vincent
Sun, 28 Oct 2007 22:00:41 +0000
I presented the “CEP is a good idea” pitch [*1] to the Business Rules forum audience, on the basis that:
- all business rules relate to events, in the sense that they apply to, or are invoked by, some kind of event
- Complex Event Processing can involve many algorithms, but especially rules - particularly when making useful decisions based on detected patterns of behavior
- the type of rule engine used for CEP is very similar to that used in the BRMS market - the main difference being the execution of long-running rule processes (statefull) rather than fire-n-forget transactional decision processes (stateless) [*2].
I haven’t seen the feedback summary yet, but we had quite a full room given there was an interesting panel (on “Emerging Trends in Enterprise Decisioning“) running at the same time in another room. We also had an extended Q&A session afterwards with some great questions from the audience, like:
A. Why doesn’t performance drop off when you add more rules? [Answer: this is the role of the rule engine, to provide scaleability]
B. How can you fit an entire corporate data model and history into a CEP process? [Answer: this is a design issue, but the tools to help here include the scaleable and distributed cache, ability to set persistence on class types, etc].
C. Are all CEP tools rule-driven? [Answer: er, no… but eventually there are probably rules of some kind in any CEP process]. [*3]
And another thing - I was pleasantly surprised to find 3 TIBCO CEP end-users in the audience! So perhaps the convergence of event processing is not just in my (or IDC’s) head!
Notes:
[*1] In some previous business rule conferences, the CEP pitch has been abley given by the IBM Amit R&D team members. So the topic is not new to the general audience (i.e. Thanks to IBM for the warm-up!).
[*2] However, most commercial BREs can run in both stateful and stateless modes. Usually it is a matter of optimization - no non-CEP BRE for example includes a persistence mechanism for stateful operation. On the other hand, you would not expect CEP-optimised BRE’s like TIBCO BusinessEvents to include special adapters for stateless operation under particular vendor implementations of J2EE application servers.
[*3] I have been struggling to find a classification scheme for CEP technologies, especially when the recent Event Processing Technology Society saw some “event processing” case studies that seemed very close to vanilla business processing / transaction processing. I had thought of using the terms “Simple Event Simple Processing” through to “Complex Event Complex Processing” to indicate the amount of 1st level event filtering involved, and then the 2nd level of assessment processing of that event. This is based on the old Comp Sci classification of parallel processing (Single/Multiple Instruction/Data) schemes. Probably you could describe query-based approaces to CEP as CESP, whereas when you add rule inferencing or process chaining to drive business decisions, you would get to CECP. And the standard event-driving-a-transaction model can be SESP or SECP…


Source...