Cloud platforms come with a variety of options and constraints certainly driven by infrastructure and business needs which indeed diversify their pricing plans and customer adoption as well. But often they also have an implicit value: designing with certain constraints in mind would facilitate scaling and replication, would provide performance gains and increased revenues. Are constraints such as request timeout or limits on outbound traffic and datastore usage, so bad after all? However, when designing for your private hosting, you might skip or ignore few of them and miss an opportunity to improve your service design. Here are few points worth to consider:
- Design for monetization: no matter which PaaS you would choose, it will always have constraints on network traffic, datastore connections, datastore space and so on, and that’s reasonable and obvious: they need to make a profit after all. When designing for the cloud you definitely need to face these constraints trying to minimize your expenses (review your data model, restrict required data, limit outbound traffic, adding cache mechanisms and so on), because you certainly want to increase as much as possible your monthly revenue. Designing for your private hosting you would not have taken so seriously some of these parameters perhaps, realizing later on their importance though.
- Design for performance: as part of the previous point, you might need to adapt your design to fit some cloud constraints and obtain a considerable performance gain which wasn’t directly linked to monetization (but it would impact it somehow after all). You need to limit your outbound traffic, your request time-out, the number of queries on your database, any long running process and so on: well, a bit of healthy pressure on these subjects would definitely not hurt your design.
- Design for latest technologies and standards: PaaS usually don’t support all existing technologies nor all frameworks/tools per technology, but they normally offer the most common ones, well known and widely used, because they obviously need to target a large community and facilitate your deploy (and their profit). These constraints may change your decision concerning build management, for instance, giving up on your well proven ant script and going for Maven on Heroku, or finally stop using a certain old-but-known library in favour of a more modern and effective one. That might have an impact on your roadmap though because of unexpected learning curves and you would not have faced this question on your nice private hosting, but it might be time to upgrade your competences while working on your project, soon or later you may appreciate that. And it could facilitate the integration of new team members in the future.
- Design for abstraction: the majority of PaaS would also require some platform/API dependency which would make your application cloud-platform dependent, something you would definitely try to avoid. Adding further layers of abstraction may solve the issue: you may use Memcache or Big table on GAE, for instance, but your service shouldn’t know that if you really want to keep a smooth portability. It’s an extra effort indeed which may affect the decision of choosing a target Paas rather than another, but that’s worth in case of future moves. You would probably not waste time and energy on your more open remote or private hosting, but you may end up later on in complex refactorings and wonder why a simple and clean abstraction wasn’t part of your initial design.
While you would reasonably ignore cloud constraints when designing for your private hosting, you might gain a quick added value just keeping in mind few of their constraints and challenge your design against them: how would your design react? Would it be easy to deploy your application elsewhere? Would your monetization improve adapting your design to any of them?
The exercise is worth the effort as long as it doesn’t sound completely artificial and you do see some potential added values. There are plenty of non functional requirements you might just have left out and possibly you might also consider to use a cloud platform for your final deploy. Pick up two or three of the most used ones (AWS, GAE, Openshift, Heroku, Cloudbees are worth to mention, but they are not the only ones) and dare your design for an hypothetical deploy.
|Reference:||How designing for the cloud would improve your service implementation from our JCG partner Antonio Di Matteo at the Refactoring Ideas blog.|