Anomalies detection
Summarize
Summary of Anomalies Detection
The Anomalies Detection feature in ServiceNow, specifically in the Xanadu release, offers proactive monitoring of both cyclical and non-cyclical metrics under the performance chart for the Impact Total package on production instances. This functionality helps you identify outliers in metrics based on historical patterns. While not every anomaly indicates an issue, it gives you the opportunity to assess criticality and configure alerts as needed.
Show less
Key Features
- Cyclical Metrics: Metrics that repeat in a regular pattern, including:
- Transaction Count
- Server Response Time
- SQL Response Time
- Semaphore Mean
- Node Memory Max
- Visualization: Anomalies are marked in red on the line chart, representing deviations from historical averages and distribution patterns.
- Job Anomaly Detection: Tracks concurrent scheduled jobs and overlays transaction counts, highlighting anomalies in red for further investigation.
- Job Criteria: A job qualifies as scheduled if it runs daily or at least once each week over the last four weeks.
- Average Transaction Count Calculation: Averages transaction counts over the past four weeks for better anomaly detection.
- Anomaly Identification: Anomalies are defined by comparing current job execution times against historical averages and standard deviations.
Key Outcomes
By utilizing the Anomalies Detection feature, you can effectively monitor and address performance issues, ensuring that you are alerted to significant deviations in job execution times. This empowers you to take timely actions based on the criticality of detected anomalies and configure alerts accordingly. For further guidance, refer to the section on configuring anomaly alerts.
Instance Observer proactively detects the anomalies for cyclical or non-cyclical metrics under the performance chart for the Impact Total package on production instances. Anomalies represent metrics outliers based on historical patterns. Every anomaly may not represent an issue, but, notifies you and you decide the criticality and configure alerts, accordingly.
Cyclical metrics occur as a complete set of events that repeat themselves regularly in the same order or in a regularly repeated period. Non-cyclical metrics are metrics that repeat themselves irregularly or in random, less predictable repeated periods. An anomaly, also known as an outlier, is a data point that is unusual, rare, or doesn't conform to the expected patterns or distribution of the data.
- Transaction count: The instance-wide sum of all UI transactions of an internal type known as UI_TYPE.
- Server Response Time: The average or mean execution time for UI_TYPE transactions.
- SQL Response Time: The reported mean of database response time measured at the application layer that starts when a query is sent to the database and finishes when the response has been received.
- Semaphore Mean: The average number of end-user transactions being processed concurrently over a one-minute period.
- Node Memory Max: The in use memory max in MB per node at a given data point in history. This value generally ranges between 1000 MB to 2048 MB.
- The x-axis represents time, and the y-axis represents the actual metrics as per date range selected. For example, the line chart displays the transaction count values over time. Anomalies are denoted by red color coding on the chart and represent the occurrence of an anomaly in the data. The placement of the red mark depends on the criteria or algorithm used to detect anomalies.
- The range represents the upper and lower boundary limits with a normal distribution of the metrics based on their historical dataset pattern.
- The mean line represents the four week average value of the metrics to compare the deviation at a given point in time.
Job anomaly detection
Job anomaly charts track the number of scheduled jobs running concurrently for each hour of the day with the overlaying metric of the average of transaction counts for each hour. Any bar in the chart that has one or multiple anomalous jobs is highlighted as red. Select the detail link of the bar chart to view the job level details.
Job Details will represent the hourly scheduled jobs along with respective average transactions of that instance.
Drill down from the hourly scheduled job count into an individual recurrence job for any hour of day and further into the execution pattern of the same job from the last seven days. This can help to perform end to end root cause for jobs that usually take a consistent amount of time to complete that suddenly experience a significant increase or decrease in execution time, indicating a possible performance issue.
- Schedule job criteria
- For the job to be considered a scheduled job, it should satisfy at least one of the following criteria:
- The job runs at least once for each day of the week.
- The job has run at least once for every week in the past four weeks.
- Average transaction count calculation
- For every hour in a given day, the sum of the transaction count of the past four weeks for the same day and same hour is averaged. For example, by fetching the sum of transaction counts from the past four Mondays for the fourth hour and averages the values for the final calculation.
- Job anomaly identification
- For the past four weeks, for every hour in which the job ran, the average duration time is calculated with the standard deviation value, which is the mean value, plus 5 multiplied by the standard deviation value.
Anomaly response
All anomalies don’t represent an issue, but, the outliers detected based on historical patterns. Configure alerts accordingly after analyzing the criticality of the anomaly detected. See Configure anomaly alerts for more information on alerts.