Doug Safford from AllState Insurance gave a fascinating overview of a large Insurance IT department’s migration from a high-overhead ($1B pa budget!) mainframe shop to an architecture-driven organisation. This primarily revolved around the introduction of a standard ESB - a business event bus in this case, with a standard middleware solution (TIBCO EMS) with a standard framework (to handle the governance needs of tracking, logging, etc). It also involved much simplification of the IT processes: going from 10 to 2 data centers, for example, and from 15,000 (!!) to just 4 (!!!) AS400-class servers.

Doug mentioned that the enablers for their Business Process Management were frameworks, services, events and rules. Of these:

• Services are carefully managed: they have deployed 4,000+ TIBCO BusinessWorks engines in the service and framework areas!
• Moving to an Event Driven Architecture allowed much more business control of their business events, such as standardised routing and de-duping. The events feed their CEP engines as well as their Data Warehouses for analytics.
• Separation of the business logic / rules was seen as critical in such a regulated industry; they could not afford to have volatile rules replicated in software code around their systems. AllState have standardised on TIBCO BusinessEvents as their rules engine (with 800+ engines today, and the fastest growing tool deployment for them).

In terms of the ROI for their investment, Doug mentioned:

• Marketing campaigns can now be deployed from new in weeks rather than 6-12 months, or from existing campaign templates in days rather than months.
• Rule changes, such as for routing leads to agents, can be done overnight, usually from the BAM reports. In one example they monitored the success rates for agents in closing business, re-routed more leads to them, to gain a 15% increase in closing rates…

I would not be at all surprised to see this AllState IT transformation to be covered in business management books and MBA programs in future!

Prof. Zicardi of ODBMS.org (remember ODBMS?[*1]) published an interesting interview with HP Fellow (and past RDBMS architect) Dr Goetz Graefe on the rising importance of “data storage” without the twirling-disks-thing. Of course this is a trend very much in evidence at customers of TIBCO with (”NoSQL”-type, or more accurately, “SQL-optional”) technologies like TIBCO BusinessEvents using distributed DataGrid technology for faster-than-disk / more-scale-than-a-single-process event correlating. And for those  looking to apply “diskless” to SOA as well as EDA, note that TIBCO ActiveSpaces now includes the TIBCO BusinessWorks interface for doing data distribution in BW processes…

With CEP we have long questioned the need to “bung everything onto disk” and instead process it (the incoming events) as they arrive… of course you will want a historic record for some events, but fundamentally you don’t usually want to follow the old-model of “put it in the database then worry about / process it later”. Unless you are an RDBMS sales guy anyway!

Notes:

*1. “Object-oriented” makes perfect sense until one tries too hard on encapsulation (of behavior). For example, a business rule relating new-order events with existing customers could be encapsulated in the order event or the customer event, but in reality should be in neither. Although you might want to be able to refer to related rules from either…

TIBCO, Oracle, IBM and F.U.Berlin presented the latest tutorial from the EPTS Reference Architecture Group today at the ACM Distributed Event Based Systems Conference being held at IBM’s TJ Watson Research Labs. This covered the current work of the EPTS Reference Architecture group on defining (and refining) common event processing design patterns like “filter”, “enrich” and “route”.

This tutorial proved interesting to prepare: we discussed patterns in terms of the different event processing technologies of the presenters: rules-oriented event processing (TIBCO BusinessEvents, IBM Websphere Decision Server Business Events component, and Prova) and stream-oriented event processing (Oracle CEP, and IBM Infosphere Streams). In future patterns work we’ll try and expand this list (as well as the patterns covered), for example to include the TIBCO BusinessEvents Query Language, Pattern Framework, and State Model, and maybe TIBCO Hawk rules.Watch this space!

Standard Event Processing Design Pattern

Good to see a post by Cap Gemini’s Roeland Loggen on a Sense-Research-Respond pattern he is observing in their System Integration business. Although I would probably put  the Research (i.e. long-term predictive analytics) as part of the feedback loop, it is basically similar to the event-decision-action cycle we’ve mentioned before. I wonder how long it will take for SI’s to start marketing their increasing prowess in, and uses of, Complex Event Processing?

Analytics Magazine on OODA

While discussing the proposed OMG Decision Modelling standard with business rules specialist Ron Ross recently, Ron reminded me of the list of “roles” that “decisions” can provide:

Classification, Evaluation, Selection, Approval, Assessment, Assignment, Allocation, Diagnosis, Prediction

Curious as to what sources might refer to this list, I “googled” for these terms and one of the first relevant results was…

Threat evaluation and weapon assignment decision support: A review of the state of the art (2006-2007)

This is a fascinating paper that targets the military domain but nonetheless is very relevant to CEP (and indeed business) methodology:

1. “Threat evaluation” is also a business task: business threats are things like poor customer satisfaction, poorly managed product marketing, changing business conditions (locally or across the business), inability to predict business changes, regulatory compliance risks, and so on.
2. “Command and control” (a.k.a. business management) methodologies are also of interest: the paper refers to Boyd’s Orient-Observe-Decide-Act (OODA) cycle, Lawson’s Sense-Process-Compare-Decide-Act cycle, and a Monitor-Assess-Plan-Execute cycle.
   • Orient means setting up measurements, or event sources
   • Observe, Sense-Process-Compare, and Monitor means event (including “complex event”) detection, and all the associated processes for complex event processing.
   • Decide means making some decision based on our observed events.
   • Act and Plan-Execute mean behaviors that in turn will likely affect the environment and future events.
3. The paper refers to the JDL Data Fusion model which may well be familiar to some CEP folk: this model describes (with a minor translation from the table in the paper) the use of and relationships between:
   • event assessment
   • complex event or entity assessment
   • situation or state assessment
   • impact or cost-benefit assessment
   • performance or goal assessment.

I would therefore claim that the relationship between events and decisions seems not only very clear, but also well recognised.