As Cloud Architectures Grow Up, Migration and Portability Will Be Key Issues

We have been working on a major upgrade project for an application that was written more than ten years ago.  Written in the era of the “latest” technologies from Microsoft such as InfoPath 2003, JScript, SharePoint 2003, ASP.NET, Windows Workflow Foundation, etc. we have been working to upgrade this entire application because all of these components are now end of life from a support perspective.

Image result for azure application architecture

We are now in the 2003 era for cloud – Microsoft, Amazon, Google, etc. are introducing new services and technologies at a blistering pace with the promise of making line of business applications easier to develop, manage and maintain.  As an industry, we have been sold on the same messages for years:

  • Why build something custom when you can use “out of the box” and configure?
  • Why have developers when you can have power-users who can self-service their own maintenance?
  • Why design your own architectures for scalability, security, etc. when you can rent our already industrial strength solution?
  • Why own your own code, platforms, etc. when we’ll allow you to rent ours at a fractional cost?
  • Why have a simple, monolithic application when you can try dozens of new services optimized to be best of breed?

As we fast forward five to ten years from now, there is a lesson from our experiences maintaining, upgrading and re-platforming applications from years ago. 

Migration matters.  Upgradeability matters.  Vendor lock-in matters. 

When upgrading this particular application, the custom code we developed upgraded without any problem.  Take the 10 year old code, put it into Visual Studio 2015 and build it and it builds.  Hosting the code in a basic web container like IIS also works without a problem.  However, when we start upgrading higher level platform components like InfoPath, SharePoint or Windows Workflow Foundation, Microsoft has “evolved” (e.g. broken backward compatibility) their platforms over the past 10 years and the “configuration” based application components break.  We’re now in the process of fixing these components and re-writing them in some cases to adopt the latest Microsoft standards.

Microsoft is no better or worse than any other vendor – we see similar lessons with Oracle, SAP, IBM, etc.  We will see the same issue with cloud vendors like Amazon, Google, Salesforce.com.  It is in the vendor’s interest to ensure that your application is tied to their architecture components.

Migrating to the cloud is easy.  Upgrading to a new cloud, migrating off the cloud or picking a new cloud vendor could be very hard indeed.

Lets imagine our organization has a line of business application that processes orders coming from the web site to our CRM system and then into our financial system.  We also want analytics so we also invest in some big data services and business intelligence tools.  In the Microsoft world, we might use ASP.NET, Azure Search, Flow, Power Apps, Dynamics CRM, HD Insight, Power BI, etc.  In the Amazon world, we might use Amazon Web Services, RedShift, Hadoop, etc. and maybe we use SalesForce as our CRM system and DOMO as our business intelligence solution.

Regardless of the platforms you pick, each introduction of a new component creates a dependency on the application that increases your migration risk, the complexity to make changes and locks you into the vendor providing you the service.  As the cloud architects advocate for micro-services architectures that split your applications into potentially dozens of services the supply chain of code for your application will become increasing complex.  While there are benefits to such architectures, tracking the dependencies and ensuring all these services work together now and into the future will become a significant challenge in the cloud era.

Avoiding future nightmares by learning from the past.

In reviewing upgrades to legacy applications, we can learn lessons that apply equally well to cloud.  Our team does many of these projects where we review, analyze and re-platform applications that were built decades ago.   Here are is what we have found:

  • Vendor lock-in is a huge challenge and creates a massive migration risk.  If you decide you don’t want to do business with Microsoft anymore because ABC is a cheaper alternative, do you have a plan to move all your data, your customizations, all the configurations, etc. to a new platform?  Microsoft, SalesForce, Amazon, etc. are all betting that in five years from now it will be more difficult to migrate than to pay the increased fees that they will inevitably charge.
  • Pure code solutions tend to upgrade easier than “configuration” based solutions.  My C++, Java or C# application can be ten years old but I can still compile it, run it and in most cases upgrade it to the latest frameworks.  These programming languages are super mature and backward compatibility is very high.  Be careful with high level “power-user” solutions that promise an easier but technically less robust solution.
  • Transparency and portability of code and data is an important principle.  For example, if I adopt Azure SQL as my database, I can still access my data reasonably easily.  I can also export it, move it around, take it back on premise, and see it as a set of well understood tables.  Can I say the same thing if my data is in SalesForce?   
  • As the number of platform components increase, the challenges with dependency management increase as well.  While your architecture could leverage dozens of cloud services to optimize the operation and hosting costs of your application, what is the cost of the increased complexity when you try to upgrade?
  • Understanding the maturity and commitment to maintain backward compatibility and support your customization model is a key requirement.  With the cloud accelerating the product development life cycle dramatically from 3 years to every 15-30 days, cloud vendors are going to threaten your architecture every 15- 30 days if they don’t maintain backward compatibility or provide an upgrade path.  What happens in 3 years if Microsoft decides that they don’t want to support Flow or Power Apps anymore? 
  • While there are some performances and scalability advantages to introducing micro-services, leveraging distributed architectures, harnessing specialized cloud services, etc. their introduction also increases complexity and dependencies.  Most organizations are not Facebook and could build a reasonably focused and robust application using a simple web server and a database only.  While this may sound old-fashioned and harder to maintain than a more sophisticated architecture and operational model, it means upgrading and maintaining the application will be easier.

Past technology revolutions show the way forward – those who invested in new fangled technology solutions a decade ago made some bad bets and bought into technologies that over-promised and under-delivered.  Applications that are simple, leverage non-proprietary and mature technologies and that are easy to migrate over time provide longevity.  Cloud services will follow the same path, except perhaps faster and with more complexity. 

Do you have an architecture that be easily maintained, upgraded and re-platformed five to ten years from now?  As you migrate to the cloud, is their a plan to avoid the same lock-in, migration and portability challenges we experienced on premise?