The Analytics Layer
Those that have worked in large networks know that translating large collections of performance data into useful knowledge is incredibly challenging. What does a a radio channel noise floor of -73 mean to the user experience. An RF engineer probably knows, at least theoretically, but with thousands of such values being reported on a regular basis it is nearly impossible to manually develop a health model for all circumstances that the various statistics convey.
That is where the Analytics Layer comes in. With access to both the values from networked devices and customer feedback (automated speed test results, customer surveys, calls to customer support, etc.) it is possible to draw conclusions by correlating source data with results data to draw conclusions that can be used to improve the self-healing process as well as to improve customer education and customer support procedures.
Besides improving the customer experience, all of the same data can be used to produce sophisticated business intelligence knowledge that can be used to determine ways to improve profitability on existing business lines, identify opportunities for other service tiers or business lines and collect information on the key benefits that users find particularly valuable so that promotional efforts can focus on those benefits.
To facilitate this level of analysis, the Analytics Layer specifies that collected data should be stored in big-data NoSQL datastores (e.g., Cassandra, Mongo, HBase) and processed using scalable clustered server infrastructures (e.g., Hadoop) and analyzed by sophisticated statistical analysis tools based on a standard statistics expression language like R.
The Analysis Layer benefits too from the fact that the information conveyed from networked devices are the out-of-standard cases that are the only ones the devices report (having done their own evaluation and self-healing) so that baselines are pre-computed and allows the processing on those exceptional cases on their implications.