"I did not get what I envisioned" from the project, the senior official acknowledged. But he said the F.B.I. today had a better understanding of its computer needs and limitations as a result of the effort." The lesson we have learned from this $170 million is invaluable," he said.

This quote is from a $170 million failed software project in the US called the Virtual Case File or VCF by the FBI.  The system was designed to give the bureau's nearly 12,000 agents around the country instant access to F.B.I. databases, allowing speedier investigations and better integration of information both within the bureau and with other intelligence agencies that must coordinate national security matters.

 

While there were several software failures for 2005, I chose this one to highlight a point.  The very first sentence, I did not get what I envisioned is a telling statement that demonstrates a fundamental reason why software projects fail.  There is an incredible amount of information written about why and how projects fail that has been well documented for dozens of years. The following reasons for software failure are a summary: poor user requirements, poor project management, no change control, too much complexity, unskilled staff, political issues, poor software development processes, and on it goes.  So if these reasons are well documented over the last 20 years, why do we still have failed software projects?

 

I would say we have a more fundamental problem than all of these reasons combined and that is we have no way of visualizing the finished software product before it is built.

 

In other industrialized engineering disciplines, we have many ways of visualizing the end product without actually building it first.  For example, models are used extensively to visualize cars, buildings, cell phones, iPODs, bridges, airplanes, trains, you name it and somewhere there is a model that visualizes what the finished product looks like.  The model may be an artists rendering or a precision built to scale model of an airplane.  The fact remains that no matter what the product is, there is a visualization of some sort that people can look at as a concrete vision of what the end product looks like before it is built.

 

What do we have in the software world to visualize what the finished software is going to look like?  Not much.  Some people would call prototyping or screen shots or using Visio as a way of visualizing software.  Unfortunately, none of these represent what the actual finished software is going to look like.

 

As a software professional, whats most embarrassing about the VCF failure and other failures is that 95% of all business software applications are nothing more than retrieving data out of a database, presenting it to a human, who may manipulate it and saving it back to the database. Thats it.  We do it over and over again. No matter if you are SAP or salesforce.com or a custom application in the business world, all that is being done is retrieving and saving data to a database, yet billions are spent each year doing this and if you believe statistics, over 50% of these projects will fail.

 

It would seem to me that the software industry is ready for some industrialization lessons learned in other engineering disciplines, with the first lesson being lets visualize what the end product is going to look like before we build it.  Here is an excellent example of a visual model that was created 100 years ago to visualize an immense and complicated structure.  To me, this model puts into perspective how little we have evolved in the software industry, with all of our technology we still cant build software right.  Maybe next year

If you have read this 7 part series, you will see how our DSL Visual Designer works for declaratively defining complete application integration models and see how our Software Factory works by taking the DSL Visual Designer tools output (an XML model definition) and using it to configure our Software Factory template, which is a set of scripts that configures a Visual Studio solution, including BizTalk server as the core integration engine, to code generate complete application integration solutions.  Our framework design pattern can be applied to any application integration scenario.  Further, I assert that the Software Factory approach can be applied to any software problem domain just as successfully as we have applied it to the application integration domain.

 

Our DSL Visual Designer is very specific in the sense that its domain language is specific to application integration scenarios.  We built our Visual Designer on top of BizTalk Servers integration engine to provide a higher level abstraction so that Business Analysts (BAs) can easily model their specific integration scenarios as BAs typically have the business domain knowledge and just need a modeling tool, specific to that business problem domain, to model the solution which is declaratively complete.  Since the model is complete (i.e. high fidelity, meaning not missing any pertinent information) and that we are in the virtual world of software, we can code generate the entire output, meaning producing the solution in a predictable and repeatable way.

 

Thats the main point of software industrialization.  As mentioned in the introduction to this series of blog posts, software industrialization means raising the level of abstraction to producing software solutions in a predictable and repeatable manner.  Producing software solutions in a predictable and repeatable manner is the bane of our industry.  In our case, using the Software Factory approach enables us to use model-driven development techniques using Domain Specific Languages (DSL) and assemble pre-built components based on a domain model pattern specifically designed for modeling application integration scenarios.

 

Because we designed and constructed over 25 application integration scenarios by hand over a course of 4 years we saw many application integration patterns based on producing these solutions for several different industry verticals and different types of integration scenarios (i.e. basic messaging exchanges to complicated, recursive workflows, all using orchestration).  In fact, in almost every project we were seeing these patterns repeat themselves over and over again.  Based on these repeating patterns (i.e. our evolution), we saw how we could embody these design patterns into a modeling tool (i.e. our DSL) and take the modeling tools output (XML definition of an application integration scenario) and configure a Software Factory to code generate the output, producing an installer package that can be easily installed on the target system, including a number of runtime applications that can monitor and manage the installed application integration solution.

 

I personally believe that this Software Factory approach will be the way that most (business) software applications will be designed (i.e. DSL models) and constructed (i.e. code generated) in the future.  Why do I say future?  Why has this approach not caught on (yet)?  This blog has discussed many themes to the reasons why.  Mostly it boils down to education.  I work with developers today that do not see the value in interface-based programming let alone the value of model-driven development or code generation or the concept of a Software Factory.

 

The education/experience gap in our own software engineering community is quite large and when some programmers (new and seasoned alike) are introduced to these concepts well, lets say that there is considerable resistance to even thinking about using this approach.

 

Now throw into the mix the vendor community that insists on producing marketecture. Our very own software vendors are doing us a disservice by selling hype to our target business community.  Now add the business community to the mix, who are the target audience of this marketecture to most, software is a complete mystery.  All they know that it is an incredible time consuming, costly, painful process that mostly results in never ending work, endless upgrades and broken promises.  Unfortunately, given the current state of the art in the software industry, this is all but true.

 

With a few exceptions, I consider the software engineering industry to be still in the dark ages, yet to go through the industrialization process that other, more mature, engineering disciplines have evolved to.  This series of posts is intended to show, in one companys way, how we evolved the industrialization of software in the application integration domain using modern software engineering techniques and tools.

Parts 1 through 5 discussed the details of our application integration framework design pattern.  A Business Analyst (BA) uses our DSL Visual Designer (as a RIA) that has our framework design pattern implemented to declaratively model specific application integration scenarios.  There is one optional point in the declaration where a small amount of custom code may be required and that is if a custom application adapter is required.  Other than that option, any application integration scenario, regardless of size or complexity can be 100% declaratively modeled using our DSL Visual Designer.

 

This goes directly to the point of why model driven development using the Software Factory approach works which in my opinion, represents the state of the art in software engineering today.  My intent in writing Parts 1 through 5 was to walk through how our application integration framework pattern can be used to model any application integration scenario.  The output of our DSL modeling tool is an XML definition that declaratively defines a specific application integration solution as defined by a BA and validated by our Software Factory schema (i.e. our framework model that we just walked through in parts 1 through 5).

 

Next I will discuss our Software Factory in detail and how it works based on a particular application integration definition.  Have a look at this logical view of our Software Factory.  It contains 3 subsystems, they are: the DSL Visual Designer that we just walked through, our Software Factory, and our runtime system.

 

You already know what the DSL Designer does.  The Software Factory takes the DSL Designer XML tools output and configures our Software Factory template.  This means that all of the properties that were set by a BA in the DSL Designer are interpreted by our Visual Studio script and populates a Visual Studio solution with pre-built code artifacts, including source and destination XSD schemas, adapters, business rules, XSL maps, any custom assemblies uploaded by the customer, source and destination namespaces, BizTalk orchestrations, and our pre-built runtime applications are added to the solution.

 

Once the Visual Studio solution and projects have been populated with the specific XML configuration information, a deployment project is added to the solution and the solution is built and packaged as a .msi, ready for deployment.  The Customer is then notified that the package is ready to be downloaded, the Customer downloads the packaged solution and runs it on the target server. During installation, we ask the Customer for credentials for the solution to run along with the source and destination addresses of where the applications to be integrated exist on the network. In our .msi, we check for the existence of BizTalk Server, SQL Server, IIS, etc., to ensure all of the services are installed and running correctly.

 

Configuring and extending Visual Studio is a relatively simple process as the Visual Studio extensibility model is well documented and provides for easy automation.  After designing and developing 25 BizTalk integration solutions, setting the same old properties, adding pre-built orchestration schedules, schemas, etc. is easy.  Sure, some projects are more difficult than others and a little manual manipulation may be required, but generally the time it takes to configure and build an application integration solution using the Software Factory approach is a matter of days instead of months, and thats the point of this series of posts.

 

Parts 1 through 6 have been captured in a Live Meeting demonstration that can be viewed online.

 

Next post we wrap up our discussion on the evolution of software industrialization.

In part 4, we discussed our application integration framework design pattern at the level of the Connection object.  A Connection represents a logical connection to each Application, and physically to the Applications Adapter.  A Connection contains 1 or more DataFlows.  A DataFlow is a collection object that contains 1 to 4 schemas (i.e. XSDs), 0 or more rules, and 0 to 2 maps depending on DataFlow communication type.

 

With respect to rules, for anyone that has used Microsofts Outlook Rules and Alerts functionality will recognize our GUI Rules Designer.  As mentioned in another post, its been done before, you just have to go look for it .  The Rules Designer contains a Rules Manager that allows the BA to add, edit or delete rules in an ordered list, a Rules Editor that allows the BA to declaratively make rules, A Fact Editor that allows a BA to declaratively name a rule, and A Rule Expression Editor that allows the BA to define expressions, and if required, provides the facility to upload custom rules (i.e. .NET assemblies).

 

This diagram shows a screen shot of our Rules Designer:

 

 

At this level we also provide the capability for BAs to upload test XML documents that can be validated against the specified schemas.  The BA can also test the (XSL) map for proper transformations.  Any errors for schema validation and/or map transformation are returned to BA in the Designer.  In fact, if the BA provides a valid XML document with test values that represent what the real values are likely to be, business rules can also be executed, in which the return values are the resulting actions that would occur for the rule processing.

 

Thats it.  The BA has now declaratively configured a complete application integration scenario using our DSL Visual Designer tool that implements our application integration framework design pattern.  As mentioned before, our Visual Designer is exposed over the internet as a Rich Internet Application (RIA).  This allows a BA easy access to their application integration solutions from anywhere in the world, in a secured manner.  This approach models the software as a service scenario where there are no requirements for an organization to use the Visual Designer other than a browser.

 

Now what?  After the BA has completed defining the application integration model and saved the model, the BA can now order the integration definition to be built (i.e. code generated) from our Software Factory.  The order includes pricing based on the number of application integration connections modeled in the scenario.  Once the order has been placed, the turnaround time could be a few hours to a few days, depending on options ordered.  The BA then returns to the Visual Designer web site and can download the resulting MSI to be installed on their target system.  Once installed, the BA and/or System Administrator can run a self-test to verify and certify the application integration solution for correctness. 

 

The runtime installation also includes other tools to manage and monitor the application integration solution.  This includes Business Activity Monitoring for monitoring the application integration scenario, a Centralized Exception Manager that handles any runtime messaging exchange exceptions, complete with notifications, and an Integration Manager that allows runtime configuration of security credentials and application end point locations.

 

With respect to making any changes to the application integration scenario, the BA can launch the Visual Designer again, load up the existing model, that has been versioned, make modifications (to a schema for example) and save, then order and get another MSI that performs the upgrade process on the installed application integration solution.

 

Next post I will discuss our Software Factory for taking application integration scenarios based on our framework design pattern (as realized by our DSL Visual Designer and output as an XML definition) and walk through the steps on how we configure our Software Factory template to code generate the specified solution.

 

In part 3, I described our application integration framework pattern at the level of BusinessFlows.  A BusinessFlow contains 2 or more Applications and 1 or more Connections.  I described how the framework pattern works with Applications and in this post we discuss Connections between Applications.

 

A Connection represents a logical connection to each Application, and physically to each Applications Adapter.  A Connection contains 1 or more DataFlows.  A DataFlow is a collection object that contains other objects, which we will describe in a minute, but its construct allows us to contain one to many DataFlow rows.  Each DataFlow row specifies the direction in which the communication of messages (i.e. data) flow.  It also specifies which Application in the integration scenario has initiated the event in the communication, either in real-time or on a scheduled (i.e. batch) basis. 

 

For example, if I have MS Great Plains on one end of the connection and the Corporate HQ Financial Application on the other end of the connection, I can specify that MS Great Plains will be the initiator of the business event and the messages (i.e. financial data) will flow from MS Great Plains to the HQ Financial Application.  Further, I can specify the DataFlow row type of communication.  For example, one-way asynchronous, one-way asynchronous, two-way request response synchronous, and two-way request with delayed response (i.e. synch on aysnch).

 

As mentioned above, a DataFlow contains a collection of objects, specifically, 1 to 4 schemas (i.e. XSDs), 0 or more rules, and 0 to 2 maps.  A map is an XSL transformation, using a visual editor that allows a BA to map from one schema format to another.  Anyone familiar with the BizTalk Mapper tool will recognize our visual editor as the same representation, except we expose our Mapper tool over the internet in our Rich Internet Application (RIA) Visual Designer.

The reason for 1 to 4 schemas is that depending on the communication type and whether it is a pass-thru or not with respect to message flow.  For example, in our MS Great Plains to Corporate HQ Financial Application, we have specified 2 message schemas, one for the Great Plains format and the other for the HQ Financial Application, in a one-way asynchronous dataflow.  We also have 1 XSL map that transforms the source Great Plains message format into the destination HQ Financial Application format.  Further we may have specified certain rules to be executed either before and/or after the mapping transformation.

 

As mentioned above, our DSL Visual Designer is exposed over the internet as a Rich Internet Application (RIA).  We also provide a facility for a BA to upload any custom message schemas that have developed.

 

In my next post, we will discuss the Rules Designer, which is important to BAs and their business as this is where any data processing of the data occurs.

In part 2, I started describing our framework design pattern for defining application integration scenarios.  At the top level is our Model object that contains one or more Business Entities.  Each Business Entity can also contain one of more Business Entities to represent any hierarchical business entity structure that occurs in the real world.  Each Business Entity contains one or more BusinessFlows.  A BusinessFlow represents Business Processes which can be a simple A2A integration or a complex workflow process.  With respect to our DSL Visual Designer, each object in the model includes a wizard and/or property sheet for the Business Analyst (BA) to enter values that describe the object being configured.

 

Looking at the next level in the object model we can see that a BusinessFlow contains 2 or more Applications and 1 or more Connections.  An Application object represents a real-world computer application.  We have a number of Application objects pre-built (see Application Types in the model) allowing the BA to select from a list. If the Application required by the BA is not in the list, then a custom Application is entered which will require custom code to describe its interface (i.e. Adapter).  Each Application has an Adapter.  An Adapter is a plug-in connector where one end of the connector knows how to natively communicate with the Applications API and the other end exposes an interface that has been transformed into XML messages (including XSDs) that the Application can exchange with other Applications.
 
In our continuing example of financial applications exchanging data, the Application may be MS Great Plains which comes with a COTS adapter (i.e. connector) that performs this transformation in communicating with Great Plains native API and exposes an XML messaging interface.  Since we are using BizTalk as the underlying messaging engine, there are many COTS adapters available that provides this connector and transformation mechanism.

In the case of a custom Application where a COTS adapter is not available, we use the provided BizTalk Adapter SDK to write custom code to build an adapter.  Note this is the only variable point in the entire application integration scenario that requires custom code to be written, that is, if a COTS adapter is not available.  The custom code is limited in scope as we are extending a pre-existing adapter framework supplied by the middleware vendor for a specific Application in which we will implement only the messaging interfaces required for the application integration scenario.

 

This is a key point to keep in mind as everything else in the application integration scenario is defined declaratively.  Since everything else is declarative, this dramatically reduces the amount of custom code that has to be written for any given application integration scenario.  In fact, even with a custom adapter, this means that the entire application integration definition is syntactically and semantically complete, which means we can code generate the entire solution and the main reason why the Software Factory approach works: model-driven development, which dramatically raises the level of abstraction for developing application integration solutions to the point where a BA declaratively defines the complete solution through a modeling tool (i.e. our DSL Visual Designer) to be generated.

 

Using model-driven development, we can define a model up-front that is both complete and of high-fidelity. We can then validate the model against a known schema and generate the output.  Our application integration framework design pattern is this schema (i.e. Software Factory schema).  Using our DSL Visual Designer, the BA models a specific application integration scenario in which the Designers output produces a valid XML document definition that describes that specific application integration scenario.  The XML definition is used as input to our Software Factory template which configures and extends Visual Studio design time components and reuses our pre-built components that are checked out from our source control system.  This includes configuring BizTalk Server and our factory template includes the instruction set (i.e. scripts) to build the complete solution into an installable package (i.e. MSI package) which is then installed on the target system.

 

Returning to our framework design pattern, a Connection object represents the actual connection between Applications and will be the topic of my next post.

In part 1, I described my experience using Software Factories, including DSLs, to predictably and repeatedly solve application integration problems.  Our Software Factory framework pattern evolved over the course of 4 years and 25 application integration projects.  Essentially we developed a product that encapsulated our framework pattern that could solve any application integration problem with minimal coding.  This was done by analyzing what is common and what is variable to every application integration scenario and designing a framework model to cover these scenarios.  My intent here is to cover each aspect of the model and discuss how and why we arrived at our overall application integration framework pattern.

 

Referring to our framework object model, the top most container is the Model object in which it contains all of the objects that make up one or more application integration scenarios.  Essentially it is an abstract (root) container, which translates to the top-level drawing canvas in our (DSL) visual designer modeling tool.

 

The first common modeling construct is that all application integrations occur within a Business entity or between Business entities.  A Business entity is an abstract object that can represent an individual company, a division of a company, a remote office, an enterprise or small group within an organization.  A Business entity may also contain other Business entities.

 

It seems obvious or even taken for granted that a Business entity models a real world construct (i.e. a real business).  However, in my experience this construct is typically either overlooked or viewed as trivial.  The reality is that application integrations do not occur in thin air, they are tied to real businesses, thus part of the problem domain we are interested in modeling.  Having this construct allows us to model any application integration scenario within or between any number of uniquely named Business entities.  For example, lets say we have 50 satellite offices that are geographically dispersed that use various accounting software packages that need to exchange financial data with a centralized financial package (i.e. Corporate HQ). 

 

Without the capability to model this in a visual designer would make it very difficult to design and construct this application integration scenario, let alone actually visualize semantically what is actually going on. While we are talking about the business entities and the logical connections between the business entities, we are also describing the many properties of each unique business entity and its relationships to each other.  This translates into unique name spaces required to define the paths between (or within) these businesses in the models XML definition output.  In other words, we have declaratively captured all of the information required for each business and to navigate from one business entity to the next.

 

Looking at the framework design pattern at the next level, we can see that each business entity has one or more Business Flows.  A Business Flow is an abstract object contained in a collection whose parent is a Business entity. The Business Flow represents a particular Business Process which may be a simple A2A integration, or may be a complex workflow process.  For example, if one of our satellite offices (i.e. business entity) had three financial applications that exchanged data between each to form an aggregation, before it sent its aggregation to the centralized financial package, then we need the capability to model this. 

 

You can see an example of what this visually looks like in this layered view of our visual designer model at the canvas layer called Business Flow.  You can see how this visual representation conforms to our framework design pattern.  You can also see how this visual representation can scale to represent any number of Business Entities and the collections of Business Flows for each Business Entity.  While this occurs in the real world, few application integration vendors toolsets have the capability to capture this critical information.

 

In part 3, we will continue walking through our application integration framework pattern.

This blog is dedicated to the advancement of what I call the industrialization of software.  Software industrialization is about raising the level of abstraction for programmers to produce quality software products using modern software engineering approaches such as Software Factories. One aspect of the Software Factory approach employs model-driven development using Domain Specific Languages (DSLs) that code generates parts (or all) of the solution for a specific problem domain.  The Software Factory approach enables programmers to produce and assemble software in a predictable and repeatable manner something we dont do well as an industry, but is done well in other industrialized engineering disciplines like civil, electronics, electrical and mechanical domains.  Or at the very least, software engineering is nowhere near as evolved as these engineering disciplines.

 

The author of this blog lived the Software Factory approach while working at a company called 5by5 Software, now called Bridgewerx.  5by5 was initially a Microsoft Systems Integrator (SI) that offered professional services in the area of application integration, designing and constructing solutions using Microsofts BizTalk Server product.  Over a 4 year time frame, we designed and constructed over 25 application integrations solutions.  These solutions included messaging exchanges and workflow of all sizes, shapes and forms for various vertical industries.

 

During this 4 year time frame, we recognized several application integration design/code patterns.  Each pattern went through the process of discovery, reuse, refinement, and then iterated through this process again for each new project we undertook.  After designing and constructing 25 BizTalk integration projects, we found an overall framework design pattern that could be applied to all integration projects.  We decided to encapsulate this design pattern in a modeling tool, which really is a Domain Specific Language (DSL) for designing and generating application integration solutions sitting on top of BizTalk Server. 

 

The idea is that the DSL runtime Visual Designer would be used by a Business Analyst, who has the business domain knowledge to model an application integration scenario and the modeling tools output (XML) would supply the complete definition of this specific modeled application integration. We used this complete definition as input to our software factory template which then grabbed reusable components, other artifacts (with some custom code) and configure BizTalk Server with the specific application integration scenario, automatically build the solution in Visual Studio and produce a Microsoft installer package (MSI) which then can be installed on the customers target machine.  We called our Software Factory and DSL, Bridgewerx.

 

The point is that the software factory approach, incorporating DSLs works.  Our 4 years of application integration domain experience and subsequent ISV product is living proof of this fact.  We were pioneers in producing a software factory for the design of and code generating application integration solutions.

 

To further substantiate this claim, I will walk through a number of our projects and describe the patterns we discovered, designed, codified, categorized and subsequently automated, until we had the entire application integration framework pattern designed. Then we encapsulated the framework pattern into a DSL and built our software factory for code generating application integration solutions. 

 

From where I sit, this is industrializing the software development process for the application integration domain.  Another claim I make is that this same approach can be successfully applied to other software application domains of interest (e.g. software factory for generating families of e-commerce applications.)

 

In part 2, we will discuss what is common and what is variable to every application integration scenario as a way of describing our framework pattern.  Then we will look at some specific integration projects we undertook and how it fits the framework pattern.