The case for cloud rightsizing: Don’t get caught with your head in the clouds

 In Cloud

My dad, a self-employed businessman for many years, taught me the value of saving and spending wisely. Those same values hold true with many organizations we speak to today.

Yet there’s a concept that extends the value of your cloud environments that many companies have failed to embrace or may have even forgotten about: rightsizing.

What is rightsizing?

Rightsizing combines organizations’ current and projected cloud environments and presents recommended AWS, Azure, or Google Cloud Platform (GCP) sizing based on workload (CPU, disk, I/O, memory).With so many instance options and storage types that vary by performance, region, and price, it’s easy to overspend on resources you don’t need. Rightsizing can ensure you’re purchasing the best possible cloud instance types for the specific workloads you’re looking to move from on-premises or virtual (data center/colocation) to the cloud. Bundling this approach with Reserve Instances and upfront purchasing plans can further drive down costs.

Common rightsizing missteps

After decades in the technology industry, I find it amazing to see how traditional data center computing has transformed itself with the expansion of virtualization and cloud. For many, VMware was the de facto standard in most modern implementations of virtualization. An organization’s biggest challenge was not whether to virtualize, but how big or small the virtualization platform should be. Value added resellers and systems integrators answered the call by using tools like VMware Capacity Planner.

For those who have never used this service, it takes a data-driven approach — older physical server workloads are outfitted with algorithms that translate the best sizing for vCPU, memory, and disk requirements for a VM projected host (on newer hardware). These algorithms generate consolidation ratios, such as the number of VMs per host we can achieve. Ultimately, this rightsizing approach helps organizations maximize every dollar spent on their next iteration of a data center 2.0.

As you move workloads and applications into AWS, Azure, or GCP. One of the biggest mistakes we see is organizations relying on AWS or Azure simple calculators. These tools aren’t enough to project the costs of existing workloads; you get what we call an “as-is” snapshot in time. “As is” translates to a 1-to-1 mapping, describing the vCPU and RAM you have today and what you want in the cloud.

But that’s not the full picture. And here’s why:

  • Most production workloads running in data centers today are not actively rightsized and tuned.
  • CPU types in the cloud are usually newer and faster than what organizations have on-premises or in the datacenter. The same applies to the solid-state drive (SSD) and high-speed interconnects that backend much of the storage used at the core infrastructure.

Cloud rightsizing done right

There are three important factors to consider when rightsizing your cloud usage: sizing, provisioning, and pricing.

  • When it comes to sizing, don’t project usage based on the size of your virtual machines. Instead, determine sizing based on specific workloads, using CPU, disk, I/O, and memory.
  • Many organizations fall into the “as-is” trap by basing their provisioning on one single snapshot in time. All provisioning should be determined by an in-depth, historical analysis.
  • As for pricing, organizations that identify actual usage vs. assumed usage can save a considerable amount of money.

Predicting performance in the cloud requires translating on-premises computing and storage performance metrics to the cloud, and identifying the cloud computing and storage options that meet these performance requirements. The way to do this is by using benchmarks.

Computing benchmarks provide an understanding of what every computing option in the cloud can handle regarding CPU and memory. Storage benchmarks offer what each storage option can manage regarding input/output operations per second (IOPS), disc capacity, occupancy, and throughput.

Benchmarks for the public cloud aren’t readily available and creating them manually is tough, if not nearly impossible on your own. However SHI partners with highly specialized software companies that can correlate metrics and benchmark every possible instance and storage permutation possible, making rightsizing a reality for many organizations.

If you have a question or need help determining what cloud resources you need, contact your SHI account team or email us at cloud@shi.com.

Related Posts: You may also be interested in...


Leave a Comment

4 × five =

Pin It on Pinterest