Thursday, December 20, 2012

On event oriented thinking



The title of this Blog is "event processing thinking", and much of the posts are my own thoughts about event processing, but today I would like to write about another topic: event oriented thinking.  
An interesting observation from recent discussions is that  in daily life we can very easily think in an event-driven way,  when the phone is ringing we are either answering it or ignoring it, when we hear about traffic jam - we try to find another way, and so on...      However - when it comes to thinking about computerized systems,  many people are programmed to think in a request-driven way, which means:  a person sends a request and the computer responses.   Thus people are trying to use what they know even if they have scenarios that are event-driven.  A very typical line of thinking is:  event is a data, we should insert it in a database, and then ask query that reflects the situation that we wish to detect.   The question is -- when are we going to ask this query.  We can ask it on any update -- but if the event is just one component in a pattern we wish to detect (example: we wish to detect when a customer sent three requests within a single day -  assuming that most customers don't send more than one request, then we can have 99.99% of the queries redundant),  we can ask the query periodically - but then again, we can both ask redundant queries, and moreover, not able to react on time, since the granularity of our timing is wrong.   There is also a notion of continuous query -- but this is not really a query, since it is not a result of a specific request.  

The nature of event-driven scenarios are: we don't know when they are going to happen, we even don't know whether they are going to happen,  but when they happen we want to do something - sometimes very fast (e.g. earthquake detection).   Furthermore, the situation we would like to detect can be realized in a pattern of many events,  and each individual event may trigger reaction, just increment some state, or even be ignored or filtered out.

The fact that people are trying to model and implement event-driven scenarios using the traditional request response is a thinking mismatch, and creates added complexity.  I'll follow up with an example in one of the next posts.  More - later.

No comments: