Cloud Rightsizing: The Ugly Truth About (Most) Recommendations

Cloud Rightsizing: The Ugly Truth About (Most) Recommendations

This is part of Solutions Review’s Premium Content Series, a collection of contributed columns written by industry experts in maturing software categories. In this submission, Virtana Senior Product Manager David McNerney outlines the ugly truth about most cloud rightsizing recommendations.

SR Premium ContentFor enterprises moving to hybrid / multi-cloud environments, rightsizing is one of the critical steps that organizations can take to control costs and optimize resources. Cloud waste is a massive problem, and the common consensus is that idle and abandoned resources can account for up to one-third of wasted cloud costs. Rightsizing is about finding the optimal cloud configuration options to ensure that you get the performance you need—within any given constraints you are operating under—at the lowest possible cost. This seems like a simple proposition, but deceptively so, as there are several factors that can make it a complicated process.

For one thing, business requirements are constantly changing, so workloads must adapt to support them, which in turn changes their operating parameters. And even when business requirements remain constant, the workloads themselves can evolve as usage changes. The other complicating factor is that cloud service providers offer more configuration options than you could possibly evaluate. We are talking about hundreds of thousands. This is why many cloud cost optimization tools provide rightsizing recommendations.

The idea behind rightsizing recommendations is to help organizations reduce the number of choices down to a manageable list of the most relevant options that will help organizations save money while maintaining desired performance levels. This really is a must-have capability – but there is a big problem that no one talks about.

Just because boxes are being checked that a feature exists doesn’t mean that it will deliver the benefits and value an organization needs; or that it considers the appropriate requirements to save money while operating at optimum performance levels. Too often, the rightsizing recommendations provided by optimization tools can’t actually be implemented, rendering that feature unusable.

There are a few reasons why this is happening, so let’s break down some of the challenges with rightsizing recommendations.

Limitations are Overlooked

As an example, a mid-sized IT security company focused on data loss prevention was running an application on a specific—read: older—version of an OS. It had not been updated in years because doing so would require an investment in a new OS and infrastructure to run it. The

application was running well, so there was simply no reason to fix what was not broken. For this company, recommendations that require a newer version of the OS were useless. And without a way to tell their tool about their OS requirement for this particular instance, they will never get a recommendation that can be implemented.

Another company had an application running a memory-heavy database, so recommendations that cut memory to any degree were non-starters. It is not that reducing memory was completely off the table, the company just needed to prove that the action would not have an adverse effect. This leads to the next issue.

Recommendations are Made on Blind Trust

Many tools say, here are our recommendations—trust us! Some will provide very basic categorization (easy/medium/hard, or baseline/aggressive) but you cannot tweak the underlying data science, which is a black box. How comfortable are you with making changes impacting mission-critical workloads for your organization based solely on trust?

Not Looking at the Big Picture

In many organizations, you have to deliver the recommendations to a different team to implement and it can sometimes be difficult to ensure adoption. For example, the development team is not traditionally used to thinking about the cost. If focusing on functionality and security, they found through app testing that doubling the size of CPU worked, and that is what they are going to go with. They’re not likely to experiment with how to get to the lowest CPU threshold that still enables the application to operate as required. They’re also not likely to want to implement a change to save the company $100 per month. This is understandable—they don’t want to risk breaking anything. But if you could prove that the application will not be negatively affected and that this is in the best interest of the business, then you’re more likely to be able to profit from those recommendations.

One-Size Does Not Fit All

Different teams have different goals. For production applications that are stable and running in volume, any opportunity to squeeze out inefficiencies can lead to significant savings. Development teams, on the other hand, benefit from experimentation. Organizations don’t want to stifle innovation for the sake of pinching pennies. Likewise, different workloads have different requirements. If you apply the same set of parameters to your customer-facing applications as you do to your backup workloads, you either risk limiting performance where you can’t afford to, or more likely, overspend when it’s not needed.

What’s Needed in a Recommendation Engine

If any of these issues sound familiar, it might be time to find a new source for rightsizing recommendations. There are a few key requirements organizations should be aware of including:

  • Customization: Companies need the ability to set parameters based on specific business or technical requirements for specific applications or workloads. If you have particular CPU, network, memory, or disk needs, you need to be able to factor those in. If you have certain constraints, such as the OS version, those need to be taken into account.
  • What-if analysis: Getting a recommendation to save money is one thing; understanding the impact beyond cost is something else altogether. It’s critical to be able to see the effect the change will have and tune the recommendation accordingly.
  • Multi-policy support: Organizations must be able to support the varying needs of all constituents across the organization. The only way to do that is to apply different policies to different areas.
  • Integration with change management: For some organizations, having recommendations integrated with change management, such as ServiceNow or Jira, may be a nice-to-have rather than a requirement, but it makes the process of applying the recommendations much simpler. Embedding them into operational workflow can improve adoption of those changes.

Enterprise leaders need solutions that will radically simplify the management of their hybrid cloud IT infrastructure. Rightsizing recommendations are about finding the optimal cloud configuration options to ensure that you get the performance you need at the lowest possible cost. After all, the only way to get value from recommendations is to actually implement them.

David McNerney
Follow Dave
Latest posts by David McNerney (see all)