Method and device for determining at least one machine involved in an anomaly detected in a complex computing infrastructure
A method for determining at least one machine involved in an anomaly detected in a computing infrastructure comprising a plurality M of machines. This determination is performed by associating with each machine a state vector constructed using metrics and logs. Each vector is also associated with a time window. The monitoring of these vectors makes it possible to detect an anomaly that is then explained via an explanation method that makes it possible to assign an involvement indicator to each machine. The use of this indicator makes it possible to determine the machine most involved.
1. A method for determining at least one machine involved in an anomaly detected in a computing infrastructure comprising a plurality M of machines, wherein M is an integer strictly greater than 1, wherein said method comprises:
configuration of a state vector structure of dimension N of a machine, N being an integer strictly greater than 2, said state vector structure distributing metric-type information and log-type information in N dimensions; determination of a time window duration; acquisition of timestamped operating data according to the configuration of the state vector structure and the time window duration; construction of a state vector, for each of the plurality M of machines from the timestamped operating data that is acquired, according to the state vector structure configured, said construction comprising
digitization of the timestamped operating data that is acquired according to the configuration of the state vector structure; normalization of the timestamped operating data that is digitized; aggregation of data by dimension;
assembly of all of each of the state vector of the plurality M of machines of N dimension into an infrastructure vector; detection of an anomaly by applying an anomaly detection method to the infrastructure vector; for the anomaly detected, the method further comprises:
application of an explanation method to the infrastructure vector producing an infrastructure explanation vector comprising M machine explanation vectors of N dimension; separation of the infrastructure explanation vector per machine; for each machine, aggregation of components of the infrastructure explanation vector to obtain an involvement indicator; determination of the machine having a highest indicator, said machine that is determined then being considered as involved in the anomaly.
2. The method for determining at least one machine involved in an anomaly detected in a computing infrastructure according to claim 1 , wherein the timestamped operating data acquired comprises metrics and lines from at least one activity log file.
3. The method for determining at least one machine involved in an anomaly detected in a computing infrastructure according to claim 1 , wherein the log-type information is digitized using a method selected from: Word2VEC, BERT, GloVe.
4. The method for determining at least one machine involved in an anomaly detected in a computing infrastructure according to claim 1 , wherein the timestamped operating data that is digitized is aggregated using on said data a method selected from: mean, quartile, minimum, maximum.
5. The method for determining at least one machine involved in an anomaly detected in a computing infrastructure according to claim 1 , wherein the time window duration is between one minute and one hour.
6. The method for determining at least one machine involved in an anomaly detected in a computing infrastructure according to claim 1 , wherein the time window duration is between one minute and ten minutes.
7. The method for determining at least one machine involved in an anomaly detected in a computing infrastructure according to claim 1 , wherein a state vector has a dimension N between fifty and one hundred.
8. The method for determining at least one machine involved in an anomaly detected in a computing infrastructure according to claim 1 , wherein a state vector has a dimension N such that a product NxM of the dimension multiplied by a number of machines does not exceed one million.
9. The method for determining at least one machine involved in an anomaly detected in a computing infrastructure according to claim 1 , wherein the anomaly detection method, applied to the infrastructure vector, is selected from one of: autoencoder and elliptic envelope, isolation forest, OCSVM.
10. The method for determining at least one machine involved in an anomaly detected in a computing infrastructure according to claim 1 , wherein the explanation method, applied to the infrastructure vector, is selected from: Shap, Lime, neural network attention layer.
11. The method for determining at least one machine involved in an anomaly detected in a computing infrastructure according to claim 1 , wherein the aggregation of the components of the infrastructure explanation vector of each machine is achieved using, on said components, a method selected from: mean, quartile, minimum, maximum.