Q&A from BCS SPA meeting on CEP

Reading Time: 2 minutes

We had a good meeting at the BCS SPA in London this week, with a great set of questions from the audience. Immo‘s notes and the slides will be placed up on the SPA web site, so we’ll simply visit some of the audience questions here:

1. What is the relationship between CEP and AI?

You don’t have to be a specialist in computer software history to observe that the use of things like “rule engines” and “neural nets” for event pattern detection has similarities with Artificial Intelligence concepts like “search”. Clearly one could argue that pattern detection is a common requirement for AI systems, and therefore the use of AI algorithms in CEP makes a lot of sense. However, no-one in the CEP community (AFAIK) is claiming any particular advances in AI research due to the development of CEP technologies.

2. What is the relationship between CEP and BI?

Dr Fan Yang was asking about CEP and BI. Most of the discussion here was around the observation that an event cache is rarely going to have the same magnitude of data as a conventional data warehouse… at least for the current generation of CEP technologies. This could, however, change in future – CEP is still a relatively new technology area and the boundaries of capabilities have not yet been reached (or more importantly, breached). So: probably there are BI reports that can use the data from a CEP system, and there may be predictive models that are too expensive to run in real-time in-situ in CEP (although this might be solved with truly distributed event stores). But for sure, the traditional BI and CEP fields overlap, with the former being more historic / after-the-event, and the latter more as-it-happens / predictive.

3. What is the relationship between CEP and military systems?

Another astute question was the military applicability and/or heritage for CEP. Sensor data fusion is real-time information merging and pattern detection, and is clearly related to CEP technology.

AFAIK there are no current military systems (as opposed to government intelligence systems) using Commercial Off The Shelf CEP systems, although I recall one commercial product being developed with US military money (your tax $ at work, etc etc).

4. How to model Events?
It was mentioned (in passing, of course) that TIBCO used standard UML class models and state models, and near-standard production rules (the latter via PRR, which isn’t finalized yet), in its CEP product. So why not standardized “event modeling”? Clearly there is a UML definition for event, but event pattern correlation etc is the subject of a new UML extension and/or profile called “EMP”. Other than that, events can be modeled as special classes (i.e. they have a timestamp, payload, and attributes). [This topic was discussed recently in the blog].

Thanks to the BCS organisers for this event.