Message log analysis for system behavior evaluation
View Patent ↗A technique is disclosed that enables the run-time behavior of a data-processing system to be analyzed and, in many cases, to be predicted. In particular, the illustrative embodiment of the present invention comprises i) transforming the messages that constitute an unstructured log into a numerical series and ii) applying a time-series analysis on the resultant series for the purpose of pattern detection. Indeed, it is recognized in the illustrative embodiment that the problem really is to detect patterns that depict aspects of system behavior, regardless of the textual content of the individual log messages. In other words, by analyzing the totality of the messages in the log or logs—as opposed to looking for pre-defined patterns of the individual messages—system behavior can be mapped and understood. The mapping helps in characterizing the system for the purposes of predicting failure, determining the time required to reach stability during failure recovery, and so forth.
1. A method comprising:
mapping, by a controller, a first plurality of messages from a message log to a mapped plurality of numerical values, wherein each numerical value from the mapped plurality of numerical values is mapped to only one message from the first plurality of messages;
performing, by the controller, a time-series analysis on the mapped plurality of numerical values with respect to when corresponding messages from the first plurality of messages occur in time; and
transmitting, by the controller, a signal that represents a characteristic of the time-series analysis;
wherein the controller comprises a processor coupled to memory.
2. The method of claim 1 wherein the mapping of the first plurality comprises:
sorting, by the controller, the first plurality of messages via a string-based sort technique into a sorted set of sorted messages; and
assigning, by the controller, a numerical value to each sorted message based on its position in the sorted set, the numerical value constituting the mapped plurality.
3. The method of claim 2 wherein the sorting is based on the Levenshtein distance between the messages in the first plurality.
4. The method of claim 1 wherein the mapping of the first plurality comprises:
receiving, by the controller, a first message in the first plurality of messages;
determining, by the controller, a distance between a first message string, which is representative of the first message, and a reference string; and
assigning, by the controller, a numerical value to the distance, the numerical value constituting the mapped plurality.
5. The method of claim 4 wherein the distance is based on one of the Levenshtein distance and the Hamming distance.
6. The method of claim 5 further comprising:
partitioning, by the controller, the first message into J multiple substrings, J being an integer greater than one; and
assigning, by the controller, weights to each substring based on the position of the string in the first message, the weighted set of substrings resulting in the first message string.
7. The method of claim 6 wherein the value of J is based on an estimated number of possible messages in the message log.
8. The method of claim 1 further comprising re-booting a data-processing system, based on the transmitted signal that represents a characteristic of the time-series analysis.
9. A method comprising:
sorting, by a controller, a first plurality of messages from a message log via a string-based sort technique into a sorted set of messages, wherein the sorted set of messages comprises all messages from the first plurality;
assigning, by the controller, a unique index to each of the sorted messages to form a non-empty set of indexes, wherein:
(i) each index in the non-empty set of indexes is assigned to only one message from the first plurality of messages, and
(ii) each index in the non-empty set of indexes is assigned to a corresponding message based on the position of the corresponding message in the sorted set of messages;
performing, by the controller, a time-series analysis on the non-empty set of indexes with respect to when corresponding messages from the first plurality of messages occur in time; and
transmitting, by the controller, a signal that represents a characteristic of the time-series analysis;
wherein the controller comprises a processor coupled to memory.
10. The method of claim 9 wherein the sorting is based on the Levenshtein distance between the messages in the first plurality of messages.
11. The method of claim 9 further comprising collating, by the controller, a second plurality of messages from the message log, resulting in a collated set of messages that comprises the first plurality.
12. The method of claim 11 further comprising deleting, by the controller, of at least one duplicate message from the collated set, resulting in a remaining set of messages that comprises the first plurality.
13. The method of claim 9 wherein the performing of the time-series analysis involves determining, by the controller, of a message density of the non-empty set of indexes.
14. The method of claim 9 further comprising re-booting a data-processing system, based on the transmitted signal that represents a characteristic of the time-series analysis.
15. A method comprising:
receiving, by a controller, a first message in a first plurality of messages from a message log;
determining, by the controller, a distance between a first message string, which is representative of the first message, and a reference string, wherein the determining includes:
(i) partitioning the first message into J multiple substrings, J being an integer greater than one and wherein the value of J is based on an estimated number of possible messages in the message log, and
(ii) assigning, by the controller, weights to each substring based on the position of the string in the first message, the weighted set of substrings resulting in the first message string;
assigning, by the controller, a numerical value to the distance, the numerical value constituting a mapped plurality of numerical values;
performing, by the controller, a time-series analysis on the mapped plurality with respect to when corresponding messages from the first plurality occur in time; and
transmitting, by the controller, a signal that represents a characteristic of the time-series analysis;
wherein the controller comprises a processor coupled to memory.
16. The method of claim 15 wherein the distance is based on the Levenshtein distance.
17. The method of claim 15 wherein the performing of the time-series analysis involves determining a message density of the mapped plurality.
18. The method of claim 15 further comprising re-booting a data-processing system, based on the transmitted signal that represents a characteristic of the time-series analysis.