The Tivoli Analytics for Service Performance product, which I first wrote about here and here, has since been rebranded under the IBM SmartCloud Analytics umbrella and is now known as IBM SmartCloud Analytics – Predictive Insights.
The latest 126.96.36.199 version introduces the following new features:
- Related Metrics: SmartCloud Analytics – Predictive Insights now allows you to open an alert for an anomalous metric and immediately discover and chart the set of related metrics. The set of related metrics is the set which IBM SmartCloud Analytics – Predictive Insights has established as the best predictors of the currently anomalous metric
- Progress Alarms: SmartCloud Analytics – Predictive Insights now produces series of progress alerts indicating your progress in ingesting data and creating an analytics model
- Multi topic – Data Segmentation: SmartCloud Analytics – Predictive Insights now allows you to group related data, for example, data related by geography or application, into discrete topics. Using multiple topics you can logically group and present anomalies
- Scalability – Distributed analytics installation: IBM SmartCloud Analytics – Predictive Insights now has the ability to scale linearly. You can now add multiple analytics components to your topology to increase the number of KPIs your system can handle
Of these new features, the ability to logically isolate groups of data into different topics looks to be a particularly significant development. When I first started looking at the beta offering I felt the inability to partition metric data to reflect customer, service or application specific configurations was a bit of a limitation. After all, if I’m 100% certain that there is no correlation relationship between two metrics, why do I want SCAPI to waste CPU cycles trying to find one? The counter to that argument was that the analytic algorithms are sophisticated enough to quickly learn that there is no correlation between those metrics, so little effort is actually wasted. Whilst that may be the case, from an operational perspective the ability to ring-fence one set of data from another is definitely welcomed. The support for multiple topics should also make dealing with different data sources easier, especially when it comes to determining the most appropriate aggregation interval. A nice touch is that anomaly alerts get tagged with the instance/topic details that generated them, providing a useful reference point for any subsequent filtering and processing.
Some useful resources:
- IBM Knowledge Center
- The Wiki on DeveloperWorks, which has some useful hints and tips
- The Wiki also contains a tutorial which covers the creation of a data model using the Mediation Tool, data extraction, training and working with the Service Diagnosis TIP portlet
- DeveloperWorks Forum