Setting up the right instance when there are over 400,000 unique EC2 SKUs (products) to choose from doesn’t just sound like a challenge–it is one. At its most basic, you must choose for 74 different product details that together determine the overall performance and price of your EC2. These details include information like: term type, lease contract length, region, instance type, instance family, storage type, network type, process type, max IOPS per volume, tenancy, operating system, dedicated EBS throughput, enhanced networking support, and more.
Each month new categories and values are added to the already robust menu of options. Just check out this EC2 SKU spreadsheet from Amazon–it’s over 300mb!
If all of that weren’t hard enough, when it comes to making convertible reservations, not every EC2 type can be converted into another (should you want or need to change). This means finding the right instance type and size early on is important–not only from an architectural standpoint, but also a financial one.
As of today, there are 15 EC2 types among 5 categories, totaling in 82 sizing options. In total there are over 400,000 rows in the official EC2 sizing file.
Official info from Amazon.com
With Metricly, analyze your EC2s and get recommendations for the right instance type per your instance’s workload and risk tolerance.
“It’s in the name, Metricly; plethora of metrics available with lots of customization for different dashboards and alerts.”
Michael S. – IT Operation Manager
Create a filter that targets the instances you want to resize using criteria such as element name, tags, or attributes. You can save these settings as a report for later use.
Tune our behavioral constraint options to exclude instances that belong to certain families or generations, or that don’t leave enough headroom for your historical workload.
Update your instances with our proposed types through the AWS console. We recommend starting with our report’s Top 10 Savings list.
Now that you have made sizing changes, use the box plot report to spot any performance bottlenecks in CPU, memory, IOPS or network utilization.
Load your saved filter from the Recommendation report (Part I, Step 1.) or create a new one.
Choose a metric such as memory or IOPS usage to see their usage aggregated based on min, max, median, and 25/75 percentile over the last hour, day or week.
Observe each element’s performance load, broken down across a min-max spectrum, to identify instances that are running too close to their full capacity.
Every time you add nodes to a cluster, deploy fresh code and new features, or simply experience heavier workloads–examining the usage of your resources becomes a critical step to balancing performance with cost savings.
With the help of our customers, we’ve come up with best practices to help form organizational habits to enforce a culture of cost awareness.
We recommend designating one person responsible for sharing a weekly report with your team to bring more control and accountability to your process.
It’s best to start simple. Each person who launches an instance tags it as owner (a group, department, application or individual). Your champion ensures all instances are tagged.
Refine your filters according to tag, attribute, and risk tolerance constraints and then save the report so that you can find it again and we can automatically share it via email or Slack.
We can rank your savings by the largest financial saving potential, so that your champion can share each week with your extended team and senior management. This step promotes a culture of cost awareness.
Not every instance should be downsized, so the champion should apply a special tag that excludes them from downsizing according to your management criteria.
Create a schedule for updating reports. As environments grow, utilization needs to change and new opportunities for cost savings may arise. Use previously saved reports or get a fresh perspective with new ones!
Analyze all of your computing resources including IOPS and memory to avoid bottlenecks.
Significantly save by rightsizing your environments and avoiding stranded capacity.
Automate the analysis and enforce a culture of cost awareness across your organization.
Not signed up? Get started with
a free 21-day trial.
Make sure your Metricly account has activated an AWS data-source, and activated the cost integration.
We require 7 days of performance data to be gathered before we make sizing reccomendations.
Sign up for free, and join DevOps who rate us 5-stars on Capterra!21-day free trial No credit card required