- Common Pitfalls Limiting the Value of SOA and BPM
- How Other Industries Approach Varying Conditions
- A Streamlined Enterprise Architecture for BPM and SOA
- Basic Principles for Enterprise Dynamicity
Basic Principles for Enterprise Dynamicity
Now that we have set the static scene for the enterprise architecture, how do we use all these elements in a way that allows a dynamic business and IT approach? Dynamicity implies flows and movements, with events and messages propagating through the enterprise and variations between and within the business components. Flows, such as the Order to Bill, flow across enterprise components and are often referred to as end-to-end processes. However, it is essential to differentiate the apparent effect of an event's chain reaction from an explicit choreography of business services owned by a specific organization within the enterprise. Both are called business processes but are different in nature and implementation.
To differentiate these business processes, we use business decompositions and map to a common terminology with a precise definition of the different types of processes at each level. For this purpose, we use the categorization of basic types of processes derived from the Business Process Modeling Notation (BPMN) standard.
Categorizing the Processes
The BPMN standard introduces three basic types of submodels within an end-to-end BPMN model:12
- Collaboration (global) processes—Business to business between enterprises or organizations
- Abstract (public) processes—Inside an enterprise across business components or internal organizations
- Private (internal) business processes—Within business components
In addition, BPMN introduces the notions of pools and lanes, which are also helpful in decomposing processes in modules and grains, allowing them to then be articulated in a flexible and dynamic fashion.
Collaboration (Global) Processes
A collaboration process is the description of the interactions between two completely independent business entities. Cross-industry standards, such as ebXML BPSS, RosettaNet, or to some extent, industry-specific standards, such as ACORD, describe such interactions sequences. The example collaboration process in Figure 1-9 shows a sequence of purchase order processing between two companies.
Figure 1-9 Example collaboration process
Even though there is a formal sequence, there is not an engine between the enterprises orchestrating that sequence. Each company/enterprise is responsible for sequencing its own part within its boundaries. EDI VANs are an early form of these collaboration processes, which are progressively disappearing because they inhibited flexibility.
Abstract (Public) Processes
These processes represent the overall interactions between different internal organizations in the enterprise owning their internal business rules or sequences. Consequently, abstract processes do not have a single business owner and would not be easily manageable if implemented as a single IT construct.
A first, obvious level of abstract processes is between business components as defined previously, but looking at the APQC decomposition, these abstract processes also occur between the processes at level 3 of the Process Classification Framework. Just as for the collaboration processes, there cannot be an engine controlling the interactions between the business components or level 3 processes because there would not be an owner of the logic in that engine. The overall sequence is the result of services calls and events occurring between each of the level 3 processes or business components. Understanding this difference between an explicit control and an implicit realization of the business process is essential to understanding the approach for implementing dynamic business processes.
In Figure 1-10, the "sales to bill" abstract process is represented by the interactions between the business components.
Figure 1-10 Abstract sales to bill process
Private (Internal) Business Processes
Within a single organization with a single identified business owner, the owner can manage the rules and sequences of tasks, which are then private to that "owner" or his delegates.
As a result, private processes are preferable for explicit flow control and automation because they are manageable assets of the enterprise, for which stake holders and life cycle can define.
They can be either explicit, if expressed in a workflow language or service choreography/orchestration language such as XML Process Definition Language (XPDL) or Web Services Business Process Execution Language (BPEL), or implicit, if resulting from the code running inside a particular application or package such as the ones provided by companies like SAP, Oracle, Amdocs, and so on. These processes chain to other private processes either by the means, events, or services calls. If they are exposed as services and consume services, they then can become components in an assembly model, such as standardized by the Service Component Architecture standard from the Open SOA organization.13 These private processes match the APQC processes that are at decomposition level 3. They operate within the boundaries of a business component, and there can be several private processes in a given business component or within a finer grained decomposition. Figure 1-11 shows an SCA assembly diagram where the customer order process (ProcessCustomerOrder) is exposed as a service and consumes other services, one of which is another process validating the customer order.
Figure 1-11 Private process exposed as a component and consuming other processes
This private process is exposed from within the eTOM level 3 as shown in Figure 1-12.
Figure 1-12 Private process in eTOM level 3
As you can see the service component in Figure 1-11 has more interfaces than the level 3 of the process decomposition in Figure 1-12, highlighting the difference between a component providing an implementation in the application map and a private process identified from the process decomposition.
Within a private process, a lane represents a sequence of tasks or activities driven by a specific participant. A participant can be any specific party role such as a buyer, seller, or call center operator, but also it can be a technical participation such a bus mediation or a program routine. The identification of a lane's nature has an impact on the selection of the appropriate technologies for implementing particular portions, services, or a private process. Pools are group of lanes and seem to imply multiple actors. However, this is not explicitly stated in the BPMN standard.
Applying Decomposition to End-to-End Process
As a consequence of decomposing processes in categories, an end-to-end process is the abstract or collaboration process resulting from the assembly of private processes that group the lanes that interact within their boundaries.
Here is an example or "order to cash" end-to-end abstract process for telecommunication operators, composed of a chain of private processes contained within the boundaries of a business component.
In Figure 1-13, we identify variations of private processes, such as the various service configurations and resource provisioning. This example also looks at private processes that are common successors in an abstract flow, such as the Set Top Box configuration required by the three services of a triple play—that is, Internet, television, and voice-over-IP calling services.
Figure 1-13 Abstract process of chaining private processes on a business map
In this specific case, the services can be submitServiceOrder with three implementations: submitServiceOrderForVoIP, submitServiceOrderForIPTV, and submitServiceOrderForADSL.
In a dynamic process approach, we must be able to add a quadruple play service such as gaming, with its specific service configuration and resources, without having to regenerate and test the "end to end" abstract process. This can only be achieved by means of dynamic binding and coordination between the various private processes.
Impact of Business Information Changes to Processes
The less you carry, the more agile you are. Similarly, the less information processes carry, the more they are able to handle variations of the enterprise information. But that information still needs to be accessible from the various business organizations of the enterprise, and the appropriate information must be available to the business components that require it for valid business needs.
To reach process agility, we have to move from a processes that carries explicitly all the information to processes that refer to information preserved elsewhere. Then, we only pass the core of the decisional information through processes, and preserve the bulk of the information using a master data management approach and information services to access that data as required by the business components and processes.
This still does not solve the variation of the interface to that information if attributes are added. An enterprise information variability approach is the necessary complement to ensure that within acceptable limits, the information model characteristics can change without affecting the interfaces.
The Enterprise Expansion Joint
This process and information dynamicity requires flexibility, and a mediation layer between components is one of the elements that provides such flexibility. The initial solution to the mediation layer in IT and a services-oriented architecture has been the Enterprise Services Bus. However, these buses often address the technical variability, looking at protocols and messaging, leaving the content interpretation to the integrated consumers or providers.
There is a need to push the mediation layer capabilities to content and business semantic adaptation, thus removing the interface's tight coupling of consumers and providers. As a simple example, the IT solution should enable the evolution of the versions of provided business services without requiring an immediate regeneration of business service consumers.
All the additional capabilities lead to what I call the enterprise expansion joint, the capacity for the business to expand using existing IT capabilities, without having to enter a long delivery process. The business variations of this expansion joint have to be bound to precisely identified limits that should be identified during the enterprise architecture phase. The enterprise architecture modeling method must then address the sizing of the business variability that this expansion layer can absorb, looking at foreseeable future business evolutions.
I had a concrete case where an enterprise wanted to deliver a new customer application and process while planning to change billing systems in a later phase. The target was to allow this change of billing systems for another packaged application without affecting any of the elements of the new customer care environment. So, the approach was to define what business services a billing system should expose to customer care in a top-down fashion and then use the mediation layer to perform the adaptation between exposed APIs and desired granularity and variability. In this specific case, the existing billing system was Tuxedo-based and required to use units of work in C for consistency, which led us to create a specific adapter, delivered in less than one month using the available adapter frameworks.
In addition to the granularity adaptation, the mediation layer may also have to address state as the reusable business services are designed to be independent of any implementation. Thus it may not expose specific data that is necessary for a specific application or package realizing the service.
In Chapter 6, "Implementing the Enterprise Expansion Joint," we discuss a set of techniques for realizing such a layer.