Sergey Nivens - Fotolia
Looking for something else?
There is such a range of public cloud instances; it can be tough to know which to choose. There are many variations from top cloud providers, such as Amazon Web Services, Microsoft Azure and Google, all of which have different specifications and costs.
And since not all cloud instances are one size fits all, enterprises need to examine their workloads -- and their budget -- to see which instance type and size will work best. In some cases, they even need to create custom instances to optimize performance and cost.
Here are five tips to help identify the best, and most cost-effective, cloud instance type for your workloads.
The right size is key
Since not all workloads are static, admins need to regularly review cloud instances and know when it is time to resize. Look out for three warning signs that a cloud instance is too small: long runtimes, an inability to meet usage spikes and the need for more cloud instances to support a workload. Review cloud usage reports to determine which resources, such as memory or virtual CPU cores, your instance needs. Test the new instance size live or in a sandbox to see if it is the right fit.
When it comes to saving money, larger instances are sometimes the way to go, since they reduce the total instance count. With a right-sized instance, IT teams can avoid bottlenecks and achieve better runtimes. When you deal with bursty workloads, take a different approach. Consider long-term instances, such as Amazon Elastic Compute Cloud (EC2) Reserved Instances, for the baseline load, larger instance types for the bursts and then instances from the spot market for remaining peaks.
Snag spare capacity on the fly
While standard cloud instances can meet most needs, Spot Instances in Amazon Web Services (AWS) EC2 and Google Preemptible VMs can slash public cloud costs.
AWS Spot Instances give enterprises a chance to bid on unused EC2 capacity. The cost of these instances varies and depends on supply and demand. According to AWS, enterprises can receive an AWS Spot Instance for up to 90% less than an On-Demand Instance. Google Preemptible VMs are similar, and are up to 70% less than Google standard instances. Google, however, terminates these VMs after 24 hours, or when resources are needed for other tasks.
While they offer deep discounts, these cloud instance types mean users might not have the resources they need on-demand. Users could also be bumped by other users during a critical moment in a project. Evaluate your applications and see if you can afford these interruptions.
Consider a custom approach
Standard cloud instances don't always meet enterprise needs. For example, if you spin up a standard instance for an application that isn't compute-intensive, you might spend money on unused resources. To reduce costs and waste consider a custom instance type.
AWS, Azure and Google offer admins the ability to customize instances with the amount of memory, CPU power and other resources they need. But before you jump in, evaluate the workloads you run. Size the cloud instances correctly to save money and preserve application performance.
The emergence of infrastructure as code also allows admins to change instances on the fly. These instances are dynamic, and admins can reconfigure them to meet application needs. This approach is only beneficial in the long-term, however, if the enterprise's cloud provider bills for usage by the second. If not, a custom instance is a better choice.
Look for usage-based cloud discounts
For more savings, familiarize yourself with your providers' cloud instance types and look for available discounts. AWS EC2 Reserved Instances save enterprises 75%, compared to AWS On-Demand Instances, according to AWS, and let users reserve capacity in an availability zone. Pay upfront to save the most money with this instance type.
Google Cloud Platform helps you cut costs with its sustained use discounts, which qualify users for a discount when their cloud instances run for at least 25% of their billing month.
While Microsoft Azure doesn't offer a direct equivalent to AWS Spot Instances or Google Preemptible VMs, there are potential discounts for users. Most are attached to existing Enterprise Agreements. To test out Azure prices, consider a free trial with the spending limit feature. Also, the Azure portal allows enterprises to track and break down their costs.
Explore GPU instances
Cloud providers continue to add instance types to meet enterprise needs -- and one of those new offerings is graphic processing units (GPUs). These instances provide parallel compute power and speed up complex processing tasks, including machine learning. AWS, Azure and Google have entered the market with their own GPU instances, based on NVIDIA technology.
But GPU instances aren't for all kinds of applications. Evaluate your needs and choose the right workloads to receive the best results. Some workloads that benefit from GPUs are business analytics, video production, artificial intelligence, virtual desktop infrastructure, super-computing and engineering simulation. For example, business analytics applications, such as those used in retail, can benefit from GPU's parallel computing power -- especially those that require quick response times during peak demand.
Speedy response time is also important in academic research and the manufacturing industry, which makes GPU instances a good fit for these markets as well.
Cut cloud costs with these five tips
Don't pay for unused resources in cloud
Is a GPU cloud instance worth the extra cost?