, 5. Physically, a page portion PPi, 5 is stored, which contains only the modified cell PCi, 3 and the copied cell PCi, 2. Accordingly, physically one page cell is less stored than logical belong to the version PVi, 5 of the database page Pi.

For the logical management of the page version PVi, 5, a new version pointer PVRs, 1 with the two page sub-pointers PPR5, i and PPR4, i is added to the page sections PPi, 5 and PPi, 4 in the above-mentioned internal directory of the database system. In addition, the database system manages, as information about the reason for storing the page cell PCi, 2, that it is a copy without being changed, and a version number difference to the page part

PPi, 3, which contains the version of the side cell PCi, 2, which serves as a copy sheet, so to speak. Further, the database system manages as information about the reason of storing the page cell PCi, 3 that it has been changed, and a version number difference corresponding to that

Side part PPi, 4 leads, which contains the next older version of the cell PCl, 3.

In FIG. 1, this is symbolized by a C and a 1 for the cell PCi, 3 in accordance with the above-mentioned explanation. For the copied cell PCl, 2, however, the following symbolism applies: The C and the 1 in parentheses were taken from the version of the side cell PCl, 2 serving as a copy template. The 2 outside the parenthesis represents the version number difference leading to the side part PP1, 3, which contains, as it were, a copy template version of the side cell PCi, 2.

After saving the side part PPi, 5, the fifth revision Rs, 1 of the database page Pi is completed and a fifth page version PVi, 5 is kept in the database system.

Revision R6, i: At a later time, the fifth version PVi, 5 of the database page Pi is called again so that once again modifications can be made during a sixth revision Rε, i. The access to the page version PVi, 5 takes place via the page version pointer PVRs, i managed for this purpose by the database system, as well as the two page sub-pointers PPRs, i and PPRs, 1.

Since only the page cells PCi, 2 and PCi, 3 are contained in the referenced page part PPi, 5, it follows that the cell PCi, 1 contained in the other referenced page part PPi, 4, also for the fifth page version PVi , 5 is still valid. Thus, the side version PVi, 5 of the side Pi is reconstructed and provided from the side cells PCi, 2 and PCi, 3 contained in the side part PPi, 5 and the side cell PCi, 1 contained in the side part PPi, 4.

PVi, 5 of the database page Pi from the page cells PCi, 1, PCi, 2 and PCi, 3 above all the other from the database system to the respective cell, that is, the respective node managed information considered. For each node manages the database system in addition to its own identification information c and the identification information c of the nodes, which comprise XML components that were in the XML tree hierarchically directly superordinate, subordinate or at the same level directly adjacent, respectively together with the corresponding ranking information.

Thus, taking into account these other identification information managed by the database system, c and ranking information during the reconstruction of the Page version PVi, 4 of the database page Pi from the page cells PCi, 1, PCi, 2 and PCi, 3, it is clear that the cell PCi, 1 is a node comprising an XML element, while the cell PCi, 2 a more comprehensive text content Node, and that corresponding to the cell PCi, 3

Node comprises a further XML element which, in the XML tree structure, is superior to the XML element which is included in the node corresponding to the cell PCi, 1 as the parent XML element. Only from the arrangement of the nodes within the data page this hierarchical structure would not have been apparent.

As a modification on the side Pi deletion of the cell PCi, 2 or the corresponding node takes place. That is, for the node comprising an XML element or the cell PCi, 1, the corresponding text content included in the node or cell PCi, 2 has been deleted.

The ranking between the nodes has changed only between PCi, 1 and PCi, 2 insofar as that

Relationship no longer exists. The ranking between PCi, 1 and PCi, 3 has not changed. The ranking information managed by the database should therefore be updated.

Since during the revision Rβ, i a modification on the side Pi takes place only in the form of a deletion of the cell PCi, 2, the physically stored side part PPi, 6 only needed to contain the modified version of the cell PCi, 2, the cell PCi , 2 is still stored in the emptied state.

However, there would then be no longer direct access to the page cell PCi, 1 for the page version R6, i via its page pointer PPRδ, i and PPRs, i of the page version pointer PVRδ, i, as it was still available via the second page sub-pointer PPR4, i of the version pointer PVRs, i in the side part PPi, 4, so that a copy of the side cell PCi, 1 in the side part PPi, 6 is included.

This further revision of the page is detected and, in response thereto, the revision counter r, which has hitherto had the value 5, for this memory page Pi is increased by the count value 1, ie to 6.

Logically, the result of this revision Rβ, i represents the page version PVi, 6. Physically, a side part PPi, 6 is stored, which contains only the deleted cell PCi, 2 and the copied cell PCi, 1. Accordingly, once again physically one page cell is stored less than logically belonging to the version PVi, 6 of the database page Pi.

For the logical management of the page version PVi, 6, a new version pointer PVR6, i with the two page sub-pointers PPR6, i and PPR5, i is added to the page sections PPi, 6 and PPi, 5 in the above-mentioned internal directory of the database system. In addition, the database system manages as information about the reason for storing the page cell PCi, 1 that it is a copy without being changed, and a version number difference leading to the page part PPi, 2, which serves as a copy template version of the page cell PCi , 1 contains. Further, the database system manages as information about the reason for storing the page cell PCi, 2 that it has been deleted, and a version number difference leading to the page part PPi, 3 containing the next older version of the cell PCi, 3.

In FIG. 1 this is according to the above-mentioned explanation with a D and a 3 for the cell PCi, 2 symbolizes. For the repeatedly copied cell PCi, l, the following symbolism applies: The C and the 1 in parentheses were taken from the version of the side cell PCi, i serving as a copy template. The 4 outside the parenthesis represents the version number difference leading to the side part PPi, 2, which contains, so to speak, the version of the side cell PCi, 1 serving as a copy template.

Thus, it becomes clear that for the copy of a page cell PCp, c, preferably a page sub-pointer PPRr, p or a version number difference is managed by the database system leading to the page part PPp, r, which responds to that revision Rr, p, during which the last modification of the cell or the last read access to the cell PCp, c took place.

After the storage of the side part PPi, 6, the sixth revision R6, i of the database page Pi is completed and in the database system, besides the previous 5, a sixth page version PVi, 5 is finally kept ready.

An advantageous aspect of the invention is that it no longer stores snapshots for entire memory pages, but instead stores memory page cell snapshots.

Thus, in accordance with the present invention, a memory page snapshot is decomposed into a number of memory page cell snapshots stored on the storage medium distributed over a number of m memory page portions.

The method according to the invention is therefore also referred to as "sliding snapshot".

Claims

claims

A method of maintaining a plurality of versions of memory pages in a memory system, wherein memory pages (Pp) are subdivided into memory page cells (PCp, C ₁ , ..., PCp, C _n ) and each of the memory page cells is assigned unique memory page cell identification information (c) in which, in response to a revision (R) of a particular memory page (Pp), a revision counter (r) for the respective memory page is changed by one count and logically a memory page version (PVp, r) is managed by the memory system, and physically storing a memory page portion (PPp, r) in the memory system, the memory page portion (PPp, r) including at least one memory page cell (PCp, c) modified during revision (Rr, p), and wherein in response to the revision (Rr , p) a memory page (Pp) a memory page version pointer (PVRr, p) is managed by the memory system, which m memory side part pointer (PPRr _max , p , ..., PPRr _max _ _{m + 1} , p) to the memory side parts (PPp, r _max , ..., PPp, r _max . _{m + 1} ), where m is a definable parameter value of m> 0.

The method of claim 1, wherein the memory page part (PPp, r) includes a copy of at least one memory page cell (PCp, c) that was last contained in the memory page part (PPp, r-m).

3. The method according to any one of claims 1 or 2, wherein for each memory page cell (PCp, c) a

Memory page (Pp) both their own Memory page cell identification information (c) and also the respective ones

Memory page cell identification information (c) of

Memory page cells located in the memory page (Pp) in their immediate vicinity and one each

Ranking information relating to a respective one of these

Memory page cells are managed by the memory system, and wherein for each memory page cell (PCp, c) in the

Memory page part (PPp, r) both their own memory page cell identification information (c) and the respective

Memory page cell identification information (c) of

Memory page cells that are in this

Memory page version (PVp, r) are in their immediate vicinity and each of a ranking information with respect to a respective one of these memory page cells from the

Be managed storage system.

4. The method according to any one of claims 1 to 3, wherein for each memory page cell (PCp, c) in the

Memory page part (PPp, r) information about the reason for their storage during this revision (Rr, p) is managed by the storage system.

5. Method according to claim 4, wherein the reasons for storing a memory page cell (PCp, c) within a memory page part (PPp, r) are at least one read access to the memory page cell (PCp, c) whose insertion into the memory page (Pp) Change and its deletion from the memory side and their freedom of modification since a number of m revisions (R) the memory side (Pp) are distinguishable.

6. Method according to one of claims 1 to 5, wherein for each memory page cell (PCp, c) contained in a memory page part (PPp, r) another memory page sub-pointer (PPRr, p) refers to the memory page part containing the memory page cell (PCp, c) in the next older modified version is managed by the storage system.

7. The method according to any one of claims 1 to 6, wherein the parameter value m than for the entire

Memory system and / or valid for each of a number of specific memory pages and / or dynamic as a function of the current utilization of the storage system with respect to the read and write accesses to the storage medium assertable.

The method of any one of claims 1 to 7, wherein each memory page cell (PCp, C ₁ , ..., PCp, C _n ) may contain a variable number of data bytes.

9. Method according to one of claims 1 to 8, wherein each memory page (Pp) can contain a variable or, for better handling of the memory page for the memory system, a maximum of a definable number of memory page cells (PCp, c).

A method of accessing versions of memory pages maintained in accordance with the method of any one of claims 1 to 9 in a memory system, taking into account the memory page version pointer (PVRr, p) managed by the memory system for the desired memory page version (PVp, r) and the m of this memory page sub-pointers included (PPRr _max , p, ..., PPRr _max _ _{m + 1} , p) the desired memory page version (PVp, r) from the memory page cells (PCp, C ₁ , ..., PCp, c _n ) contained in the referenced memory page parts (PPp, r _max , ..., PPp, r _max _ _{m + 1} ) and provided where, if a memory page cell (PCp, c) is included in a plurality of the m referenced memory page parts (PPP.r _max ,..., PPp, r _max _m _{+ 1} ), that memory page cell (PCp, c) is the newer one This memory side parts is used.

11. The method of claim 10, wherein reconstructing and providing the desired memory page version (PVp, r) under

Consideration of the memory system to the respective

Memory page cell (PCp, c) managed information about the reason for its storage during each

Revision (Rr, p) takes place, and / or taking into account the memory system to the respective

Memory page cell (PCp, c) managed

Memory page cell identification information (c) of the memory page cells located in the respective

Memory page version (PVp, r) in their immediate

Environment as well as the respectively administered

Ranking information relating to a respective one of these

Memory page cells.

12. Storage system, prepared for carrying out the method according to one of claims 1 to 11.

13. A computer-readable medium having a computer program stored thereon and defining instructions controlling a computer or the storage system of claim 12 such that it is suitable for performing a method according to any one of Claims 1 to 11 execute.