In the last post I set the context for the HSVO project and how I winded up a member of it via my contract at NRC, setting the stage to help any potential readers understand all the partners and the objectives they had going in.

National Research Council of Canada, Institute for Information Technology (NRC-IIT) in Fredericton, NB (Photo credit: Wikipedia)

Now I’ll talk about the major problems faced on the project (also suffered by Healthcare IT & Medical Schools all over the world and in any shared E-Learning initiatives in general). Hopefully, any time you need to have so many partner systems work together based on the anecdotes shared here, you’ll quickly realize the importance of establishing an Electronic Data Interchange (EDI) amongst your partners, specifically, a simple to parse Canonical Data Model (CDM). These constant communication and data formats must form the foundation for any cross-organizational initiatives like this involving high levels of integration between two or more devices, services, systems, apps and/or courses.

How we did it

NRC management was essentially the lead on the project as far as owning the largest piece being delivered (i.e. their mission assigned by the HSVO was to “provide the glue that would join all these separate devices being built by our research partners”). After about 6 months of initial research, the plan formulated by my superiors before I was even hired on by NRC as an Application Developer to implement it, was to extend their existing SAVOIR 1.0 source code. It is common for government organizations like NRC to look to past work and projects/experiments from within their portfolio which can potentially be leveraged or extended beyond their initial expiry date and use cases. Nothing wrong with that, but its not always the best fit, especially when dealing with complex problems which might be better served by green fielding a totally new solution based on Agile development using direct interaction with the customer (in our case, our research partners, but primarily NOSM who the entire project was tailored to in the name of the promises of “E-Learning for Remote Medicine”).

That said, SAVOIR 1.0 was basically a glorified Application Launcher which acted as a sort of dock that could run any application installed on your Operating System with a single-click. Think RocketDock but without any of the slick animations, and, implemented in Java so it could at least work cross-platform on Windows, Unix/Mac, or Linux. It was originally intended to simplify the lives of Architects as part of a separate project that finished up in 2008. Behind its code were also two basic Web Services:

• User Management (called “SAVOIR_UserMgmt“) this allowed the system to keep track of who was using the SAVOIR dock to run their applications and which applications they launched at which time. It also allowed system administrators to turn on an optional desktop-based login popup in Java if desired, to force specific users to login with their username/password before using the applications in the SAVOIR dock; however of course if the applications were installed outside of SAVOIR’s install pack, then those applications could be launched anonymously as usual at any time, by going directly to their “exe” file on the system, or clicking on a shortcut (such as in their Program Files menu).
• Session Management (called “SAVOIR_SessionMgmt“) this allowed a unique identifier to be attached to each running instance of the SAVOIR dock, regardless of whether or not anyone was actually logged in, or if the login feature was turned on by the system administrator for the network on which SAVOIR was running.

We delivered what could really be called SAVOIR 1.5 after approximately my first 6 months on the job, with some heavy modifications to the core Java Swing GUI including the ability to drag & drop to the dock or slide/move applications up and down in launch order, and, the ability to also launch particular websites/webapps in the browser for the first time. This browser launch feature was probably the one we were most proud of, and only became reasonably easy to do in Java 1.6+. Now, any link could be placed on the Dock, either of the form:

• file://path/application.exe (for locally installed apps)
• http(s)://domain:port/path/#hash?param1=abc&paramN=etc (for apps on the web)

This gave the SAVOIR launcher a great deal of flexibility, we even developed a couple canned demos of the new capability which we thought would delight our partners. By being able to popup a simple search window (also in Java) and navigate straight to deep search results of the CMA Guidelines Infobase, PubMed medical journal archive, Wolfram Alpha as a calculation tool, and Wikipedia as a general info resource, as per the user’s selection.

The reception was lukewarm at best though, which came to a bit of a surprise to me after all the hard work myself and the team had put in, doing exactly what our superiors had requested. What I began to learn was that there was a growing disconnect between what our partners (who these partners were is covered in my last post) were expecting and what we were delivering. What our partners were telling us, was that they did not simply want a customized version of SAVOIR 1.0 rather what they wanted was a more unique application with some intelligence that basically not just got out of the way so they could use their applications in a particular order, but they wanted something that could do alot more heavy lifting and facilitate communication between their separate applications.

A brief word on EAI
Enterprise Application Integration (EAI) is the dilemma we face when trying to integrate large complex applications (but the cocepts really apply to applications of any size). With two applications, the integration is easy. Expose one or both as an API and send data, one or two ways, as needed:

It quickly becomes tough to manage as you add in more applications to support/integrate:

If you have any kind of external scalability requirement to support alot of external services/devices, just forget about point-to-point integrations:

Of course the ESB vendors like MuleSoft, WS02, TIBCO and even Apache ServiceMix evangelists promise their ESB products will instantly make your EAI efforts look and feel like this:

Not quite. Even with proper use of JMS for messaging, HTTP for addressing, and an ESB to broker transformations, what we learned the hard way is that despite all the hype of the ESB providers, you can’t expect to just plug in the services to an ESB and walk away laughing, problem solved (its possible that certain members of our team who sha’nt be named bought this notion a little too much though). You simply won’t be able to avoid the need to gain at least a rudimentary amount of working knowledge of each of the given devices or APIs you want to integrate, before you can even try to do anything meaningful with the data inputs they send and outputs they can receive. What that also means is that you’ll need to gain some basic domain knowledge and/or work directly with a domain expert in the initial design stages (akin to Agile development). We lacked that domain expertise at NRC, and at first management also relucted on opening direct lines of communication between our developers and main clients (which from the broader group were NOSM and McGill). At first the reaction was all around frustration, then lots of talk of from the medical experts on our project needing something they were calling “The Eye of Sauron” followed by a rash decision to implement a Rule Engine (in our case Drools) which could live on the ESB to help determine the routing of messages based on business logic defined by our HSVO partners. We struggled with this, not just the implementation and integration of the Rule Engine as yet another endpoint on the ESB and whether that was helping us or hurting us in the long-run. Perhaps most significantly what we struggled with was how to simplify the Domain-Specific Language (DSL) Drools required as input down to a simple enough format that it could be either hand-coded in Excel (one of Drools’ supported input formats) then uploaded periodically before running E-Learning classroom simulations (aka “Scenarios” as the team dubbed them), or, to a consistent enough format that the rules could be automatically generated in the backend by an Authoring Tool (yet another piece of new technology that would have to be built to support this growing monstrosity). Next came discussions about the need for a “Rosetta Stone” behind the scenes of SAVOIR, which I codified as a “Term Dictionary” which was capable of mapping concepts between endpoints (i.e. Blood Pressure in one system may be the input “BP”, but in another system it might be “kPa”; in some it could be split by S-Systolic and D-Diastolic while others it could be together as “S/D mmHg”). Finally, a need for a “Unit Converter” to convert different medical units of measurement (i.e. metric to imperial to bridge gaps from Canada/UK/Ireland to United States, differences between Med School SOAP notes, etc). So much for an Application Launcher huh? We went through several student placements but none of them made a very significant dent in the Rule Engine (aka. Eye of Sauron) piece, or, the Message Translator/Mapper/UnitConverter (aka. Rosetta Stone); so it was up to myself and Roger Sanche at NOSM to come up with a way to get this beast working. Working with the domain experts daily we realized the main problem. Interaction in one application/service/device must be able to fire-off events in a predictable way to the Message Bus and get routed through to the next application/service/device yes, but we are doing this in support of multiple learners at multiple locations, this was the key. It was one of the hardest things I’ve ever done but after many late nights we pulled off a working integration that would show the recipe. We called it the SAVOIR SDK, mostly because we had spent our creative reasoning powers and were too exhausted come up with anything else.

What we realized is that if we could apply Sturgeon’s Law (aka. KISS, DRY, YAGNI) and simply start thinking of each of the complex devices we were trying to integrate as nothing more than differing set of inputs/outputs, with some unique endpoint “addressing needs” based on their own scalability and protocols (i.e. HTTP based web app that can support thousands of users at once, or, a UDP camera that only one person or classroom could use at a time) then we could start to see a different more manageable picture. Finally we could realize something similar to the ESB simplification we were promised; to get SAVOIR 2.0 anywhere near our partners’ desired level of integration, we needed to create just three simple CDM and thin wrappers (SDKs) for our partners to use to send and receive messages in a single consitent format. That simple realization finally got us the big breakthrough we needed, and a matching of Scenario (path to scenario being worked through), sessionID (learner unique identifier), and instanceID (learner’s device’s OS/browser/tab identifier) as attributes of the services we were describing would be the last step to the messaging delivery problem. I’ll be the first to admit, its such a simple concept. To this day I wish I’d bought the infamous EAI book sooner, but it was a chance landing on the book’s website that lead us down the right path. Check out the schema we put together:
Read the rest of this entry »