EP3360034A1

EP3360034A1 - Dynamically distributed backup method and system

Info

Publication number: EP3360034A1
Application number: EP16778810.8A
Authority: EP
Inventors: Francis Pinault
Original assignee: ROBERTO GIORI CO Ltd
Current assignee: ROBERTO GIORI CO Ltd
Priority date: 2015-10-08
Filing date: 2016-10-07
Publication date: 2018-08-15
Also published as: US10678468B2; EP3153961A1; US20180284991A1; WO2017060495A1; BR112018006134A2; CN108139869A

Abstract

The invention relates to the field of IT, and in particular to distributed data storage across a plurality of storage servers. A distributed backup method includes the following steps: dividing the data so as to obtain data blocks; determining, for each block, a particular server from the plurality of storage servers; and memorising each block in the determined server. According to the invention, the determination of the particular server is a function of a current time instant. It can also be a function of a private key of the user. The key is used to form a mask. The latter is offset as a function of the current time instant. Then, the offset mask and the complementary mask thereof are applied, respectively, to two server distribution tables in order to identify the servers to be used for each of the data blocks. The blocks can change servers at each new time instant.

Description

METHOD AND SYSTEM FOR DYNAMICALLY DISTRIBUTED BACKUP

FIELD OF THE INVENTION

The present invention relates to the computer field, and more particularly to a system and method for storing data associated with a user in a computer network comprising a plurality of storage servers. In other words, it is distributed or distributed storage or backup of data over a network.

BACKGROUND OF THE INVENTION

The idea of using a large computer network, such as the Internet, to secure data storage is not new. In the journalistic article "Using the Internet as a distributed storage system" (http://www.lemondeinformatique.fr/ actualites / lire-utilization-internet-comme-système-de-storage-reparti-22932.html), it was already indicated that the idea, as simple as it is ambitious of a certain society, is to slice a volume of data to be archived and to distribute their storage on all the resources available on the Internet. This approach, dubbed DSG (for "dispersed storage grid") is based on an algorithm developed at MIT in the late 70s. It allows to split the data slice, and gives each of them the ability to regenerate lost segments. The reliability of the system would give it record availability: less than an hour of unavailability over a million years.

Today, the storage or backup of data on the cloud (or

"Cloud" according to the English terminology) is widely used.

However, despite many data encryption techniques, the level of security, in terms of confidentiality of data specific to a user, proposed by existing solutions may be unsatisfactory. In particular, an attacker may attempt to recover, on the storage servers used, the data blocks forming an initial secret data, without effective countermeasures being taken in particular because the recursion of access to this data may allow with the time and sophisticated spy devices to thwart the data ciphers. Document US 2007/073990 describes distributed storage of blocks forming a file on servers. A list of servers is determined from a seed associated with the file. When a server is added or removed, a new assignment of the blocks to the available servers is performed, limiting a redistribution of the blocks to only blocks concerned by the server added or removed according to this new assignment.

SUMMARY OF THE INVENTION

An object of the present invention is thus to improve the security of a personal data during its storage distributed over a plurality of servers.

For the purpose of this invention, a first aspect of the present invention relates to a method of storing data associated with a user in a computer network comprising a plurality of storage servers, the method comprising the following steps:

dividing the data to obtain a plurality of data blocks; determining, for each data block, a respective one of the plurality of storage servers; and

store each block of data at the respective storage server,

characterized in that the determination of the respective server is a function of a current time instant.

In particular, the determination of the respective server as a function of a current time instant can be carried out for each block of data, so that the storage server used to store each respective block of data varies periodically in time.

Correlatively, the invention relates, according to a second aspect, to a system (which can be integrated into a simple user terminal) for storing data associated with a user in a computer network comprising a plurality of storage servers, the system comprising at least one microprocessor configured to execute, in a system execution environment, the following steps:

dividing the data to obtain a plurality of data blocks; determining, for each data block, a respective one of the plurality of storage servers; and store each block of data at the respective storage server,

The method or system according to the invention thus makes it possible to increase the security of personal data, for example confidential data, encrypted or not, or personal programs.

This increased security is obtained by the time dependency of the storage server used to store each block of data resulting from the division of the personal data. As a result, the storage server used for storing a particular data block may vary in time, i.e., it is determined according to one or more dynamic distribution laws. The task of locating and retrieving the blocks of data is thus greatly complexified for a malicious person.

Optional features of the method according to the invention are further defined in the dependent claims. The system according to the invention may also comprise means configured to implement these optional features.

In one embodiment, a new respective storage server is determined, at each new time instant, for each data block, so as to store the data block at a new storage server at each new time instant.

This provision specifies the dependence, at the time, of the determination of the server to use.

In a particular embodiment, the method further comprises the following steps in response to a request for access to the data associated with the user:

identify storage servers storing, at a given time point, the data blocks;

retrieving the data blocks from the respective storage servers thus identified, to reform said data; and

in the event of detection of an error on the data reformed from the recovered data blocks, identifying new storage servers storing, at a following time instant (the one immediately following the given time instant), the data blocks, then retrieve the data blocks from the respective new storage servers thus identified, to reform said data.

This arrangement makes it possible to process the storage discontinuity of the data blocks during a change of time instant. Indeed, depending on whether the request for access to the data received in the vicinity of this change in time is processed more or less quickly, the data blocks may have been moved from one server to another, according to the new distributed storage scheme applicable at time T + 1.

Thus, the data is reconstituted using the scheme applicable at time T, and if this data is erroneous (lack of coherence, error on an identification criterion such as a user identity in the reconstituted data, etc.), a reconstruction is performed using the scheme applicable at time T + 1.

In another embodiment, the determination of the respective server is further dependent on a private binary key associated with the user. It can be any cryptographic key associated with the user, which is used in its binary form.

This arrangement makes it possible to encrypt the distribution scheme of the storage servers according to each user, and thus makes it more difficult for an attacker to carry out the operations to be implemented to identify the storage location of each of the data blocks.

In a particular embodiment, the step of determining the storage servers includes a step of applying the binary key as a mask to a first server distribution table to identify storage servers to be used for a portion of the data blocks. respective server distribution table associating a server with each block of data.

The knowledge of the key thus becomes indispensable for the identification of each storage server used.

According to one particular characteristic, the step of determining the storage servers further comprises a step of applying a complementary bit key as a mask to a second server distribution table to identify storage servers to be used for the other blocks. respective data. In particular, said second server distribution table can associate a server with each data block and can be formed from the same elementary table as the first server distribution table. For example, Distribution tables are generated by repeating (and concatenating) the elementary table, the second distribution table being the continuity of the first distribution table with regard to the repetition of the elementary table.

These provisions make it possible to determine which storage servers to use in a very secure manner.

According to a particular embodiment, the mask formed by the binary key is shifted relative to the first or second distribution table of the servers by a number of positions depending on the current time instant, before being applied to the first or second second server distribution table. The current time instant is thus used as disturbing reference in the application of the mask (user's binary key), increasing the security of the distributed storage of the personal data.

According to another particular embodiment, the mask is formed of a repetition (possibly partial) of the binary key so as to reach the size of the first or second distribution table servers, that is to say the number blocks of data to store.

An attacker will need to know the user's key to try to locate the servers where each block of data is stored.

According to yet another particular embodiment, the method further comprises a step of determining a server distribution basic table by duplication of which the server distribution table or tables are obtained,

method in which the step of determining the elementary table is a function of a performance index associated with each storage server and a confidence index associated with the geographical location of each storage server.

Thus, a strategy for prioritizing the use of certain servers can be implemented, for example to favor efficient servers and / or located in areas with low geographical risk (eg seismic risk or geopolitical risk).

According to one particular characteristic, the length of the elementary table is a function of the sum of weights associated with the storage servers, the weight associated with a storage server being determined from the performance and trust indices of the storage server considered. According to another particular characteristic, the step of determining the elementary table comprises the following steps:

determining, for each storage server, a frequency of repetition of an occurrence (for example via an identifier) of the storage server in the elementary table as a function of the weight associated with the said storage server;

filling the elementary table by repeating, for each iteratively considered server and according to its determined repetition frequency, a server occurrence within the elementary table until a repetition number equal to the weight associated with the server in question.

The elementary table thus makes it possible to obtain a complex and interlaced distribution of the servers, in proportions equal to their respective weights, that is to say to their confidence and performance indices. Also, such an elementary table is complex to recreate for a malicious person, while ensuring equity between the servers given their characteristics.

Note that during the formation of the elementary table, if a position

(in the elementary table) is already occupied by an occurrence of a storage server during a repeat of another storage server, we can decide to shift the occurrence of this other storage server to the next position free, then repeat the repetition from this new position. To achieve the number of repetitions / occurrences desired while the end of the elementary table is reached, it can be expected to continue the repetition by looping again on the beginning of the elementary table. For example, if the length of the elementary table equals the sum of the weights, it is necessary, to fill the whole elementary table, that each server is present in a number of occurrences equal to its own weight.

In one embodiment of the invention, the step of dividing the data comprises the following steps:

dividing the data into elementary blocks of data;

duplicate the elementary blocks into duplicate blocks;

interleaving the duplicate blocks to obtain said plurality of data blocks.

This arrangement makes it possible to introduce a redundancy of the elementary blocks forming the initial personal data, and consequently makes it possible to improve the storage reliability in the system. BRIEF PRESENTATION OF FIGURES

Other features and advantages of the invention will become apparent in the description below, illustrated by the accompanying drawings, in which:

FIG. 1 illustrates an example of a hardware architecture in which the present invention can be implemented, notably in the form of computer programs;

FIG. 2 illustrates an example of a computer network comprising a plurality of storage servers in which the invention can be implemented;

FIG. 3 illustrates, using a flow chart, the general steps of a method of distributed backup of a piece of data according to embodiments of the invention;

FIG. 4 illustrates, in the form of a flow chart, steps for the determination of storage servers of the method of FIG. 3;

FIG. 5 illustrates an example of implementation of the steps of FIG. 4;

FIG. 6 illustrates, in the form of a flow chart, steps for determining an elementary table of the method of FIG. 3;

FIG. 7 illustrates an example of implementation of the steps of FIG. 6; and

FIG. 8 illustrates, in the form of a flow chart, an example of general steps of a method of accessing a data item saved according to the method of FIG. 3.

DETAILED DESCRIPTION OF EMBODIMENTS OF THE INVENTION FIG. 1 illustrates an example of a hardware architecture in which the present invention may be implemented, notably in the form of computer programs. For example, this hardware architecture can be part of a device or user terminal, such as an onboard computer or not, laptop, mobile terminal, a mobile tablet, or from a server offering services. distributed backup of data and access to this data.

The hardware architecture 10 comprises in particular a communication bus 100 to which are connected:

a processing unit 1 10, denoted CPU (for Central Processing Unit), which can comprise one or more processors; at least one non-volatile memory 120, for example ROM (for Read Only Memory), EEPROM (for Electrically Erasable Read Only Memory) or Flash, for storing computer programs for the implementation of the invention and possible parameters used for this one;

a random access memory 130 or cache memory or volatile memory, for example RAM (for Random Access Memory), configured to store the executable code of methods according to embodiments of the invention, and to store registers adapted to memorize, at least temporarily, variables and parameters necessary for the implementation of the invention according to embodiments;

an interface 140 of input / output I / O (for Input / Outpuf), for example a screen, a keyboard, a mouse or other pointing device such as a touch screen or a remote control enabling a user to interact with the system via a graphical interface; and

a communication interface COM 150 adapted to exchange data for example with storage servers via a computer or communication network.

The instruction codes of the program stored in non-volatile memory 120 are loaded into RAM 130 for execution by the processing unit CPU 1 10.

The non-volatile memory 120 also stores confidential information of the user, for example a private key in binary form. Of course, in order to improve the protection of such a private key, it can be stored in a secure element (SE for Secure Element) type smart card, equipping the system according to this hardware architecture or on an HSM ( Hardware Security Module).

The present invention is part of the backup (or storage) distributed data on storage servers of a communication network, typically an extended computer network, such as the Internet.

FIG. 2 illustrates an example of a computer network 20 comprising a plurality of M storage servers S _x . In the nonlimiting example of the figure, four (M = 4) storage servers Si, S ₂ , S ₃ and S ₄ are represented. The servers are synchronized on the same reference clock.

A user terminal 21, presenting the hardware architecture of FIG. 1, enables a user to request the safeguarding of personal data, sometimes confidential, encrypted or not, and access this personal data once it stored distributed in the network 20.

The user terminal 21 can implement the present invention to manage the distributed storage of such personal data and its subsequent access. Alternatively, the user terminal 21 can access a distributed data backup service and subsequent access to this data, proposed by a server S of the network 20. In both cases, all the parameters (indices of confidence and performance, user keys, etc.) discussed subsequently can be stored on such a server S, and be retrieved, if necessary, by the user terminals.

The general principles of distributed data backup include dividing the data to obtain a plurality of data blocks; determining, for each data block, a respective one of the plurality of storage servers; and storing each block of data at the respective storage server.

In this context, the present invention provides for increasing the protection, and therefore the security, of the data thus stored according to these solutions, by making a determination of each respective server according to a current time instant, that is to say say according to time.

The result is a location of each block of data that can vary over time, making recovery by an attacker more difficult.

FIG. 3 illustrates, using a flow chart, general steps of an exemplary method according to embodiments of the invention. These steps are implemented in a system according to the invention, which may be the user terminal 21 or the server S of FIG.

In step 30, a request for storing a personal data DATA is received from the user (via the user terminal 21 if necessary).

This data is personal in that it is attached to a user or group of users. It may consist of a plurality of elementary data, for example confidential. Typically, the personal data is encrypted.

The personal data DATA forms a file of size LENGTH.

In step 31, the DATA data is divided to obtain a plurality of data blocks. This step is broken down into three sub-steps: dividing the data DATA into elementary blocks of data DD _N ; duplicate the elementary blocks into duplicate DVD blocks to provide a sufficient level of redundancy; and interleaving the duplicate blocks to improve the reliability of the storage mechanism. The data DATA can be divided into Nb blocks of constant size Lfix, applicable to any data DATA in time: Nb = Γ LENGTH / Lfixe 1, where Γ Ί is the function returning the whole part by excess.

As a variant, the data DATA can be divided into a plurality of blocks of the same size Lvar, this block size being variable in that it can depend on one or more parameters, for example chosen from the following parameters: the size LENGTH of the DATA data, the user, the operator of the distributed backup and data access service, etc. The number of blocks obtained is then:

Nb = T LENGTH / Lvar 1.

The user of variable lengths further improves the security of the data to be backed up.

By way of example, the variability of the size of the block as a function of the LENGTH size of the data DATA can follow one of the following formulas:

Lvar = TLmin + (LENGTH / Nbmax)],

where Nbmax is a predefined parameter and Lmin is a predefined minimum integer size. In this case, the number Nb of data blocks obtained tends to Nbmax plus the data DATA is large;

Lvar = V (LENGTH) for LENGTH <Nbmax ²

Lvar = LENGTH / Nbmax for LENGTH≥ Nbmax ² ,

in which case the number Nb of data blocks obtained is min (rV (LENGTH) 1, Nbmax), and therefore tends to Nbmax plus the DATA data is large.

For example, the variability of the block size according to the user may consist of using a unique identifier ID of the user (for example a social security number, a passport number or identity card, etc.) that one normalizes in a predefined interval [0; Nbmax], to calculate this length:

Nb = ID - Nbmax.LlD / NbmaxJ, where L J is the function returning the integer part by default, and thus:

Lvar = TLENGTH / Nbl = rLENGTH / (ID-Nbmax. | _ID / Nbmax_ |) 1.

Alternatively, an integer of the interval [0; Nbmax] can be randomly assigned to each user and used to define the value Nb. The variable size then flows directly: Lvar = TLENGTH / Nbl.

For example, the variability of the size of the block according to an operator may consist in providing different levels (value Nb) of dividing the data according to subscription options or performance. The use of a division into one large number of blocks is safer, but requires more calculations as described later. Also, such a level of division can be reserved for premium subscribers.

Of course, these various examples can be combined with one another to produce variable lengths of division of the data DATA into data blocks DrDMt.

Optionally, these blocks of data can be duplicated to provide data redundancy, making the reconstitution of DATA data more reliable. In one embodiment, the redundancy law can be fixed, defining the number of duplications RD by a fixed number, for example 3.

According to another embodiment, the applied redundancy law may be variable in that it may depend on one or more parameters, for example according to CS confidence indices, assigned to the M servers S ,. For example, the whole number of duplications can be:

RD = RDmax + - L (ΣCSi) / M + 1 J

with RDmin <RD≤RDmax # RDmin and RDmax being two predefined values; and moy () being the function that returns the median or mean value.

Given the RD number of duplications, Nb '= RD.Nb D'-iD blocks are obtained from n elementary blocks DD _N.

Optionally also, the m DVD blocks can be interleaved to improve the reliability of the backup system with regard to errors occurring in the processing of DVD blocks.

For example, the interleaving of the produced DVD data blocks may be monotone of depth P, meaning that each group of P elementary blocks D, is duplicated RD times. For example, for RD = 3 and P = 4, we consider each group of 4 blocks successively as follows:

D2D3D4 D1 D1 D1 D2D3D4 D2D3D4 D ₅ D ₆ D ₇ D ₈ D ₅ D ₆ D ₇ D ₈ D ₅ D ₆ ...

As a variant, a complex interweaving of depth P can be implemented, meaning that for each group of P elementary blocks, their duplications are mixed. For example for RD = 3 and P = 6:

D1 D4D2D5D3D6 D5D2D4D1 D6D3 D6D1 D3D4D2D5 D ₈ D ₁₀ D ₇ D ₁₁ D ₉ D ₁ 2 ... Following the step 31, a basic table of distribution servers S ,, denoted TABLE _E, is obtained in step 32 .

The elementary table TABLE _E consists of an ordered plurality of L _TAB LE entries, each identifying one of the servers S ,. This table elementary TABLE _E can be a predefined table recovered in non-volatile memory of the system. As a variant, it can be determined according to the method of FIG. 6 described below in order, in particular, to favor trusted and / or efficient servers.

An example of a basic table of length is reproduced here for the purposes of illustration, where only the index i of the server S is reported, when M = 4:

Following step 32, a private key of the user is obtained in step 33. It is preferably a cryptographic key obtained from an elliptic curve. This key, noted K, is securely stored in the system embodying the invention, for example using a secure element type smart card.

As explained below, the private key K is used to determine the servers to be used to store each D 'block.

Then in step 34, a respective storage server is determined, for each data block D ', among the plurality of storage servers, according to a current time instant Tu. The current time instant is defined with an accuracy directly dependent on a chosen time unit.

For example, the time instant can be defined by the current time if a time unit of the order of the hour is chosen. In this case, the day is divided into 24 successive instants, identified by their respective hours Tu = 0 to 23. Since the storage according to the invention depends on the time, such a temporal unit makes it possible to modify the storage location of the blocks D '. , twenty-four times a day.

Alternatively, a time unit of the agenda can be used, so as to modify the storage location of the blocks D ', thirty or thirty-one times a month.

These proposed time units have the advantage of being very long compared to the processing time of steps 31 to 35 for determining new locations for storing data blocks. Indeed, a ratio greater than 1000 (such processing by computer means generally taking less than a few seconds) is thus obtained, making it possible to reduce the risks of ambiguity relating to the passage of a transition from one time instant to the next, when receiving a request to access the data DATA.

However, mechanisms for managing these risks can be implemented as described later with reference to Figure 8. Step 34, an embodiment of which is described in greater detail hereinafter with reference to FIG. 4, thus makes it possible to identify a storage server for each data block D ', resulting from the division of the initial data. DATA, and this according to the current time instant Tu.

It follows, in step 35, the proper storage of each block of data at the respective storage server and determined. Conventional secure communication techniques with the storage servers S are preferably implemented.

The method continues in step 36 where the system waits for the next time instant, for example the beginning of the next hour or the next day. When a new time instant is reached, steps 31 to 35 are reiterated to determine a new respective storage server for each data block D'i, and thus store the data block at the new storage server during this new time. time instant. Preferably, the data blocks are erased from the old storage servers on which they were stored during the old time instant just ended.

It can thus be seen that the distributed backup of the data DATA by blocks evolves dynamically, making the task of locating the data blocks D 'difficult for a malicious person.

Note that the new execution of steps 31, 32 and 33 may simply consist in recovering the result of a previous execution of these steps, when these do not involve the current time instant as a parameter (for example the elementary table distribution can change over time).

Step 35 is itself dependent on the current time instant, ensuring that the storage servers identified for each block of data to backup evolve over time.

FIG. 4 illustrates an embodiment of the step 34 of determining the storage servers for saving the data blocks D 'at the current time instant Tu. This determination takes into account, besides the current time instant Tu, the private key K, the elementary distribution table TABLE _E and the size LENGTH of the data DATA to be saved.

A first step 40 consists in obtaining a first distribution table TABLE1 from the elementary table TABLE _E , by duplication of the latter in order to obtain a table TABLE1 of length equal to Nb '(that is to say a table TABLE1 of the same length as the number of data blocks D ', to be saved). FIG. 5 illustrates an example of a TABLE _E elementary table and of the first TABLE1 distribution table thus obtained, for M = 4 (four servers), with 41 D blocks.

Then, the next step 41 is to obtain a MASK bit mask from the private key K and the current time instant Tu. Since this mask MASK will be applied to the first distribution table TABLE1, it has the same size Nb 'as this one.

In the example of Figure 5, the private key K is used in its binary form (continuation of and Ό '), here a 32-bit key. Then, the mask MASK is formed by the repetition of the binary key K, until reaching the size Nb 'of the first server distribution table. In the figure, the nine bolded bits are from a K key repetition.

Then, in step 42, the mask MASK is applied to the first distribution table TABLE1 servers to identify storage servers to use for a portion of the respective data blocks D ',. According to embodiments, it is at this stage that the current time instant Tu is taken into account to disturb the identification of the storage servers to be used.

In particular, it may be provided to shift the mask MASK relative to the beginning of the first distribution table servers TABLE1 a number of positions according to the current time instant Tu, before being applied to this server distribution table .

As shown in FIG. 5, the mask MASK is shifted from Tu positions before application to the TABLE1 (offset indicated by K "Tu); and the result RESULT1 of this masking operation (those of the mask identify the servers of the table TABLE1 to keep) identifies only a portion of the storage servers to use.

In step 43, a second server distribution table is obtained

TABLE2 of size Nb 'from the table elementary TABLE _E by duplication of the latter. In order to obtain a table TABLE2 different from the first table TABLE1, the second distribution table can simply be the continuity of the first distribution table with regard to the repetition of the elementary table, as illustrated in FIG.

5.

Then, in step 44, a second MASK2 mask formed, for example, of the binary (bit-to-bit) complement of the first mask MASK is obtained. The second mask also has a size equal to Nb '. In step 45, the second mask MASK2 is applied to the second distribution table TABLE2 in the same way as in step 42, so as to identify the storage servers to be used for the other data blocks D ', (those for which step 42 could not identify such servers). Indeed, the use of the complementary of the first mask ensures that ultimately each of the blocks D ', is associated with a respective storage server.

Figure 5 identifies the result of this operation by the reference

RESULT2.

Of course, other approaches can be implemented such as the use of other masks generated from the private key K and the instant Tu, and the repetition of masking operations as the set of blocks D data has not been assigned respective storage servers.

The process of FIG. 4 ends at step 46 by merging the results RESULT1 and RESULT2 of the masking operations, so as to obtain a RESULT grid for locating the Nb 'storage servers.

This grid thus identifies the storage server S to be used for each of the Nb 'data blocks D'.

We will now describe, with reference to FIG. 6, an implementation of the step 32 for determining the distribution table of the TABLE _E servers by duplication of which the distribution tables of the servers TABLE1 and TABLE2 are obtained.

In this method, the step of determining the elementary table is a function of a performance index associated with each storage server and a confidence index associated with the geographical location of each storage server.

Also, we assume a set of properties attached to the servers S, of Figure 2.

Each server S, is associated with a geographical location Two servers can have the same location LS _j , hence N <M.

A confidence index CS, is associated with each location LS _j . This confidence index is representative of a local stability with regard to the accessibility of servers located there. For example, this confidence index can be established as in patent EP 2 433 216, for example on a scale of 0 (low confidence) to CS _ma x = 10 (strong confidence), taking into account seismic risks, flooding , geopolitical, etc. for the location considered. Of course, other ranges of values are possible.

Each storage server S, is thus indirectly associated with a CS index of confidence, linked to its geographical location.

In addition, a performance index PS is associated with each storage server S, depending for example on the performance of access to this server and / or the performance of the server itself.

There are a large number of techniques to scale server performance, including storage performance, memory performance, processor performance, network performance, and process performance. Also they will not be described in detail here.

The PS performance index is preferably established on a scale ranging from 0 (poor performance) to PS _ma x = 10 (high performance). Of course, other values are possible.

Note that the PS performance index can vary over time:

PSi = f (Tu), in which case for example step 32 is re-executed entirely at each new time instant (after step 36).

As illustrated in FIG. 6, the step of determining the elementary distribution table of TABLE _E servers begins with a step 60 of obtaining a weight WS, associated with each storage server S 1. This weight can in particular be representative of the associated confidence and performance. Also, the weight WS, associated with a storage server S, can be determined from the performance and confidence indices of the storage server in question, for example by combining the indices CS, and PS, associated with S ,.

For example: WS = (CSi.PSi) / (CS _max .PS _max) for a weight ranging from 0 to 1.

As a variant, to obtain a weight varying between 0 and CS _max or between 0 and PS _max , one of the following formulas can be used:

WS; = (CSi.PSi) / PS _max,

WS; = (CSi.PSi) / CS _max .

In addition, if CS _my x = PS _my x, the following formula can be used to set an average value between performance and trust of storage servers:

Having the weights WS ,, step 61 consists in determining the LTABLE length of the elementary table according to the sum of weights associated with the servers storage. For example, L _T ABLE = Σi = i..M (WSi). If WS is between 0 and 10 (CSmax = Smax = 10), then the table length is at most 10.M.

Then, in step 62 an index 'x' is initialized to 1. This index is used in the loop of the algorithm described after, to treat all the storage servers: x = 1 ... M.

Then in step 63, a repetition frequency F _x is determined for each storage server according to the weight WS _X associated with the storage server S _x considered. As used subsequently, F _x is representative of a frequency of occurrence of the storage server S _x in the elementary distribution table TABLE _E , when it will be necessary to create it.

Thus, the higher the weight WS _X is high (trusted server and / or performance), the higher the repetition frequency can be chosen to favor trusted servers and / or performance.

For example, 1 / F _X = L (L _T ABLE / WS _X ) _ |. In other words, it is envisaged to repeat the server S, all L (L _T ABLE / WS _X ) J positions within the TABLET table.

Then in the following steps, TABLE table _E is formed by repeating WS _x times each server S _x with a frequency F _x .

For example, in step 64, the padding of TABLE _E for the server S _x is initialized by TABLE _E (x) = x. The first position in the elementary table TABLE _E therefore informs the server Si.

If this entry of the table is already used, we take the first available entry following TABLE _E (x).

The position of the input entered is stored in a variable 'z'. In addition, a counter of the number of NBOCC occurrences is initialized to 1.

Then it is checked in step 65 if all the occurrences of the server S _x have been added to the TABLE table _E : "NBOCC = WS _X ? ".

If not, then step 66 provides for populating the next occurrence of the server S _x in the TABLE _E.

To do this, the position of the next entry is determined:

z <- (z + 1 / F _X ) mod (L _T ABLE).

If the corresponding entry TABLE _E (z) is already filled in, again we take the first available next entry (by looping back to the beginning of the table if necessary), in which case we memorize its index in the variable z. Then, the entry TABLE _E (z) is filled in to indicate the storage server S _x : TABLE _E (z) = x, and the variable NBOCC is incremented.

The method then loops back to step 65 making it possible to fill all the occurrences of the server S _x in the elementary table TABLE _E.

When all these occurrences have been filled in (exit 'yes' from the test

65), step 67 determines if all the servers have been processed: "x = M? In which case the process of Figure 6 is terminated. If not, the next storage server is considered by incrementing index x (step 68) before looping back to step 63.

Given the definition of L _TAB LE, all the entries in TABLE _E finally inform a storage server.

Figure 7 illustrates the process of Figure 6 for M = 4 servers, with the following weights WSi = 4, WS ₂ = 3, WS ₃ = 8 and WS ₄ = 6. The elementary table TABLE _E is filled by repeating, for each server S, iteratively considered and according to its determined repetition frequency (F,), an occurrence of the server within the elementary table until reaching a number of repetition NBOCC equal to the weight WS, associated with the considered server.

Step 61 makes it possible to obtain L _TA BLE = 4 + 3 + 8 + 6 = 21

The first loop (x = 1) of steps 63 to 66 makes it possible to obtain

1 / Fi = L (L TABLE / WSi) J = L21 / 4j = 5

Then have TABLE _E (1) = 1, TABLE _E (1 + 5 = 6) = 1, TABLE _E (6 + 5 = 1 1) = 1 and

TABLE _E (1 1 + 5 = 16) = 1. As at this stage NBOCC = 4 = WSi, no other occurrence of the server Si is added to the elementary table of distribution TABLE _E , and in particular at the level of the entry TABLE _E (16 + 5 = 21) identified, on the figure, by the symbol '♦'.

In the second loop (x = 2), 1 / F ₂ = L21 / 3J = 7, then TABLE _E (2) = 2, TABLE _E (2 + 7 = 9) = 2. As the entry TABLE _E (16) is already filled in (for the server Si), this entry is passed (chip '·' in the figure) and the next available input is chosen TABLE _E (17) = 3. At this point NBOCC reaches WS ₂ = 3, which ends the loop for the server S ₂ . We see here that there are fewer occurrences of the server S ₂ compared to the server Si, due to the fact that the latter has a higher weight (4 against 3).

In the third loop (x = 3), 1 / F ₃ = L21 / 8J = 2, then TABLE _E (3) = 3,

TABLE _E (5) = 3, TABLE _E (7) = 3. As the entry TABLE _E (9) is already filled in (for the server S ₂ ), the following available input is chosen TABLE _E (10) = 3. Then the occurrences according to 1 / F ₃ are repeated: TABLE _E (12) = 3, TABLE _E (14) = 3. As the entries TABLE _E (16) and TABLE _E (17) are already filled in, the following available entry is chosen TABLE _E (18) = 3. The last occurrence to reach NBOCC = WS ₂ = 3 is given: TABLE _E (20) = 3.

Finally, during the fourth loop (x = 4), 1 / F ₄ = | _21 / 6j = 3, then TABLE _E (4) = 4, TABLE _E (8) = 4 (since TABLE _E (7) is already filled in), TABLE _E (13) = 4 (since TABLE _E (1 1) and TABLE _E (12) are already filled in), TABLE _E (19) = 4 (since TABLE _E (16) to TABLE _E (18) are already filled in). Since z + 1 / F ₄ = 22 is greater than A _BLE , we loop back to the beginning of the elementary table TABLE _E where we find the first available input TABLE _E (15) to inform the server S ₄ . Finally, the last occurrence of the server S ₄ is indicated in the last available entry TABLE _E (21).

This gives the completely filled TABLEE elementary table _E , which can be used in step 32 described above.

The method of accessing a data item saved according to the algorithm of FIG. 3 is now described with reference to FIG. 8. As mentioned above, this algorithm includes a mechanism for managing the risks of ambiguity relating to the passage of a message. transition from one time instant to the next, upon receipt of a request to access the data DATA.

The algorithm starts at step 80 by receiving a request to access the data DATA by a user U. If necessary, the division, redundancy and interleaving mechanisms (step 31) of the data DATA are set to particularly for the purpose of knowing the number Nb 'of data blocks D' to be recovered.

A variable Ίοορ 'is initialized to 0, to serve as a mechanism for managing temporal transitions.

The temporal instant Tu of reception of the request is memorized. The following steps identify the storage servers storing, at this time, the data blocks forming the data DATA to access.

In particular, in step 81, the elementary table TABLE _E is obtained in a manner similar to step 32. Then in step 82, the private key K of the user is obtained in a manner similar to step 33. Then at step 83, the data block storage servers D 'are determined similarly to step 34, for the time slot Tu.

In step 84, the data blocks D 'are recovered from these determined storage servers by conventional mechanisms (for example, requests secure). Then, during step 85, the data DATA is reformed from the blocks D ', thus recovered.

The next step, 86, is to check the consistency of the result of step 85. Several elements can be checked to identify a possible error. For example, the verification may relate to the identification of the user U which must be identical to that indicated in the data reformed DATA (for example if the DATA data is encrypted, the use of a public key of the user U can verify the authenticity). According to another example, checksum verification may be carried out (for example if the end of the data DATA consists of a checksum of the rest of the data). Other checks can be carried out such as the dating of the last storage stored compared to a traceability stored operations performed for this user.

In case of inconsistency or error found on the data reformed from the recovered data blocks (test 87), the process continues to test 88 to check if we have just tested the data at the time instant Tu (loop = 0) or at the next time instant Tu + 1 (loop = 1). If loop = 0, the time instant Tu: Tu -Tu + 1 is incremented in step 89 and loops back to step 83 so as to identify new storage servers storing, at the next time instant (that immediately following the time of reception of the access request), the blocks of data, then to recover (step 84) the data blocks from the respective new storage servers thus identified, to reform (step 85) said data .

If loop = 1 (test 88), then an error message is returned to the user in response to his request (step 90).

In the absence of error in the test 87, the reformed data is returned to the user in response to his request (step 91).

It can be seen that if the data DATA can not be correctly reconstructed using the distribution scheme of the blocks D ', at the time instant Tu, it is reconstituted using the distribution scheme valid for the following time instant. Tu + 1. Thus, even if the access request is received close to a time transition by which the data blocks D 'are moved from servers, the method makes it possible to retrieve said DATA data in a secure manner.

The embodiments of the invention which have just been described make it possible to determine, in a wide area network, distributed backup virtual locations according to one or more dynamic distribution laws. This approach offers a high level of security of data saved in a way distributed. Various mechanisms make it possible to improve this security, such as the redundancy of data blocks, the use of the user's identity to vary certain parameters.

The foregoing examples are only embodiments of the invention which is not limited thereto.

Claims

A method for storing a data (DATA) associated with a user in a computer network (20) comprising a plurality of storage servers (S,), the method comprising the steps of:

dividing (31) the data to obtain a plurality of data blocks (D ,, determine (34), for each data block, a respective one of the plurality of storage servers, and

storing (35) each block of data at the respective storage server,

characterized in that the determination, for each data block, of the respective server is a function of a current time instant (Tu), so that the storage server used for storing each respective block of data varies periodically in time.

The method according to claim 1, wherein a new respective storage server is determined, at each new time instant, for each data block dividing said data, so as to store the data block at a new data server. storage at each new time instant.

The method of claim 2, further comprising the following steps in response to a request for access to the data associated with the user:

identifying (83) storage servers storing, at a given time point (Tu), the data blocks;

retrieving (84) the data blocks (D ,, D ',) from the respective storage servers so identified, to reform said data (DATA); and

if an error on the data reformed from the retrieved data blocks is detected, identify new storage servers storing the data blocks at a later time, and then retrieve the data blocks from the new data servers. respective storage thus identified to reform said data.

4. Method according to one of claims 1 to 3, wherein the determination of the respective server is further function of a private bit key (K) associated with the user.

The method of claim 4, wherein the step of determining the storage servers includes a step (42) of applying the binary key as a mask (MASK) to a first server distribution table (TABLE1) to identify storage servers to be used for a portion of the respective data blocks, said first server distribution table associating a server with each block of data.

The method of claim 5, wherein the step of determining the storage servers further comprises a step (45) of applying a complementary bit key as a mask (MASK2) to a second server distribution table ( TABLE2) for identifying storage servers to be used for the respective other data blocks, said second server distribution table associating a server with each data block and being formed from the same elementary table (TABLE _E ) as the first server distribution table.

7. The method of claim 5 or 6, wherein the mask formed by the binary key is shifted relative to the first or second distribution table servers a number of positions depending on the current time instant, before being applied to the first or second server distribution table.

8. Method according to one of claims 5 to 7, wherein the mask is formed of a repetition of the binary key so as to reach the size (Nb ') of the first or second server distribution table.

9. Method according to one of claims 5 to 8, further comprising a step of determining (32) a basic server distribution table (TABLE _E ) by duplication of which the server distribution table or tables are obtained

method in which the step of determining the elementary table is a function of a performance index (PS) associated with each storage server and a confidence index (CS) associated with the geographical location (LS) each storage server.

10. The method of claim 9, wherein the length (L _TAB LE) of the elementary table is a function of the sum of weight (WS) associated with the storage servers, the weight associated with a storage server being determined from performance and trust indices of the storage server considered.

1 1. The method of claim 9 or 10, wherein the step of determining the elementary table comprises the steps of: determining (63), for each storage server, a frequency (F,) of repetition of a storage server occurrence in the elementary table according to the weight associated with said storage server in question;

filling (64, 66) the elementary table by repeating, for each iteratively considered server and according to its determined repetition frequency, a server occurrence within the elementary table until reaching (65) a repetition number (NBOCC ) equal to the weight associated with the server in question.

12. Method according to one of claims 1 to 1 1, wherein the step of dividing the data comprises the following steps:

dividing the data (DATA) into elementary blocks of data (D,); duplicate the elementary blocks into duplicate blocks;

interleaving the duplicated blocks so as to obtain said plurality of data blocks (D ',).

13. A data storage system (DATA) associated with a user in a computer network (20) comprising a plurality of storage servers

(S,), the system comprising at least one microprocessor (1 10) configured to execute, in a system execution environment, the following steps:

dividing the data to obtain a plurality of data blocks (D ,, D ',); determining, for each data block, a respective one of the plurality of storage servers; and

store each block of data at the respective storage server,

The system of claim 13, wherein the microprocessor is further configured to determine a new respective storage server at each new time instant and for each data block dividing said data, so as to store the data block at the same time. a new storage server at each new time instant.

15. The system of claim 13 or 14, wherein the determination of the respective server is further function of a private bit key (K) associated with the user.