If you’re a returning reader to my series, thank you for reading this far. We have a couple more posts in store for you. If you’re a new visitor, you can find previous posts below:
Part 1 - Introduction to Hybrid IT
Part 2 - Public Cloud Experiences, Costs, and Reasons for Using Hybrid IT
Part 3 - Building Better On-Premises Data Centers
Part 4 - Location and Regulatory Restrictions
In this post, I’ll be looking at how I help my customers assess and architect solutions across the options available throughout on-premises solutions and the major public cloud offerings. I’ll look at how best to use public cloud resources and how to fit those to use cases such as development/testing, and when to bring workloads back on-premises.
In most cases, modern applications that have been built cloud-native, such as functions or using as-a-service style offerings, will have a natural fit to the cloud that they’ve been developed for. However, a lot of the customers I work with and encounter aren’t that far along the journey. That’s the desired goal, but it takes time to refactor or replace existing applications and processes.
With that in mind, where do I start? The first and most important part is in understanding the landscape. What do the current applications look like? What technologies are in use (both hardware and software)? What do the data flows look like? What does the data lifecycle look like? What are the current development processes?
Building a service catalogue is an important step in making decisions about how you spend your money and time. There are various methods out there for achieving these assessments, like TIME analysis or The 6 Rs
. Armed with as much information as possible, you’re empowered to make better decisions.
Next, I usually look at where the quick wins can be made—where the best bang for your buck changes can be implemented to show return to the business. This usually starts in development/test environments and potentially pre-production environments. Increasing velocity here can provide immediate results and value to the business. Another area to consider is backup/long-term data retention.
Development and Testing
For development and test environments, I look at the existing architecture, are these traditional VM-based environments? Can they be containerized easily? Is containerization where possible a good step toward more cloud native behavior/thinking?
In traditional VM environments, can automation be used to quickly build and destroy environments? If I’m building a new feature and I want to do integration testing, can I use mocks and other simulated components to reduce the amount of infrastructure needed? If so, then these short-lived environments are a great candidate for the public cloud. Where you can automate and have predictable lifecycles into the hours, days, and maybe even weeks, the efficiencies and savings of placing that workload in the cloud are evident.
When it comes to longer cycles like acceptance testing and pre-production, perhaps these require a longer lifetime or greater resource allocation. In these circumstances, traditional VM-based architectures and monolithic applications can become costly in the public cloud. My advice is to use the same automation techniques to deploy these to local resources with more reliable costs. However, the plan should always look forward and assess future developments where you can replace components into modern architectures over time and deploy across both on-premises and public cloud.
As I mentioned, the other area I often explore is data retention. Can long-term backups be sent to cold storage in the cloud? The benefits offered above that of tape management for infrequently accessed data are often prominent. Restore access may be slower, but how often are you performing those operations? How urgent is a restore from, say, six years ago? Many times, you can wait to get this information back.
Continuing the theme of data, it’s important to understand what data you need where and how you want to use it. There are benefits to using cloud native services for things like business intelligence, artificial intelligence (AI), machine learning (ML), and other processing. However, you often don’t need the entire data set to get the information you need. Look at building systems and using services that allow you to get the right data to the right location, or bring the data to the compute, as it were. Once you have the results you need, the data that was processed to generate them can be removed, and the results themselves can live where you need them at that point.
Lastly, I think about scale and the future. What happens if your service/application grows beyond your expectations? Not many people will be the next Netflix or Dropbox, but it’s important to think about what would happen if that came about. While uncommon, there are situations where systems scale to a point that using public cloud services becomes uneconomical. Have you architected the solution in a way that allows you to remove yourself? Would major work be required to build back on-premises? In most cases, this is a difficult question to answer, as there are many moving parts and relies on levels of success and scale that may not have been predictable. I’ve encountered this type of situation over the years, usually not to the full extent of complete removal of cloud services. I commonly see this in data storage. Large amounts of active data can become costly quickly. In these situations, I look to solutions that allow me to leverage traditional storage arrays that can be near-cloud, usually systems placed in data centers that have direct access to cloud providers.
In my final post, I’ll be going deeper into some of the areas I’ve discussed here and will cover how I use DevOps/CICD tooling in hybrid IT environments