Cloud Optimization is the place!

So. you've arrived this far and embraced the cloud model by bringing a bunch (or thousands) of workloads into the public cloud.You have sweated through the assessment, preparation, test and migration phases and are now in control of all your operations in one single place. This is what cloud is about, right?

Well, yes and no.

Thing is, this is where the real work should start. Because you are paying those VMs and workloads and PaaS by the minute, it is mandatory (end ethical, from a footprint and environmental point of view) that your use of them is as accurate and precise as possible. Not only to save big bucks, but because then you can move more workloads and stuff into the cloud and increase your IT efficiency, since you can afford much more than what the original assessment and planning process told you.

Let me give you some numbers that come from a real life experience (a large corporation with thousands of employees, that moved their entire IT operations into the public cloud. You'd never guess who this is :) )

After the first workload migration this is what could be found:

• The average CPU usage was below 6% (yes you've read that right! and by the way this is industry standard).
• There was a common tendency to over-provision the size of a machine, which is typical of on-premise workloads since you have to plan and account for future growth.
• A staggering number of VMs was kept on with no need for it (think of weekends, nights and out of hours, and most of all, think of dev and test VMs that lay unused for long runs).
• Another incredible number of VM was not even _ever_ used! Those VMs were there from legacy departments and nobody knew what they were and so nobody wanted to take responsibility. Until you can prove that it hasn't been used in the past x months/years.

These thoughtful discoveries, as you can imagine, prompted a series of actions that generated a proper cycle of cloud optimization, that can be summarized as follows:

1. Identification. Looking at recommendations and virtual machine options - they are updated in performances and new models periodically, so it makes sense to assess the original choice, on at least a quarterly basis.
2. Examination of data. Reviewing and validating the new choices: can the new size support the workload with no performance drawback? Is the code resilient enough to turn the server off nightly?
3. Use of tools. Automation tools are a must in the process, since they help building trust in the overall cycle and simplify operations.
4. Assessment of results. Check that the costs are indeed dropping.

As you can see, this is a common cycle of optimization that can be applied to nearly everything. The only difference with cloud is that the results are immediate, as seen in your monthly bill. Migrating from on-premise data centers means changing the long-term mentality where you could take your time making the assessment until the next wave of purchases: here you need to assess and act fast, but the results come along just as quickly. The internal IT culture also needs to come along, by rewarding cost savings and automation that bring short and mid-term cost savings, and in general, by spreading the word on how everyone needs to be responsible for resource optimization.

One evident example of it is the "snooze": switching on dev-test VMs only when used. (Or switching off all dev and test machines in the beginning). Having a process and a tool that can address this behavior (which for on-premise data center is obviously the opposite) can bring tons of optimization to the plate, without affecting a single bit of operations!

In addition, when updating or migrating an application, PaaS should always be considered first, as it has the most efficient and cost-wise model than pure IaaS: the hardware is only handled by usage and you don't even have to care about operating VMs as the platform takes care of it all. In case of serverless, executing a piece of code that can scale, when needed, to millions of instances without you having to put a finger on a VM should suffice to convince you of the validity of the model.

And let me give you some real experience numbers, so that the numbers we're talking about immediately prompt your decision to start optimizing:

• 38 percent reduction in cloud spending;
• 30,000 Azure virtual machine snooze requests;
• Increase in CPU utilization efficiency across infrastructure environment by over 400 percent;
• A decrease in the number of operating system instances;
• Migration to platform as a service (PaaS)–based solutions heavily promoted;
• An optimization-centered culture, which by the way, is alone the ultimate result in a cloud optimization exercise.

As I heard someone saying, cloud is not a place, it's a model. It's time to embrace the cloud model fully and start to enjoy the real benefits of it.

[source & additional info here]