Method for filtering attack streams targetting a connectivity module

1. A method for filtering attack flows targeting a connectivity module receiving a plurality of connection inflows, the method comprising:

determining a plurality of aggregates, each aggregate resulting from a combination of the plurality of connection inflows received during a predefined time period, said time period considered to determine each aggregate being different from that considered to determine other aggregates, and time periods considered to determine the plurality of aggregates all being comprised in a first observation time window;

determining a plurality of first measurement vectors that are each associated with one of said aggregates and that contain values of first characteristic parameters of the one of said aggregates with which it is associated;

projecting the plurality of first measurement vectors into at least one sub-space defined by said first characteristic parameters;

determining another aggregate resulting from a combination of a plurality of connection inflows received during another time period, said another time period following the first observation time window;

determining another first measurement vector that is associated with the another aggregate and that contains values of first characteristic parameters of the another aggregate;

projecting the another first measurement vector into at least one sub-space defined by said first characteristic parameters, wherein a sub-space is a grid of cells defined by a first axis and a second axis, the first axis corresponding to a one of the first characteristic parameters and the second axis corresponding to a second one of the first characteristic parameters;

determining an abnormality score depending on the result of the projection of the another first measurement vector and of the projection of the plurality of first measurement vectors, wherein the abnormality score corresponds to a distance between (i) a cell within the at least one sub-space to which the first measurement vector is projected and (ii) one or more closest dense cells in the subspace where a proportion of the projection of the plurality of first measurement vectors is higher than a predetermined percentage; and

comparing the abnormality score to a first threshold and to a second threshold,

wherein

when the abnormality score is higher than the second threshold, the presence of an attack is detected in the another time period,

when the abnormality score is lower than the first threshold, the absence of an attack is detected in the another time period,

when the abnormality score is between the first threshold and the second threshold then the abnormality score is in a zone of doubt as to a presence of attack flows and the following steps are performed:

determining a plurality of second measurement vectors that are each associated with one of said aggregates and that contain values of other characteristic parameters of the aggregate with which it is associated, said other characteristic parameters being different from said first characteristic parameters,

determining another second measurement vector that is associated with the another aggregate and that contains values of other characteristic parameters of the another aggregate, and

detecting the presence or absence of an attack by analyzing the another second measurement vector.

2. The method for filtering attack flows as claimed in claim 1 , further comprising:

defining a second observation time window, said second observation time window corresponding to the first observation time window shifted by said other time period;

determining a new aggregate, the new aggregate resulting from a combination of a plurality of connection inflows received during a new time period, said new time period following the second observation time window;

determining a new first measurement vector that is associated with the new aggregate and that contains the values of the first characteristic parameters of said new aggregate;

deleting the projection of the first measurement vector associated with the aggregate resulting from the combination of the plurality of connection inflows received during the time period located in the first observation time window but outside the second observation time window;

projecting a new first vector into at least said one sub-space defined by said first characteristic parameters; and

determining a new abnormality score depending on a result of said projection of the new first vector into the at least said one sub-space.

3. The method for filtering attack flows as claimed in claim 1 , wherein provision is made to project the plurality of first measurement vectors into a plurality of sub-spaces, and wherein the abnormality score is determined from a sum of abnormality functions, each abnormality function of the abnormality functions being determined per sub-space depending on the result of said projection of the plurality of first measurement vectors and on the projection of the another first measurement vector and on an average abnormality, the average abnormality also being determined from said projection of the plurality of first measurement vectors and from the projection of the another first measurement vector.

4. The method for filtering attack flows as claimed in claim 3 , further comprising, when the presence of an attack is detected, the following:

comparing the abnormality functions determined for all the sub-spaces;

selecting at least one sub-space having a highest abnormality function;

separating the another aggregate into a plurality of separate connection flows;

determining a plurality of identification vectors of the plurality of separate connection flows, the identification vectors containing the same first characteristic parameters as the first measurement vectors;

projecting said plurality of identification vectors into said at least one selected sub-space;

identifying at least one illegitimate connection flow, said illegitimate connection flow corresponding to the connection flow associated with an abnormal projection of an identification vector; and

filtering the illegitimate connection flow.

5. The method for filtering attack flows as claimed in claim 1 , wherein at least one of the first characteristic parameters comprises one of following data:

a number of connection-flow sources;

an average number of connection-flow sources per sub-network;

a proportion of transmission requests;

a proportion of received error requests;

an average size of the received error requests; and

an average lifetime of the received error requests.

6. The method for filtering attack flows as claimed in claim 1 , wherein at least one of the another characteristic parameters comprises one of following data:

a hit rate associated with access to a first cache-memory level;

a hit rate associated with access to a second cache-memory level;

a miss rate associated with access to another cache-memory level; and

a proportion of memory used.

7. A device for filtering attack flows targeting a connectivity module from a plurality of connection inflows, comprising:

processing circuitry configured to

determine a plurality of aggregates, each aggregate resulting from a combination of the plurality of connection inflows received during a predefined time period, said time period considered to determine each aggregate being different from that considered to determine other aggregates, and time periods considered to determine the plurality of aggregates all being comprised in a first observation time window;

determine a plurality of first measurement vectors that are each associated with one of said aggregates and that contain values of first characteristic parameters of the one of said aggregates with which it is associated;

project the plurality of first measurement vectors into at least one sub-space defined by said first characteristic parameters, wherein a sub-space is a grid of cells defined by a first axis and a second axis, the first axis corresponding to a one of the first characteristic parameters and the second axis corresponding to a second one of the first characteristic parameters;

determine another aggregate resulting from a combination of a plurality of connection inflows received during another time period, said another time period following the first observation time window;

determine another first measurement vector that is associated with the another aggregate and that contains values of first characteristic parameters of the another aggregate;

project the another first measurement vector into at least one sub-space defined by said first characteristic parameters;

determine an abnormality score depending on the result of the projection of the another first measurement vector and of the projection of the plurality of first measurement vectors, wherein the abnormality score corresponds to a distance between (i) a cell within the at least one sub-space to which the first measurement vector is projected and (ii) one or more closest dense cells in the subspace where a proportion of the projection of the plurality of first measurement vectors is higher than a predetermined percentage; and

compare the abnormality score to a first threshold and to a second threshold,

wherein

when the abnormality score is higher than the second threshold, the presence of an attack is detected in the another time period,

when the abnormality score is lower than the first threshold, the absence of an attack is detected in the another time period,

when the abnormality score is between the first threshold and the second threshold then the abnormality score is in a zone of doubt as to a presence of attack flows and the processing circuitry is configured to:

determine a plurality of second measurement vectors that are each associated with one of said aggregates and that contain values of other characteristic parameters of the aggregate with which it is associated, said other characteristic parameters being different from said first characteristic parameters,

determine another second measurement vector that is associated with the another aggregate and that contains values of other characteristic parameters of the another aggregate, and

detect the presence or absence of an attack by analyzing the another second measurement vector.