In a previous post we opined that discrete events (i.e. representing a point in time, not a period in time) were what we mostly meant by the term “event”, and “historic events” over a period of time were in fact “states”. To some extent this might be irrelevant as you might still need interval logic operators in your CEP system to determine whether or not some states overlap (or, if you prefer, if some complex events with durations overlap). And there is a precision issue here due to the granularity of the clock(s) being used…
Boundary case issues with datetimes can be somewhat fuzzy: however these are often not too important in business. For example, your tax form submission date limit will tend to have some well defined specifications on what your tax authorities will accept or not. However there are other opportunities for “fuzzy event processing” – which usually mean “fuzzy event patterns” – that might be useful.
I am reminded of this last point by some recent discussions about the role of technologies like TIBCO Netrics in complex event processing. For example Netrics provides a “fuzzy matching engine” which is usually used for data cleansing in things like MDM systems. A sample use case is that when provided with partial details of a person and their address, the engine can quickly match (in real time) and respond with an ordered list of possible candidates. In event processing this could be useful in the way that Google Instant search does your searching in a predictive fashion: it might be we have enough information for a “match” before all the events are received. This could complement real-time analytics and the such like for intelligent, reactive CEP systems.
