My thoughts on topic #2

The first point that jumped out at me from this week’s reading was the proposed roles of Solution Architect (SA), Application Architect (AA) and Enterprise Solution Architect (ESA) as part of an organization’s Enterprise Architecture (EA) function.  The explanation of the responsibilities, focus areas and deliverables makes lots of sense.  I can see the benefits of dividing the focus on a single applications architecture by an SA, generalized focus on multiple application architectures for a business area of the AA role and enterprise wide application architecture trends and planning focus of the ESA role. 

The issue I have with these definitions are not the definitions themselves.  I find it hard to believe that any organization would establish all of these roles at any given point of their EA program development.  The article did concede that smaller organizations would likely have individual staff performing multiple roles at once.  I think additional research would be valuable into how real organizations divide these roles among their staff.  I would be interested to see how the roles are divided at organizations of similar sizes. 

I would anticipate the results to show that larger organizations would implement the SA and ESA roles, due to efficiencies of scale due to their size supporting more specialization and needs to communicate ESA level concepts across more SA staff.  Smaller organizations would likely not be able to support so much specialization due to their size and communications would need to distribute across a smaller number of EA staff.  I would think smaller organizations would implement the AA role only.

A second piece of the reading that stood out to me was the Application Architecture Future Trend paper describing the experience of Amazon’s architecture development.  Beyond the overall picture presented describing the now trend setting approach of Amazon’s development of AWS, one arguably small detail caught my focus.  Amazon’s approach of developing all functions with APIs for use even by internal users appears to avoid all the weaknesses I have seen in my career in using this approach.

For many years I have seen this approach described as a best practice, but in practice, I have seen it very little.  Further, in my own personal experience, the approach never quite lived up to the theoretical benefits fully.  Primarily, the idea that providing an API interface insulates the participating endpoints from being impacted by each other’s changes never quite work out.  Two main shortcomings I have seen are

• Parameter changes to the API calls drive over time the need to change the calling application, in contradiction of the promise to prevent this
• Additional functionality and data returned from calls causes problems for calling applications when additional functionality and data are not needed

The personal revelation that I discovered in the paper involved the use of XML documents in the API calls.  Formatting parameters as XML documents enable more flexible means of calling the APIs that allow parameter changes without changing the API signature, which addresses my first issue.  And the concept of enabling the caller to provide an XSLT parameter to empower the caller to limit returned data to meet its needs addresses issues with data overload and efficiency was an eye opener.  I will be adding this to my own toolkit.

One big question that was raised for me came from the article “Time to Retire the 3 Tier Architecture”.  The title caught my eye partly because I have already come to the conclusion myself (I think the time is long past actually).  The other part was my own struggle to understand the new environment, specifically when it comes to the Data tier.  Replacing the UI tier due to the changes in UI devices seems easy to understand.  For the mid-tier, replacement also makes sense with the proliferation of SOA services, micro services, etc.  However, I have not seen a good reference architecture for the data tier.  I have some exposure to Big Data and the concept of Eventual Consistency, but not a good detailed architecture for how to implement either as a replacement for a good old fashioned self-contained relational DB.  One main capability I have not seen replicated in other approaches is speed.  Generalized services always seem to be much slower than relational DBs with any large amount of data, and the complexities of managing eventual consistency with all but the simplest data sets seems daunting.

My question is, has anybody found a good reference for what either of these data approaches would look like in a replacement for the traditional data tier?