Root cause analysis for protection storage devices using causal graphs
View Patent ↗Embodiments for finding a root cause of an anomaly in a network environment by representing assets in the network environment as respective nodes in a causal graph, wherein the nodes have a measurable quality that can be tracked and arcs between pairs of nodes represent causal relationships between nodes of the node pairs designating source nodes as processes at the top of a hierarchy of tracked processes, and sink nodes as processes at the bottom of the hierarchy and having characteristics of interest in the environment; detecting anomalies in the tracked processes embodied in the sink nodes; traversing the causal graph in a reverse order from a node in which an outlier is detected; and analyzing nodes along the traversal path to identify a node of the highest hierarchy that shows unusual behavior as the root cause.
1. A method of finding a root cause of an anomaly in a network environment, comprising:
representing assets in the network environment as respective nodes in a causal graph, wherein the nodes have a measurable quality that can be tracked and arcs between pairs of nodes represent causal relationships between nodes of the node pairs;
designating source nodes as processes at the top of a hierarchy of tracked processes, and sink nodes as processes at the bottom of the hierarchy and having characteristics of interest in the environment;
detecting anomalies in the tracked processes embodied in the sink nodes;
traversing the causal graph in a reverse order from a node in which an outlier is detected; and
analyzing nodes along the traversal path to identify a node of the highest hierarchy that shows unusual behavior as the root cause in a prioritized list of possible root causes.
2. The method of claim 1 further comprising defining a time series to track changes of the tracked processes across time in a defined time scale.
3. The method of claim 2 further comprising displaying all anomalies detected in the environment through a graphical user interface against the time series scale.
4. The method of claim 3 wherein the time series comprises log data generated for data transactions of the nodes.
5. The method of claim 4 further comprising transforming the log data into time-series data using aggregations.
6. The method of claim 2 wherein the detecting step comprises finding anomalies in the time series using one of a smoother process or a forecasting process.
7. The method of claim 6 wherein the outlier comprises a detection on an edge of the time series.
8. The method of claim 7 wherein the detecting step comprises a forecasting process utilizing different forecasting models.
9. The method of claim 8 further comprising:
comparing results between forecasting models of the different forecasting models based on a test data set;
selecting a model with a minimal error as a best forecasting model;
using the best forecasting model to determine a difference between a predicted outlier value and an actual outlier value to determine a residual value; and
declaring the outlier to be an anomaly if the residual exceeds a defined threshold value by a defined margin.
10. The method of claim 9 further comprising assigning weights to anomalies based on a significance of respective residuals for a weighted calculation.
11. A system of finding a root cause of an anomaly in a network environment, comprising:
a graph generator representing assets in the network environment as respective nodes in a causal graph, wherein the nodes have a measurable quality that can be tracked and arcs between pairs of nodes represent causal relationships between nodes of the node pairs, wherein source nodes are designated as processes at the top of a hierarchy of tracked processes, and sink nodes are designated as processes at the bottom of the hierarchy and having characteristics of interest in the environment;
a detector detecting anomalies in the tracked processes embodied in the sink nodes; and
a root cause analyzer traversing the causal graph in a reverse order from a node in which an outlier is detected, and analyzing nodes along the traversal path to identify a node of the highest hierarchy that shows unusual behavior as the root cause in a prioritized list of possible root causes.
12. The system of claim 11 further comprising a time series defined to track changes of the tracked processes across time in a defined time scale, and a graphical user interface displaying all anomalies detected in the environment against the time series scale.
13. The system of claim 12 wherein the time series comprises log data generated for data transactions of the nodes, and wherein the log data is transformed into time-series data using aggregations.
14. The system of claim 12 wherein the detector finds anomalies in the time series using one of a smoother process or a forecasting process using different forecasting models, and wherein the outlier comprises a detection on an edge of the time series.
15. The system of claim 14 wherein the detector compares results between forecasting models of the different forecasting models based on a test data set; selects a model with a minimal error as a best forecasting model; uses the best forecasting model to determine a difference between a predicted outlier value and an actual outlier value to determine a residual value; and declares the outlier to be an anomaly if the residual exceeds a defined threshold value by a defined margin.
16. The system of claim 15 wherein the detector further assigns weights to anomalies based on a significance of respective residuals for a weighted calculation.
17. The system of claim 11 wherein the network environment comprises a data protection network having a backup server and data storage devices storing data backed up from data sources by the backup server.
18. The system of claim 17 wherein the anomalies comprise significant changes in data backup performance through the backup server.
19. A computer program product, comprising a non-transitory computer-readable medium having a computer-readable program code embodied therein, the computer-readable program code adapted to be executed by one or more processors to perform a method of finding a root cause of an anomaly in a network environment, the method comprising:
representing assets in the network environment as respective nodes in a causal graph, wherein the nodes have a measurable quality that can be tracked and arcs between pairs of nodes represent causal relationships between nodes of the node pairs;
designating source nodes as processes at the top of a hierarchy of tracked processes, and sink nodes as processes at the bottom of the hierarchy and having characteristics of interest in the environment;
detecting anomalies in the tracked processes embodied in the sink nodes;
traversing the causal graph in a reverse order from a node in which an outlier is detected; and
analyzing nodes along the traversal path to identify a node of the highest hierarchy that shows unusual behavior as the root cause in a prioritized list of possible root causes.