Kumar,
You need to distinguish between Capacity Planning and System/Server Health and Resource Utilization. Tim is correct and that the number of concurrent sessions is the primary guidance used for capacity planning of the ISE deployment or given node. However, there is no guarantee that customer will achieve max potential capacity based on various factors in real environment including network health, auth load and excessive errors, features implemented, and adherence to best practices.
For planning, there should always be a buffer in capacity to accommodate failover, sub-optimal distribution of sessions, activity bursts, anomalous conditions, and individual feature scale. For server health, the metrics cited are typical ones to watch. Certainly endpoint distribution (charts added in ISE 2.2), auth activity, alarms, and latency are also key metrics.
Starting in ISE 2.2, we have exposed the Key Performance Metrics (KPM) into Admin UI reports (previously requires export from CLI) as well as ISE Counters and Thresholds which expose the Profiler statics (previously requires display from CLI) which provide more details on RADIUS load and Profiling activity--two of the more impactful indicators of system load. Process usage has also been broken out into greater detail and exposed to Admin UI reports.
Unfortunately, many application level metrics are not exposed via SNMP, but you can set various alarm thresholds for triggering a log or email. These can then be further processed by loggers to trigger SNMP Traps. Some basic SNMP system MIBs and traps are available today. For SNS appliances, monitoring via the CIMC is an option. For virtual appliances, monitoring via the hypervisor tools is available.
Hope that helps.
Craig
