EC2 Recommendations to Benefit Your Cloud-Native Strategy

One of the key ways Apptio supports customer efforts to maximize the business value of technology investments is through helping them take advantage of cloud-native technologies such as Kubernetes and managed services. We are excited to announce product enhancements to Cloudability that further these efforts for Amazon EC2 by embracing scaling best practices and optimizing machine types designed for modern workloads.

A foundational aspect of all public clouds, both for end-user consumption and for backing popular cloud services such as managed databases, is the virtual machine (VM). For AWS, these VMs are called EC2 instances and make up the biggest category of usage. In fact, across Apptio’s customer base, EC2 instance usage makes up over 20% of total AWS spend, making it the biggest category of cloud usage.

Running VMs on public cloud comes with maximum flexibility, allowing you to continue traditional on-prem patterns, simply running large general-purpose machines around the clock regardless of workload demands, or adopt new strategies that have been enabled by the cloud and which open opportunities for novel solutions and use cases. With so much of your AWS spend dedicated to EC2 instances, it’s vital you look for, and act when you find, opportunities to rightsize this infrastructure. Cloudability now provides new capabilities to optimize EC2 usage with rightsizing recommendations for Auto Scaling groups (ASG) and GPU-backed accelerated computing instances.

Deliver application availability and cost efficiency by rightsizing EC2 Auto Scaling groups

AWS provides many options for scaling resources in and out with EC2 Auto Scaling groups (ASG), offering the most flexibility and potential benefit. ASGs have controls for hosting a workload with multiple EC2 instances simultaneously. The operations teams who manage these groups are responsible for defining policies that consider both application availability and cost efficiency. Important settings include what instance type to use, the baseline number of instances in the group, and, crucially, rules for when to scale out the instance count and back in again.

Any optimization efforts need to consider the entire group of EC2 instances together — not individually, as you might normally do when rightsizing. This is especially true since EC2 instances that are members of an ASG are often ephemeral — being instantiated and then terminated based on scaling events — while the group itself lives on.

With this in mind, Apptio is excited to launch rightsizing for EC2 Auto Scaling groups in Cloudability. Within the AWS section of Cloudability rightsizing, you’ll now find a new tab for rightsizing ASGs (rightsizing recommendations for individual instances remain in the regular EC2 tab). The platform automatically maps EC2 instances to their respective ASGs. For each hour of the reporting period, Cloudability evaluates how many instances were running for each ASG and the resource consumption across important metrics like CPU and memory. Cost-saving recommendations are then made at the level of the ASG, which includes changing both the number of instances (min and max) and the instance type. To help engineers choose the right combination, the details page has additional information with the utilization charts (CPU, memory, etc.) each overlayed with the historical instance count, making it straightforward to understand scaling trends and correlate this to resource utilization.

Optimize machine learning and other modern workloads

Adopting public cloud comes with opportunities to use specialized VMs, getting benefits like automatic bursting and processors designed by the cloud vendor to deliver the best price performance. One of these unique opportunities that cloud customers are embracing is accelerating their VMs with graphics processing units (GPUs), powering modern workloads such as graphics-intensive processing and machine learning. Over the last four years, our customers have increased usage of this instance category by 300% as a portion of overall EC2 instance usage.

For AWS, most GPU-backed EC2 instances belong to the Accelerated Computing category and include multiple generations of the P and G instance types. As these instance types are relatively expensive, it is important that teams optimize this usage, ensuring there’s adequate budget for ongoing initiatives. This requires understanding the consumption of the GPU and its associated memory, in addition to regular metrics such as CPU utilization.

Cloudability now offers full support for rightsizing these GPU-backed instances. GPU consumption data can be ingested via a specific CloudWatch custom metric or via Datadog for customers who have already configured this integration.

The rightsizing recommendations then surface two common paths for customers to reduce costs. Where the provisioned GPUs are excessive, the instance may switch to a better matched GPU-backed type — for example, a P3.8xlarge (4 GPUs) could possibly be switched to a G5.8xlarge (1 GPU) for an 80% saving — while if the GPU is idle over the term a valid option could be to switch the P3.8xlarge to an M6i.8xlarge (a far cheaper instance with no GPU) for even greater savings. To support validation of these cost-saving options, additional charts have been added for GPU and its associated memory to the rightsizing details page.

Continue benefiting from your cloud-native strategy

You are paying a needlessly high price for the flexibility and scalability of EC2 if you aren’t rightsizing your virtual machines. But rightsizing isn’t a one-and-done activity — not while AWS continues to roll out new EC2 instance types and provide new controls for scaling EC2 instances in and out. Maximizing the benefits of your cloud-native strategy means not only adopting these types of vendor innovations but also optimizing how and when you apply them. Stay tuned over the coming months as the Apptio engineering team continues to enhance Cloudability to deliver insights that help you do just that.