Bruce Allen

Cloud hosts make it relatively easy to manage data centrally while distributing it across the globe for regional access. Edge locations can be synchronized with the principal data repository via configuration, a no-code solution. Some cloud providers offer data caching co-located alongside 5G cellular towers to optimize responsiveness in smartphone applications. Your technical account manager can help you navigate the myriad options; just provide details about how your data is created and accessed. Unlike on-premises, if your needs or usage patterns change, you can switch data services with no capital outlay for equipment. This is a key value of the cloud that should not be overlooked.

Major cloud providers offer price/performance flexibility to suit different scenarios. One example is historical data retained exclusively for regulatory compliance. These data are retained for years and may never be accessed. In this situation, cold storage such as AWS S3 Glacier can be cost-effective, especially when you consider the service's built-in redundancy and backup. Cloud storage billing is generally based on 1) data size, 2) performance, 3) write I/O, and 4) read I/O, and 5) contract discounts. Extreme I/O can make on-premises storage the best choice. However, near-instantaneous storage provisioning makes it possible to build in the cloud first and change later, optimizing for observed access and storage patterns.

Approaches to protecting data using access controls and encryption depend upon factors such as whether that data must be accessible from outside the cloud host. For example, data generated by a cloud app may never need to leave the cloud, allowing you to use provider-specific options that can be simpler to set up and manage. The opposite extreme includes hybrid environments where the variations among AWS, GCP, and Azure introduce more options, limitations, and complexity. Invest time to understand your options and keep things as simple and consistent as possible. Runbooks are your safety net and should be kept updated as you build and maintain your data reservoir.

By definition, cloud computing adds to a technically complex domain an extra layer of abstraction. Actively watch and listen for anyone who seems to not fully understand the topic(s). Follow up after meetings with individuals who might be hesitant to speak up among other stakeholders. Putting structured feedback mechanisms and timings into a series of meetings is a best practice to keep communication flowing between sessions. Cloud is not unique in how to collect and respond to feedback. However, valuing feedback keeps all participants engaged and helps ensure that decisions made are understood by the whole team.

Begin by analyzing needs of the business and severity of potential damage to operations and business reputation. Business critical workloads require defining the RPO and RTO to plan for resiliency and recovery from disasters caused by man, nature, or Murphy's Law. Prioritize and invest in your most essential systems and keep recovery run books as simple as possible. Discuss worst-case scenarios and plan for each. Get agreement from the business on the above to drive the inevitable cost/benefit choices of options per system. Research and consider the cloud provider's built-in redundancy and fail-over capabilities across geographies. Plan for vengeful attacks that corrupt data as well as temporary outages.

Breaking apart monolithic solutions into discrete services allows scaling each portion individually rather than the entire application reducing runtime costs. Identify the hot spots and cold spots in current workloads and consider different hosting options for each. Look into existing cloud services that could replace functionality originally written inside your app code. Be willing to build and learn iteratively even if that is new to your team. Be prepared for the areas of complexity to move from tight dependencies in a monolith to orchestrating many loosely-coupled micro-services. Avoid the trap of avoiding analysis before starting to code; do not presume a "direct port" will produce quick and lasting results.