Detection of resource bottlenecks in user devices using artificial intelligence and causal graphs
Techniques are provided for detection of resource bottlenecks in computing devices. A particular resource bottleneck can be identified as a cause of a current device issue based on a dynamic evaluation, by an anomaly detection module, of performance metrics of a computing device. Once a particular resource bottleneck is identified as anomalous, one or more corresponding adjustments to configuration settings for the computing device can be identified to mitigate the current device issue using a causal graph that represents the dependencies among (i) various issue types for a computing device, (ii) performance metrics of the computing device to evaluate for each issue type, and (iii) one or more resources that may be a cause of a given issue type. The corresponding adjustments to the computing device can be automatically identified based on the resource determined to be the cause of the given device issue type.
1. A method, comprising:
obtaining a causal graph comprising connections between: (i) a plurality of performance issue types for a computing device in a first hierarchical level of the causal graph, (ii) one or more performance metrics of the computing device to evaluate for each performance issue type in a second hierarchical level of the causal graph, and (iii) one or more resources in a third hierarchical level of the causal graph that may be a cause of a given performance issue type;
obtaining values of at least some of the one or more performance metrics of the computing device;
identifying an anomaly indicating a current performance issue type by: (i) applying the obtained values of the performance metrics to an anomaly detection module that generates an anomaly score for each of the corresponding performance metrics, and (ii) analyzing the anomaly score for each of the corresponding performance metrics to identify the anomaly indicating the current performance issue type;
in response to the identifying the anomaly indicating the current performance issue type:
automatically processing the causal graph starting with a first node in the first hierarchical level of the causal graph corresponding to the current performance issue type and traversing to one or more next performance metric nodes, in the second hierarchical level of the causal graph, connected to the first node, wherein each next performance metric node identifies one of the values of the performance metrics of the computing device to evaluate;
automatically identifying an anomalous performance metric node in the causal graph having an anomaly based at least in part on a comparison of the value of the performance metric associated with each traversed next performance metric node to one or more corresponding predefined criteria; and
following the anomalous performance metric node to a next resource node of the causal graph to automatically identify a given resource that is a cause of the current performance issue type;
wherein the method is performed by at least one processing device comprising a processor coupled to a memory.
2. The method of claim 1 , wherein the one or more performance metrics comprise one or more of at least one device level performance metric and at least one process level performance metric.
3. The method of claim 1 , wherein the causal graph is one or more of obtained from one or more domain experts to represent a logical approach of the one or more domain experts, and obtained using one or more semi-supervised techniques.
4. The method of claim 1 , wherein the values of the one or more performance metrics of the computing device are obtained from a time series telemetry data stream for each of a plurality of key performance indicators during an execution of the computing device.
5. The method of claim 1 , wherein the given resource that is a cause of the current performance issue type comprises a node in the causal graph in a substantially maximal position of the causal graph.
6. The method of claim 1 , wherein the causal graph comprises a directed graph having a plurality of nodes each corresponding to a different variable, wherein an arc is present between a given first node and a given second node when the variable of the given second node responds to changes in the variable of the given first node.
7. The method of claim 1 , wherein the identified anomaly comprises one or more of a trend change anomaly, a high stress anomaly, a surge anomaly, and a low value anomaly.
8. The method of claim 1 , wherein the next resource node further comprises one or more corresponding adjustments to one or more configuration settings for the computing device to mitigate the current performance issue type.
9. An apparatus comprising:
at least one processing device comprising a processor coupled to a memory;
the at least one processing device being configured to implement the following steps:
obtaining a causal graph comprising connections between: (i) a plurality of performance issue types for a computing device in a first hierarchical level of the causal graph, (ii) one or more performance metrics of the computing device to evaluate for each performance issue type in a second hierarchical level of the causal graph, and (iii) one or more resources in a third hierarchical level of the causal graph that may be a cause of a given performance issue type;
obtaining values of at least some of the one or more performance metrics of the computing device;
identifying an anomaly indicating a current performance issue type by: (i) applying the obtained values of the performance metrics to an anomaly detection module that generates an anomaly score for each of the corresponding performance metrics, and (ii) analyzing the anomaly score for each of the corresponding performance metrics is analyzed to identify the anomaly indicating the current performance issue type;
in response to the identifying the anomaly indicating the current performance issue type:
automatically processing the causal graph starting with a first node in the first hierarchical level of the causal graph corresponding to the current performance issue type and traversing to one or more next performance metric nodes, in the second hierarchical level of the causal graph, connected to the first node, wherein each next performance metric node identifies one of the values of the performance metrics of the computing device to evaluate;
automatically identifying an anomalous performance metric node in the causal graph having an anomaly based at least in part on a comparison of the value of the performance metric associated with each traversed next performance metric node to one or more corresponding predefined criteria; and
following the anomalous performance metric node to a next resource node of the causal graph to automatically identify a given resource that is a cause of the current performance issue type.
10. The apparatus of claim 9 , wherein the values of the one or more performance metrics of the computing device are obtained from a time series telemetry data stream for each of a plurality of key performance indicators during an execution of the computing device.
11. The apparatus of claim 9 , wherein the given resource that is a cause of the current performance issue type comprises a node in the causal graph in a substantially maximal position of the causal graph.
12. The apparatus of claim 9 , wherein the causal graph comprises a directed graph having a plurality of nodes each corresponding to a different variable, wherein an arc is present between a given first node and a given second node when the variable of the given second node responds to changes in the variable of the given first node.
13. The apparatus of claim 9 , wherein the identified anomaly comprises one or more of a trend change anomaly, a high stress anomaly, a surge anomaly, and a low value anomaly.
14. The apparatus of claim 9 , wherein the next resource node further comprises one or more corresponding adjustments to one or more configuration settings for the computing device to mitigate the current performance issue type.
15. A non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to perform the following steps:
obtaining a causal graph comprising connections between: (i) a plurality of performance issue types for a computing device in a first hierarchical level of the causal graph, (ii) one or more performance metrics of the computing device to evaluate for each performance issue type in a second hierarchical level of the causal graph, and (iii) one or more resources in a third hierarchical level of the causal graph that may be a cause of a given performance issue type;
obtaining values of at least some of the one or more performance metrics of the computing device;
identifying an anomaly indicating a current performance issue type by: (i) applying the obtained values of the performance metrics to an anomaly detection module that generates an anomaly score for each of the corresponding performance metrics, and (ii) analyzing the anomaly score for each of the corresponding performance metrics to identify the anomaly indicating the current performance issue type;
in response to the identifying the anomaly indicating the current performance issue type:
automatically processing the causal graph starting with a first node in the first hierarchical level of the causal graph corresponding to the current performance issue type and traversing to one or more next performance metric nodes, in the second hierarchical level of the causal graph, connected to the first node, wherein each next performance metric node identifies one of the values of the performance metrics of the computing device to evaluate;
automatically identifying an anomalous performance metric node in the causal graph having an anomaly based at least in part on a comparison of the value of the performance metric associated with each traversed next performance metric node to one or more corresponding predefined criteria; and
following the anomalous performance metric node to a next resource node of the causal graph to automatically identify a given resource that is a cause of the current performance issue type.
16. The non-transitory processor-readable storage medium of claim 15 , wherein the values of the one or more performance metrics of the computing device are obtained from a time series telemetry data stream for each of a plurality of key performance indicators during an execution of the computing device.
17. The non-transitory processor-readable storage medium of claim 15 , wherein the given resource that is a cause of the current performance issue type comprises a node in the causal graph in a substantially maximal position of the causal graph.
18. The non-transitory processor-readable storage medium of claim 15 , wherein the causal graph comprises a directed graph having a plurality of nodes each corresponding to a different variable, wherein an arc is present between a given first node and a given second node when the variable of the given second node responds to changes in the variable of the given first node.
19. The non-transitory processor-readable storage medium of claim 15 , wherein the identified anomaly comprises one or more of a trend change anomaly, a high stress anomaly, a surge anomaly, and a low value anomaly.
20. The non-transitory processor-readable storage medium of claim 15 , wherein the next resource node further comprises one or more corresponding adjustments to one or more configuration settings for the computing device to mitigate the current performance issue type.