When its first version was published twenty years ago the prognosis for OMG’s UML (Unified Modeling Language) was of rapid and wide expansion. It didn’t happen, and notwithstanding a noteworthy usage, UML has not become “the” unified modeling language. Beyond the diverse agendas of methods and tools providers, this falling short may have something to do with a lack of robust semantics, as illustrated by UML 2.5’s halfhearted attempt to define individuals.
UML 2.5 Aborted Attempt with Individual
UML 2.5 has often been presented as an attempt to redress the wayward and increasingly convoluted paths taken by the previous versions. Yet, beside some useful (and long needed) clarifications and adjustments, its governing group failed to agree on some compact and unambiguous semantics and had to content itself with perfunctory guidelines introduced as an afterthought.
As a matter of fact the OMG committee may have tried to get its semantics in order, as suggested by the chapter “On semantics” making the point right away with a distinction between the things to be described and the categories to be applied:
“A UML model consists of three major categories of model elements [classifiers, events, and behaviors], each of which may be used to make statements about different kinds of individual things within the system being modeled (termed simply “individuals”in the following)”.
That straightforward understanding (UML is meant to describe individual objects, events, or behaviors) could have provided the semantic cornerstone of a sound approach. Surprisingly, and for obscure reasons, it is soon smudged by syntactical overlapping and semantic ambiguity, the term “individual” being used indifferently as adjective and noun, and then appears to be restricted to classifiers only. That leaves UML with a dearth of clear semantics regarding its scope.
Individual as a Semantic Master Key
The early dismiss of individual as a constitutive concept is unfortunate because, taking a leaf from Archimedes, itcould have been the fulcrum on which to place the UML lever.
To begin with, no modeling language, especially one supposed to be unified, can do without some convincing semantics about what it is supposed to describe:
Analysis (aka extensional) models are descriptive as they deal with external individuals, things or behaviors, capturing their relevant aspects.
Design (aka prescriptive) models are prescriptive as they describe how to build software components and execute processes.
As analysis and design models serve different purposes they clearly differ. Nonetheless, since UML is meant to straddles the divide between business and systems realms, some rigorous mechanism must ensure a persistent and consistent mapping of individuals.
That could have been neatly achieved with a comprehensive and unified interpretation of individuals, combined with a clear taxonomy of the aspects to be modeled:
Individuals are whatever occurrences (in business or systems contexts) with identities of their own.
These individuals (objects, events, or behaviors) can be specified with regard to their structure and relationships.
The logical primacy of this approach is reinforced by its immediate, practical, and conclusive benefits for business processes modeling on one side, model based engineering processes on the other side.
A Key to Business Processes Modeling
As far as business processes are concerned, modeling the part played by supporting systems turns around few critical issues, and these issues can be dealt more clearly and consistently when set in reference to individuals taxonomy (objects, behaviors, events) e.g:
Functional or non functional requirements ? The former can be associated with individuals, the latter cannot.
Architecture or application ? The former affect the specification of interactions between individuals, the latter affect only their local features.
Synchronous or asynchronous ? Specifications can only be made with regard to life-cycles and time-frames: system (objects), process (behaviors), or instant (event).
Structures or Relationships ? The former are bound to individuals’ identity, the latter are used between different individuals.
Inheritance or Delegation ? The former is used for the specification of structural or functional features, the latter for individuals’ behaviors.
More generally that understanding of individuals should greatly enhance the alignment of systems functional architectures with business processes.
A Key to Model Based Systems Engineering
As should be expected from a lack of semantic foundations, one of the main characteristics of the UML community is its fragmented practices, regrouped around diagrams (e.g Use case or Class) or task (e.g requirements analysis or code generation).
The challenge can be directly observed for model based system engineering and software development: with the exception of Statecharts for RT modeling, Class diagrams are the only ones used all along engineering processes; the others, when used, are reduced to documentation purposes. That bottleneck in development flows can be seen as the direct consequence of UML restricted semantics: since behaviors are not identified as individuals in their own right, their description cannot be directly translated into software artifacts, but have to be understood as part of active objects descriptions before being translated into class diagrams. Hence the apparent redundancy of corresponding diagrams.
As a corollary, reinstating a unified semantics of individual for both classifiers and behaviors could be the key to a seamless integration of the main UML diagrams. Most important, that would bear out the cross benefits of combining UML and MBSE.
A Key to Enterprise Architecture
Enterprise architecture can be defined in terms of territories and maps, the former materialized by the realities of enterprise organization and business operations, the latter by the assortment of charts, blueprints, or models used to describe enterprise organization, business processes, IT systems, and software applications. Along that understanding the whole endeavor depends on the ability to manage the continuity and consistency of charting; and that cannot be done without a unified and persistent identification mechanism of objects and processes across business, enterprise, and systems.
Use cases are meant to describe dialogues between users and systems, yet they usually don’t stand in isolation. On one hand they stem from business processes, on the other hand they must be realized by system functionalities, some already supported, others to be newly developed. As bringing the three facets within a shared modeling paradigm may seem a long shot, some practical options may be at hand.
To begin with, a typical option would be to use the OMG’s Meta-Object Facility (MOF) to translate the relevant subset of BPM models into UML ones. But then if, as previously suggested, use cases can be matched with such a subset, a more limited and pragmatic approach would be to introduce some modeling primitives targeting UML artifacts.
For that purpose it will be necessary to clarify the action semantics associated with the communication between processes and systems.
Contrary to models, which deal with instances of business objects and activities, meta-models are supposedly blind to business contents as they are meant to describe modeling artifacts independently of what they stand for in business contexts. To take an example, Customer is a modeling type, UML Actor is a meta-modeling one.
To stress the point, meta-languages have nothing to say about the semantics of targeted domains: they only know the language constructs used to describe domains contents and the mapping rules between those constructs.
As a consequence, whereas meta-languages are at their best when applied to clear and compact modeling languages, they scale poorly because of the exponential complexity of rules and the need to deal with all and every language idiosyncrasies. Moreover, performances degrade critically with ambiguities because even limited semantics uncertainties often pollute the meaning of trustworthy neighbors generating cascades of doubtful translations (see Knowledge-based model transformation).
Yet, scalability and ambiguity problems can be overcame by applying a core of unambiguous modeling constructs to a specific subset, and that can be achieved with use cases.
Use Cases & UML diagrams
As it happened, the Use Case can be seen as the native UML construct, the original backbone around which preexisting modeling artifacts have been consolidated, becoming a central and versatile hub of UML-based development processes:
Sequence diagrams describe collaborations between system components but can also describe interactions between actors and systems (i.e use cases).
Activity diagrams describe the business logic to be applied by use cases.
Class diagrams can be used for the design of software components but also for the analysis of business objects referenced by use cases.
State diagrams can be used for the behavior of software components but also for the states of business objects or processes along execution paths of use cases.
Apart of being a cornerstone of UML modeling, use cases are also its doorway as they can be represented by a simple sequence diagram describing interactions between users and systems, i.e between business processes and applications. So the next step should be to boil down the semantics of those interactions.
Use Cases & Action Semantics
When use cases are understood as gateways between business users and supporting systems, it should be possible to characterize the basic action semantics of messages in terms of expectations with regard to changes and activity:
Change: what process expects from system with regard to the representation of business context.
Activity: what process expects from system with regard to its supporting role.
Crossing the two criteria gives four basic situations for users-to-system action semantics:
Reading: no relevant change in the environment has to be registered, and no activity has to be supported.
Monitoring: no relevant change in the environment has to be registered, but the system is supposed to keep track on some activity.
Achievement: the system is supposed to keep track on some specific change in the environment without carrying out any specific activity.
Accomplishment: the system is supposed to keep track on some specific change in the environment and support associated activity.
These action primitives have to be wrapped into interaction ones relating to changing expectations of users and systems.
When use cases are positioned at the center of UML diagrams, those situations can be neatly modeled with state machines for actors’ expectations, business objects, and execution.
Use Cases & Modeling Patterns
If use cases describe what business processes expect from supporting systems, they can be used to map action semantics to UML diagrams:
Reading: class diagrams with relevant queries.
Monitoring: activity diagrams with reference to class diagrams.
Achievement: class diagrams with associated state diagrams.
Accomplishment: activity diagrams with associated state diagrams and reference to class diagrams.
While that catalog cannot pretend to fully support requirements, the part it does support comes with two critical benefits:
It fully and consistently describes the interactions at architecture level.
It can be unambiguously translated from BPM to UML.
On that basis, use cases provide a compact and unambiguous kernel of modeling constructs bridging the gap between BPM and UML.
As can be understood from their theoretical basis (Pi-Calculus, Petri Nets, or State Machines), processes are meant to describe the concurrent execution of activities. Assuming that large enterprises have to keep a level of indirection between operations and business logic, it ensues that activities and business logic should be defined independently of the way they are executed by business processes.
For that purpose two basic modeling approaches are available: BPM (Business Process Modeling) takes the business perspective, UML (Unified Modeling Language) takes the engineering one. Yet, each falls short with regard to a pivotal conceptual distinctions: BPM lumps together process execution and business logic, and UML makes no difference between business and software process execution. One way out of the difficulty would be to single out communications between agents (humans or systems), and specify interactions independently of their contents and channels.
That could be achieved if communication semantics were defined independently of domain-specific languages (for information contents) and technical architecture (for communication channels). As it happens, and not by chance, the outcome would neatly coincide with use cases.
Linguistics & Computers Parlance
Business and functional requirements (see Requirements taxonomy) can be expressed with formal or natural languages. Requirements expressed with formal languages, domain-specific or generic, can be directly translated into some executable specifications. But when natural languages are used to describe what business expects from systems, requirements often require some elicitation.
When that challenge is put into a linguistic perspective, two school of thought can be considered, computational or functional.
The former approach is epitomized by Chomsky’s Generative Grammar and its claim that all languages, natural or otherwise, share an innate universal grammar (UG) supporting syntactic processing independently of their meanings. Nowadays, and notwithstanding its initial favor, that “computer friendly” paradigm hasn’t kept much track in general linguistics.
Alternatively, the functionalist school of thought considers linguistics as a general cognitive capability deprived of any autonomy. Along that reasoning there is no way to separate domain-specific semantics from linguistic constructs, which means that requirements complexities, linguistic or business specific, have to be dealt with as a lump, with or without the help of knowledgeable machines.
In between, a third approach has emerged that considers language as a functional system uniquely dedicated to communication and the mapping of meanings to symbolic representations. On that basis it should be possible to separate the communication apparatus (functional semantics) from the complexities of business (knowledge representation).
Processes Execution & Action Languages
Assuming the primary objective of business processes is to manage the concurrent execution of activities, their modeling should be driven by events and their consequences for interactions between business and systems. Unfortunately, that approach is not explicitly supported by BPM or UML.
Contrary to the “simplex” mapping of business information into corresponding data models (e.g using Relational Theory), models of business and systems processes (e.g Petri Nets or State Machines) have to be set in a “duplex” configuration as they are meant to operate simultaneously. Neither BPM nor UML are well equipped to deal with the task:
The BPM perspective is governed by business logic independently of interactions with systems.
Executable UML approaches are centered on software processes execution, and their extensions to action semantics deal essentially on class instances, features value, and objects states.
Such shortcomings are of no serious consequences for stand-alone applications, i.e when what happens at architecture level can be ignored; but overlooking the distinction between the respective semantics of business and software processes execution may critically hamper the usefulness, and even the validity, of models pertaining to represent distributed interactions between users and systems. Communication semantics may help to deal with the difficulty by providing relevant stereotypes and patterns.
Business Process Models
While business process models can (and should) also be used to feed software engineering processes, their primary purpose is to define how concurrent business operations are to be executed. As far as systems engineering is concerned, that will tally with three basic steps:
Dock process interactions (aka sessions) to their business context: references to agents, business objects and time-frames.
Specify interactions: references to context, roles, events, messages, and time related constraints.
Specify information: structures, features, and rules.
Although modeling languages and tools usually support those tasks, the distinctions remain implicit, leaving users with the choice of semantics. In the absence of explicit guidelines confusion may ensue, e.g between business rules and their use by applications (BPM), or between business and system events (UML). Hence the benefits of introducing functional primitives dedicated to the description of interactions.
Such functional semantics can be illustrated by the well known CRUD primitives for the creation, reading, updating and deletion of objects, a similar approach being also applied to the design of domain specific patterns or primitives supported by functional frameworks. While thick on the ground, most of the corresponding communication frameworks deal with specific domains or technical protocols without factoring out what pertains to communication semantics independently of information contents or technical channels.
But that distinction could be especially productive when applied to business processes as it would be possible to fully separate the semantics of communications between agents and supporting systems on one hand, and the business logic used to process business flows on the other hand.
Communication vs Action Semantics
Languages have developed to serve two different purposes: first as a means to communicate (a capability shared between humans and many primates), and then as a means to represent and process information (a capability specific to humans). Taking a leaf from functional approaches to linguistics, it may be possible to characterize messages with regard to action semantics, more precisely associated activity and attendant changes:
No change: messages relating to passive (objects) or active (performed activity) states.
Change: messages relating to achievement (no activity) or accomplishment (attendant on performed activity).
Communication semantics can then be fully rounded off by adding changes in agents’ expectations to the ones in the states of objects and activities, all neatly modeled with state machines.
It must also be noted that factoring out the modeling of agents’ expectations with regard to communications is in line with the principles of service oriented architectures (SOA).
Additionally, transitions would have to be characterized by:
Time-frame: single or different.
Address space : single or different.
Mode: information, request for information, or request for action.
Organizing business processes along those principles, would enable the alignment of BPMs with their UML counterpart.
Use Cases as Bridges between BPM & UML
Use cases can be seen as the default entry point for UML modeling, and as such they should provide a bridge from business process models. That can be achieved by if use cases are understood as a combination of interactions and activities, the former obtained from communications as defined above, the latter from business logic.
One step further, the distinction between communication semantics and business contents could be used to align models respectively to systems and software architectures:
Communication semantic constructs could be mapped to systems architectures, ideally to services oriented ones.
Uses cases are meant to describe how users interact with systems, classes are meant to describe software components, including those executing use cases. It ensues that classes are introduced with the realization of use cases but are not supposed to appear as such in their definition.
The Case for Use Cases
Use cases (UCs) are the brain child of Ivar Jacobson and often considered as the main innovation introduced by UML. Their success, which largely outdoes UML’s footprint, can be explained by their focus and simplicity:
Focus: UCs are meant to describe what happens between users and systems. As such they are neatly bounded with regard to their purpose (UCs are the detailed parts of business processes supported by systems) and realization (UCs are implemented by software applications).
Simplicity: while UCs may eventually include formal (e.g pre- and post-conditions) and graphical (e.g activity diagrams) specifications, they can be fully defined and neatly circumscribed using stick actors (for the roles played by users or any other system) and ellipses (for system behaviors).
As it often happens to successful innovations, use cases have been widely interpreted and extended; nonetheless, the original concepts introduced by Ivar Jacobson remain basically unaltered.
The Point of Use Cases
Whereas focus and simplicity are clearly helpful, the primary success factor is that UCs have a point, namely they provide a conceptual bridge between business and system perspectives. That appears clearly when UCs are compared to main alternatives like users’ stories or functional requirements:
Users’ stories are set from business perspective and lack explicit constructs for the parts supported by systems. As a consequence they may flounder to identify and describe business functions meant to be shared across business processes.
Conversely, functional requirements are set from system perspective and have no built-in constructs linking business contexts and concerns to their system counterparts. As a consequence they may fall short if business requirements cannot be set upfront or are meant to change with business opportunities.
Along that understanding, nothing should be done to UCs that could compromise their mediating role between business value and system capabilities, the former driven by changes in business environment and enterprise ability to seize opportunities, the latter by the continuity of operations and the effective use of technical or informational assets.
Business Objects vs Software Components
Users’ requirements are driven by concrete, partial, and specific business expectations, and it’s up to architects to weld those diverse and changing views into the consistent and stable functional abstractions that will be implemented by software components.
Given that double discrepancy of objectives and time-scales, business analysts should not try to align their requirements with software designs, and system analysts should not try to second-guess their business counterparts with regard to future business objects. As a consequence, respective outcomes would be best achieved through a clear separation of concerns:
Use cases deal with the business value of applications, mapping views on business objects to aspects of classes.
Functional architectures deal with assets, in particular the continuous and consistent representation of business objects by software components as described by classes.
As it happens, that double classification with regard to scope and purpose should also be used to choose a development model: agile when scope and purpose can be united, phased approach otherwise.
Whereas UML has been brought to existence by very wise men under very propitious skies, the initial enthusiasm and first successes have never been transformed into wider acceptance and customary usage; subsequent updates and extensions didn’t help and may even have triggered some anticlimax. More than fifteen years after its launch, the utilization of UML remains limited, both in breadth (projects developed) and depth (features effectively used). Moreover, the UML house is deeply divided and there isn’t much consensus among the few that use it comprehensively and consistently, principally to support domain specific languages (DSL).
Certainly, there must have been a wrong turn somewhere, possibly at the UML2 crossing when the OMG committee lost sight of users modeling needs and took the road to meta-models. Considering UML’s shrinking stamp and dwindling relevancy, that road appears more and more like a dead-end; but it may be still possible to get back on track and retrieve the Us of the UML: unified semantics for all and sundry users.
Where to Look
Whether on driving or back seats, respectively for model driven or agile methods, models are widely accepted as a necessary constituent of development processes. Nonetheless, and despite being the only official standard, UML standing appears to falter, up to be already seen as a cold case. As suggested by Ivar Jacobson (“The road ahead for UML“), one of its main drawback would be its lack of modularity with regard of users needs. If that flaw is to be fixed, the question is where to look: directly at language level, or at supporting mechanisms.
Given the broad consensus that surrounded the initial project, one should at first look for a sound and stable subset to be used as a backbone and fleshed out according specific contexts, purposes or users. As a matter of fact that is what stereotypes and profiles are meant to do, except that without a well-defined backbone of unambiguous constructs, the only possible outcomes are domain specific languages. So, one should first consider how the separation of concerns could be better supported by language constructs.
Separation of Concerns
Despite its roots in the Object Oriented paradigm, UML has demonstrated its adaptability to all and every method or domain. Unfortunately, being a Jack of all trades often means a master to none, and the use of UML is clearly frustrated by its versatility; that translates either into shallow usage of ambiguous semantics, or into extensions targeting specific domains or technologies.
On the ground, three mechanisms can be used to make for the lack of focus: stereotypes, views, and customization.
UML stereotyping mechanism support predefined constructs for problem (business objects and processes) or solution (system architecture and object design) spaces. Stereotypes can be grouped into profiles, e.g for specific business domains or technical architectures.
Views (or perspectives) organize access to models according contents: logical, physical, conceptual, pragmatic, etc
Tool customization organizes access to models according users purposes and skills: analyst, architect, designer, developer, etc.
While those approaches have their benefits, they are set independently of languages constructs, either as UML extensions (stereotypes and profiles), or defined from outside by development methodologies (views) or projects organization (customization). As a consequence, they have little or no effect on the simplicity or efficiency of UML; they may even add to confusion and complexity when overlapping stereotypes are introduced to support multiple taxonomies, e.g technical architectures and business domains.
That may point to a clear direction: given the potency of the stereotyping mechanism and its pivotal role in UML utilization, significant benefits could be achieved through a better integration into core language constructs, even if that entails some constraints or limitations. Two straight modifications should be considered:
Model layers: language constructs should be re-organized along architectural concerns for enterprise (business processes), system (functionalities), and platforms (components).
Stereotypes visibility: language constructs should support the distinction between local taxonomies and “unified” ones, the former set with limited scope and visibility, the latter meant to be applies across layers.
Given the growing intricacy, ubiquity and diversity of systems, UML complexity and versatility should clearly be in demand, and the problem is to harness those capabilities according the needs and skills of the different kinds of users.
That’s arguably a critical flaw of UML, which lumps together essential with secondary constructs, as well as definite with ambivalent semantics. That brings weighty consequences, both for users and models:
Steep or even abrupt learning curve: confronted to a wall of mixed constructs users have to master the whole upfront, whatever their needs and skills.
Blurred concerns: describing various specific contents with the same ambivalent constructs will either distort language semantics, or blur concerns specificities.
Corrupted transformation: whatever the modeling tools, the bad apples of inputs will usually corrupt the whole of outputs. In other words any advance in model driven development requires a sound backbone of unambiguous language constructs.
As noted above, language constructs can be regrouped along two perspectives, one directly associated with users architectural concerns, the other reflecting the scope and visibility of targeted artifacts. While there is no particular reason to match complexity levels with architectural concerns, mapping them to granularity has a clear rationale. Such a “born again” UML would distinguish between two levels of language constructs:
Those pertaining to objects and activities identified by architectures, whatever their nature: enterprise, systems or platforms.
Those used to describe internals of objects and activities independently of their aspects and behaviors at architecture level.
That re-configuration would bring modularity to the language, enabling a smooth learning curve. More importantly, a clear-cut separation of concerns will enable some kind of Just-In-Time model transformation: instead of cumulative noises (b), one will get separate transformations for models architectural backbone on one hand, contingent specificities on the other hand (a). And that could be a real game-changer for lean and fit models.
While that could be achieved by different means, a simple solution would be to use the stereotyping mechanism to describe supporting structures of enterprise, functional, and technical architectures.
Transformation vs Portability
Model transformation is about changing contents within the same environment, portability is about moving the same contents across different environments; and despite apparent similarities, they deal with different concerns, set by users for the former, by tools vendors for the latter.
Transformation is normally performed under a single corporate roof according agreed semantics; as a corollary, it is meant to cover the full contents of models. That’s not the case for portability, whose primary objective is the exchange of consolidated contents between heterogeneous environments; while sources and targets may have to share the whole of their models, a sound policy should make room for selective portability of specific or confidential contents.
The Meta-Object Facility (MOF) is the solution of choice for portability. As a meta-language it is used to describe language constructs at source and target environments; mapping rules can then be defined and bridges built between environments. As it is, those bridges usually scale very poorly due to the exponential complexity of rules having to cover all and every model idiosyncrasies; and that’s unfortunate for portability which, instead of focused targets, has to deal with overweight models cluttered with useless contents (b).
That situation would be greatly improved (a) if the wheat of consolidated constituents could be separated from the chaff of ambiguous or irrelevant contents. On a broader perspective that will open the way leaner and fitter models.
One step back may put UML back on track
There is something of a consensus among the software engineering community regarding (1) the benefits of models and (2) the failures of UML. As should be expected, that consensus translates into fragmented modeling practices and, more generally, software engineering methodologies. Obviously there isn’t much of a future for UML along that path, but the case is still open and the trend can be reversed by putting users needs back on UML driving seat.
Model driven software engineering (aka MDA, aka MBSE, aka MDE) has had a great future for quite some time, yet there isn’t much consensus about what that could be and, in particular, about what kind of models should be in the driving seat.
Pending some agreement about model contents (e.g specific ?) and capabilities (e.g executable ?), the driving of software engineering processes will probably remain more practices than principles.
Models are shadows of reality, with their form and contents set by contexts and concerns. They can be characterized by their purpose and capabilities.
Regarding purpose, models fall in two groups: descriptive models deal with problems at hand, prescriptive models with solutions.
Descriptive models are partial and biased representations of actual contexts. Partial because they only deal with relevant objects, activities and features; biased because the selection is made on purpose.
Prescriptive models are complete descriptions of artifacts.
Regarding capabilities the distinction is between intensional and extensional languages:
Extensional (aka denotative) languages deal with sets of identified instances of objects and activities. As they condone partial or ambiguous statements, they are best used for descriptive models.
Intensional (aka connotative) languages deal with the semantics and constraints of symbolic representations. Due to their formal capabilities they are best used for prescriptive models.
Along that reasoning System Models can be characterized along architecture contexts: business processes (enterprise), functionalities (systems), and platform implementations (technologies):
Business models are descriptive and built with extensional languages (business is often said to bloom on discrepancies). They are necessarily partial as they target specific contexts and concerns.
Functional (aka analysis) models are prescriptive and built with intensional languages as they must specify the semantics and constraints of symbolic representations. Yet they are not necessarily complete since they don’t have to cover every details of business processes or implementations (cf traceability).
Implementation models (aka design) are prescriptive with no use for extensional capabilities since the relevant physical objects, i.e extensions, are system components directly derived from system specifications. However, they must support complete and formal descriptions of component features.
Models don’t Talk Alone
Models are built with logographic languages that should not be confused with phonetic ones: whereas the latter convey information sequentially, the former build semantics from different sources; that enables models to be read from different perspectives. Contrary to programs whose semantic is bound to a fixed sequential execution, models don’t talk alone, but must be chatted to. Even when their readings are sequential, the sequences are governed by readers, not by models.
That point is pivotal if model transformation, arguably a pillar of model driven development, is to be supported along different perspectives and according different concerns.
Besides, it must be noted that while models can be fully translated into (and reversed from) sequential representations (e.g with XMI), transcripts are just projections and should not be confused with models as such.
Models don’t Walk Alone
Like talks, walks are sequential as they advance step by step. Hence, while some models can be walked (aka executed), such walks are only instances of behaviors supported by models.
That should be especially clear for system models which describe architectures combining agents and devices as well as software components, contrary to programs which are limited to software components structure and sequential (or parallel) execution.
Just like XMI transcripts should not be confused with original models, “executable” models should not be confused with fully fledged ones because they simulate the behaviors of agents and devices as if they were software components. While that may be useful for models targeting software components within a given architecture, ignoring the specificities of software, agents, and devices would be pointless when the objective is to design a system architecture.
Defining models as abstractions may be correct but not very helpful when deciding what kind of abstraction should drive development processes. First, the question is to define how concerns and purposes should be used to define abstractions, in other words to set apart significant from non-significant information. Then, in order to avoid flights for always higher abstractions without burying models into ground details, some principles are needed regarding specialization and generalization.
When systems are considered, abstraction levels are set by enterprise organization, systems functionalities, and platform technologies, with concerns expressed by business, functional, or technical requirements. Given a hierarchy of concerns, models must be set at the proper level of abstraction: below that level part of information will be redundant or irrelevant; above, useful features and classifications may be overlooked as some information will be either wanting or will not discriminate between variants.
Such levels can be identified by selectively applying specialization and generalization to constrained hierarchies:
Inheritance should not cross model layers: hierarchies of business, functional and technical artifacts must be kept separate.
Structures and behaviors pertain to different concerns: abstractions of objects and aspects should be managed independently.
Specialization should be applied when subset of identified objects or behavior are associated with specific features. Generalization should be introduced when models must be consolidated.
Such an approach brings significant benefits for reuse, arguably one of the main objective of abstraction. And that appears clearly when developments are governed by models and architecture components and mechanisms are to be shared across model layers.
Driving Models and Roadmaps
Systems engineering has to meet three kinds of requirements: business needs, system functionalities, and components implementation. In a perfect world there would be one stakeholder, one architecture, and one time-scale. Unfortunately, projects may involve different stakeholders, target different architectures, and be set along different time-frames. In that case milestones and roadmaps are to be introduced in order to bring all expectations and commitments under a common roof, with models providing the glue between them. That may be achieved with models defined along MDA layers:
Computation Independent Models (CIMs) describe business objects and processes independently of supporting systems.
Platform Independent Models (PIMs) describe how business processes are supported by systems seen as functional black boxes, i.e disregarding the constraints associated to candidate technologies.
Platform Specific Models (PSMs) describe system components as implemented by specific technologies.
Interestingly, both extremities of development processes are textual descriptions, informal at the beginning, formal at the end, with models providing bridges in between. As noted above, those bridges are not always necessary: texts can be directly translated into instructions (as illustrated by voice command devices), or project teams with shared ownership can develop programs according users’ requirements (as promoted by Agile methods). Yet, the question should always be asked, and when textual requirements cannot be directly developed into programs the first task should be to draw the shortest modeling path.
Roadmaps and Meta-models
Model driven tools may appear under different guises yet most rely on meta-models and the Meta Object Facility (MOF). Given that meta-models are models of models, they are supposed to focus on a subset of relevant features selected on purpose, which, for driving models, would be some kind of road signs governing models transformation. What that could be? Two approaches are to be considered:
Language translation: as presented by the report of the Dagstuhl Seminar, meta-models can be designed according their generic transformation capabilities and used to single out language constructs in order to transfer model contents into another language.
Separation of concerns: as development advances and models take into account different concerns, meta-models can be used to monitor and control the selective processing of corresponding contents. That could be achieved if transformations were governed by traffic signs singling out relevant features and waving aside the leftovers.
Each option points to a different perspective, the former targeting tools providers, the latter aiming at modellers. Whereas MDA layers (for business, functionalities, and technology) clearly point to models built with the Unified Modeling Language according organizational, functional and technical concerns, most of current implementations follow the language option; while those tools may be (theoretically) open at technical level, they usually rely on domain specific languages. By narrowing functional scopes and relying on automated translation to bridge the gaps, solutions based upon domain specific languages can only provide local solutions to fragmented problems. That road could be a bumpy one for model driven engineering.
Reusing artifacts means using them in contexts that are different of their native ones. That may come by design, when specifications can anticipate on shared concerns, or as an afterthought, when initially unexpected similarities are identified later on.
Planned or opportunistic, reuse brings benefits in terms of costs, quality, and continuity:
Cost benefits are most easily achieved for component engineering but may also be obtained upstream with model reuse and patterns.
Quality benefits are first and foremost rooted at model level, especially when components implementation is supported by automated tools.
Continuity benefits are to be found both along the business (semantics and business rules) and engineering (functional architecture and platform implementations) perspectives.
Reuse policies may also bring positive externalities by inducing a comprehensive approach to software design.
Yet, those policies will usually entail costs and may as well bear negative externalities:
Artifacts designed for reuse are usually most costly to develop, even if part of additional costs should be ascribed to quality management.
Excessive enforcement policies may significantly hamper projects ability to meet business needs in time.
Managing reusable assets usually induces overheads.
In order to assess those policies, economics of reuse must be set across business, engineering or architecture perspectives:
Business perspective: how to factor out and reuse artifacts associated with the knowledge of business domains when system functionalities or platforms are modified.
Engineering perspective: how to reuse development artifacts when business contents or system platforms are modified.
Architecture perspective: how to use system components independently of changes affecting business contents or development artifacts.
That can be achieved by managing development assets along model driven architecture: CIMs for business and enterprise architecture, PIMs for systems functional architecture, and PSMs for systems technical architecture.
Contexts & Concerns
Whatever their inception, reused artifacts are meant to be used independently of their native context and concerns: opportunistic reuse will map a specific purpose to another one, planned reuse will map a shared concern to a specific purpose. As a corollary, reuse policies must be supported by some traceability mechanism linking concerns and purposes across contexts and architectures.
From the enterprise perspective, the problem is to reuse the knowledge of business domains and processes. For that purpose different mechanisms can be considered:
The simplest solution is to reuse generic components, with the business knowledge directly transferred through parameters.
Similarly, business rules can be separately edited and managed in business contexts before being executed by system components.
One step further, business semantics and rules can be fenced off with domains and applied to different objects and applications.
Finally, models of business objects and processes can be capitalized and managed as reusable assets.
Once business knowledge is duly capitalized as functional assets, they can be reused along the engineering perspective:
System functionalities: functional patterns are (re)used to map functional requirements to functional architecture, and services are (re)used to support business processes.
Software implementations: Component-based development and information hiding.
Along the architecture perspective information hiding is generalized to systems, and reuse is masked by the definition of services.
It must be noted that, contrary to misleading similarities, refactoring is the opposite of reuse: instead of building from well understood and safe artifacts, it tries to extract some reusable chunks from opaque and unsafe components.
Enterprise and business knowledge may affect the full scope of system functionalities: boundaries (e.g users authorizations), controls (e.g accounting rules), and persistency (e.g consistency constraints). Whereas there isn’t much to argue about the benefits of reusing enterprise and business knowledge, costs may significantly diverge depending on the way corresponding assets are managed:
Domain specific knowledge are rooted at requirements level. That’s typically achieved when use cases are introduced to describe how systems are meant to support business processes. With different use cases targeting different aspects of the same business objects and processes, overlaps must be identified and factored out in order to be reused across processes.
Business knowledge may also be global, i.e shared at enterprise architecture level, defining objectives, assets and organization associated with the continuity of corporate identity and business capabilities within a regulatory and market environment.
In any case, the challenge is to map business knowledge to system models, more precisely to embed reused descriptions of business objects and process to corresponding development artifacts. At architecture level the mapping should target objects or processes identities, at domain level the focus will be on aspects and views.
As epitomized by service oriented architectures, business architectures can be mapped to system ones through delegation, either directly (business processes calling on services), or indirectly (collaboration between services). That will establish a clear distinction between shared (aka global) and domain specific knowledge, and consequently between respective economics of reuse.
Given that shared knowledge must be reused across domains and applications, there can be no argument about benefits. That will be achieved by a messaging model built on a need-to-know basis. And since such model is an intrinsic feature of the functional architecture, it incurs no additional overheads.
Specific knowledge for its part is managed at domain level and therefore masked behind services interfaces. Whatever reuse occurs there remains local and an intrinsic part of domain design.
Things are not so clear when business knowledge is not managed by services but distributed across domains, mixing specific and global knowledge. Managing reusable assets would be easy were the distinction between business and functional requirements to coincide with the one between shared and specific knowledge; unfortunately that’s seldom the case, and requirements, functional or business, will have to be sorted out at architecture or design level.
Whereas Service Oriented Architectures (SOA) put functionalities in the driving seat, Enterprise Application Integration (EAI) gives the lead to applications for which it provides adapters. As maintenance of integration adapters is a very poor substitute for knowledge management, reuse is mostly limited to legacy applications.
At design level knowledge is weaved into canonical data models (entities), functional architectures (controls), and user interfaces frameworks (boundaries).
On one hand, tracing reuse to requirements may be problematic as they are by nature concrete and unstructured, hence not the best support for generalization or the factoring out of shared features. Assuming business analysts can nonetheless separate reusable wheat from specific chaff, knowledge management at this level will require a dedicated framework supporting comprehensive and differentiated traceability. Additional overheads will have to be taken into account and compared to potential benefits.
On the other hand, canonical data models and functional architectures are meant to provide unified views of shared objects and semantic domains. Yet, canonical models are by nature unwieldy as they carve structures, features, and connections of business objects, without clear mechanisms to combine shared and specific knowledge. As a corollary, their use may reduce flexibility, and their management usually induces significant overheads.
With enterprise and business knowledge capitalized as development assets, the engineering case for reuse may appears indisputable, but business cases are often much more controversial due to large overheads and fleeting returns. Taking cues from Barry Boehm (“Managing Software Productivity and Reuse”), here are some of the main pitfalls of artifacts reuse:
Repository delusion: knowledge being by nature contextual, its reuse is driven by circumstances and purposes; as a consequence the availability of large repositories of development assets will probably be ignored without clear pointers rooted in contexts and concerns.
Confusion between components (or structures) and functionalities (or interfaces): under the influence of the object oriented paradigm, the distinction between objects and aspects is all too often forgotten. That’s unfortunate as this difference is congruent with the one between business objects on one hand, business operations on the other hand.
Over generalization: reuse is usually achieved by factoring out useful aspects or factoring off useless ones. In both cases the temptation is to repeat the operation until nothing could be added to the scope. Such “flight for abstraction” will inevitably overtake the proper level of reuse and begets models void of any anchor to business relevancy.
Scalability: while reuse is about separation of concerns and complexity management, those two criteria don’t have to pull in the same direction. When they don’t, variants will be dispersed across artifacts and their processing will suffer a combinatorial explosion if the system has to be scaled up.
Obsolescence: shelf lives of development assets can be defined by each or both business or technical relevancy. Assuming spans either coincide or are managed independently, they should be explicitly taken into account before any reuse.
Those obstacles can be managed providing that models:
Provide some built-in traceability mechanisms between knowledge and artifacts.
Sustainable system development is the ability to meet present business requirements while enhancing system capability to support future ones. Clearly reuse is not the only factor of sustainability, with architectures, returns, and risk management being pivotal. But the economics of reuse encompass most of other factors.
Architectures are clearly first to be considered, as epitomized by MDA:
Reuse of development assets rooted in enterprise architecture is not an option: system functionalities are meant to support business processes as they are (a).
At the other end of the development process, reuse of software designs and components across technical architectures should bring benefits in quality and costs (c).
In between reuse of system functionalities is necessary to guarantee the robustness and continuity of functional architectures; it should also leverage the benefits of reuse of enterprise and development assets (b).
Regarding returns, reuse through generic components, rules engines, or semantic domains can be directly supported by development tools, bypassing explicit models of functional architectures. That makes their costs/benefits analysis both simpler and well circumscribed. That is not the case for system functionalities which stand at the hub of perspectives. As a consequence, costs/benefits should be analyzed as a whole:
Regarding business assets, a clear distinction must be maintained between specific and shared knowledge, reuse being considered for the latter only.
Regarding the reuse of business assets as functional ones, services clearly offer the best returns. Otherwise costs/benefits are to be assessed, from reuse of vocabulary and semantics domains (straightforward, limited overheads), to canonical models and enterprise application integration (contingent, significant overheads).
Economics of reuse will ultimately be set by traceability mechanisms linking enterprise and business knowledge on one hand, components designs on the other hand. Even for services (c), if at a lesser degree, the business case for reuse will be decided by leveraged benefits and non cumulative costs. Hence the importance of maintaining the distinction between identified structures and associated aspects from business (a) and functional (b) requirements, to components interfaces (d) and structures (e).
Finally, reuse may also play a significant role in risk management, especially when risks are managed according to their source:
Changes in business contexts can usually occurs along two frequency waves: short, for market opportunities, and long, for an organization’s continuity. Associated risks could be better managed if corresponding knowledge were managed accordingly.
System architectures are meant to evolve in synch with organization continuity; were technological environment or corporate structures subject to unexpected changes, reusable functional assets would be of great help.
Given that enterprise IT can no longer be self-contained and operate in isolation, reusable designs may provide buffers to technological risks and help exploiting unexpected business opportunities.