Method and apparatus for root cause and critical pattern prediction using virtual directed graphs
An approach to root cause determination in a complex systems based on monitoring and event data is disclosed. It includes a historical analysis of events with their probabilistic correlations. Applying information measures between the random variables which embody those events one can detect origins of problems and generate real-time recommendations for their locations in a hierarchical system. Estimation of system bottlenecks, as well as the risk of “black swan”-type events are also computed. The processes are based on a statistical processing of a virtual directed graph produced from historical events.
1. A method for determining a causative event in an information technology infrastructure comprising:
collecting historical event data with a computer;
generating an undirected graph, each node being an event and each edge being a joint probability of two events historically occurring simultaneously;
generating a directed graph, each node being the event and each edge being a conditional probability of a second event occurring if a first event occurs;
reducing the undirected graph by discarding edges based on mutual information calculated for each edge;
reducing the directed graph by discarding edges that correspond to discarded edges of the undirected graph; and
from the directed graph, creating and outputting an historical list of recommendations to a user, the historical list of recommendations comprising an at least one event deemed to be the most likely causative event.
2. The method of claim 1 , further comprising
comparing active events against the historical events of the directed graph;
identifying current active events that correspond to events in the historical list of recommendations; and
from the identified active events, creating and outputting a list of active mode recommendations comprising an at least one event deemed to be potentially a causative event.
3. The method of claim 1 , wherein the creating and outputting of the historical list of recommendations comprises determining entropy for the nodes of the directed graph.
4. The method of claim 3 , wherein the at least one event deemed to be the most likely causative event are the events having the lowest entropy.
5. The method of claim 3 , wherein the creating and outputting of the historical list of recommendations further comprises determining a relative recommendation index.
6. The method of claim 5 , wherein the at least one event deemed to be the most likely causative event are the events having the highest relative recommendation index.
7. The method of claim 5 , wherein the creating and outputting of the historical list of recommendations further comprises:
determining a list relative rank;
wherein the at least one event deemed to be the most likely causative event are the events having the highest list relative ranks.
8. The method of claim 1 , further comprising determining at least one of root nodes, critical nodes, critical paths, extreme paths, and critical sectors from at least one of the historical list of recommendations and the directed graph.
9. The method of claim 1 , further comprising determining bottlenecks from at least one of the historical list of recommendations and the directed graph.
10. The method of claim 1 , further comprising determining black swan events from at least one of the historical list of recommendations and the directed graph.
11. A non-transitory machine readable medium with instructions stored thereon for determining a causative event in an information technology infrastructure comprising:
collecting historical event data with a computer;
generating an undirected graph, each node being an event and each edge being a joint probability of two events historically occurring simultaneously;
generating a directed graph, each node being the event and each edge being a conditional probability of a second event occurring if a first event occurs;
reducing the undirected graph by discarding edges based on mutual information calculated for each each edge;
reducing the directed graph by discarding edges that correspond to discarded edges of the undirected graph; and
from the directed graph, creating and outputting an historical list of recommendations to a user, the historical list of recommendations comprising an at least one event deemed to be the most likely causative event.
12. The machine readable medium of claim 11 , further comprising
comparing active events against the historical events of the directed graph;
identifying current active events that correspond to events in the historical list of recommendations; and
from the identified active events, creating and outputting a list of active mode recommendations comprising an at least one event deemed to be potentially a causative event.
13. The machine readable medium of claim 11 , wherein the creating and outputting of the historical of recommendations comprises determining entropy for the nodes of the directed graph.
14. The machine readable medium of claim 13 , wherein the at least one event deemed to be the most likely causative event are the events having the lowest entropy.
15. The machine readable medium of claim 13 , wherein the creating and outputting of the historical list of recommendations further comprises determining a relative recommendation index.
16. The machine readable medium of claim 15 , wherein the at least one event deemed to be the most likely causative event are the events having the highest relative recommendation index.
17. The machine readable medium of claim 15 , wherein the creating and outputting of the historical list of recommendations further comprises:
determining a list relative rank;
wherein the at least one event deemed to be the most likely causative event are the events having the highest list relative ranks.
18. The machine readable medium of claim 11 , further comprising determining at least one of root nodes, critical nodes, critical paths, extreme paths, and critical sectors from at least one of the historical list of recommendations and the directed graph.
19. The machine readable medium of claim 11 , further comprising determining bottlenecks from at least one of the historical list of recommendations and the directed graph.
20. The machine readable medium of claim 11 , further comprising determining black swan events from at least one of the historical list of recommendations and the directed graph.