I think writing historical data of Mbeans/attributes etc to some file/s will be great to see what happened when the system being monitored went high on resources or came crashed. This will enable JManage to be used without active human monitoring. I know Alerts/graphs area there with selective attrs, but then user can't see what was the pattern when high resource consumption or crash happened (unless somebody is actively monitoring and that too for few minutes only). And then make the file available to view via Web, preferably in graph/s. Files can be rolled-out by date/size and number of files etc..
For ex, a user can specify what all mbeans/attrs he's interested in, set their polling frequency, and then JManage start writing attr values to a CSV file, which can be seen anytime via a browser (graph/table) within a particular time-frame.
This can be integrated with my earlier feature-request - to invoke some operations when some threshold exceed, like, if we get thread dump at the time when system was on high resources (preferably periodic dumps), then along with above information, it'd be much easy to understand (and hopefully debug) the underlying issue.